paper_id string | arxiv_id string | title string | markdown dict | reviews list | scores dict | metadata dict | meta_review dict | decision dict |
|---|---|---|---|---|---|---|---|---|
sIE2rI3ZPs | 2410.24206v1 | Understanding Optimization in Deep Learning with Central Flows | {
"content": "## Abstract\n\nAbstract Optimization in deep learning remains poorly understood, even in the simple setting of deterministic (i.e. full-batch) training. A key difficulty is that much of an optimizer’s behavior is implicitly determined by complex oscillatory dynamics, referred to as the “edge of stabilit... | [
{
"id": "d5JziU6soC",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 4,
"presentation": 4,
"summary": "The paper investigates optimization methods in deep learning, focusing on the \"edge of stability\" regime. Since the gradient flow differential equation fails to... | {
"rating": "5;5;8;8;8",
"rating_avg": 6.8,
"confidence": "4;3;3;4;3",
"confidence_avg": 3.4,
"soundness": "3;2;2;4;3",
"soundness_avg": 2.8,
"contribution": "2;3;4;4;4",
"contribution_avg": 3.4,
"presentation": "2;4;4;4;4",
"presentation_avg": 3.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.816587"
} | {
"id": "CjJz2SXYyg",
"metareview": "The paper introduces a novel framework for understanding the long-term dynamics of optimization in deep learning, particularly in the \"edge of stability\" regime. By modeling the time-averaged behavior of GD and RMSProp, the paper reveals some new insights, including implicit r... | {
"decision": "Accept (Poster)"
} |
sL2F9YCMXf | 2410.21357v1 | Energy-Based Diffusion Language Models for Text Generation | {
"content": "## Abstract\n\nAbstract Despite remarkable progress in autoregressive language models, alternative generative paradigms beyond left-to-right generation are still being actively explored. Discrete diffusion models, with the capacity for parallel generation, have recently emerged as a promising alternativ... | [
{
"id": "V3y1N8sPYQ",
"initial_rating": 6,
"confidence": 2,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents a new Discrete Diffusion Model that models an energy function $E_\\phi$ to improve sampling procedure from an existing Diffusion Model. The au... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;3;4;2",
"confidence_avg": 3.25,
"soundness": "3;3;4;2",
"soundness_avg": 3,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.817880"
} | {
"id": "diEKmAxPlh",
"metareview": "This paper introduces Energy-based Diffusion Language Models (EDLM), which addresses token dependencies in discrete diffusion models by incorporating an energy-based model for reranking during the denoising process. The paper's main strengths are its clear presentation, strong e... | {
"decision": "Accept (Poster)"
} |
sMWkTWh2JF | 2410.02751v1 | ReLIC: A Recipe for 64k Steps of In-Context Reinforcement Learning for Embodied AI | {
"content": "## Abstract\n\nAbstract Intelligent embodied agents need to quickly adapt to new scenarios by integrating long histories of experience into decision-making. For instance, a robot in an unfamiliar house initially wouldn’t know the locations of objects needed for tasks and might perform inefficiently. How... | [
{
"id": "l0oZWW4DMb",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces ReLIC (Reinforcement Learning In Context), a new approach that enables embodied AI agents to adapt to new environments using up to 64,000 ste... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "3;3;3",
"confidence_avg": 3,
"soundness": "3;2;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;2;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.818673"
} | {
"id": "1K7akjMtth",
"metareview": "The paper presents an approach for using large context transformers to attend to large amounts of experience. The reviewers have a split opinion on this paper, which was not fully resolved during the discussion phase. Hence I have taken a closer look at the paper. On the positiv... | {
"decision": "Reject"
} |
sMyXP8Tanm | 2406.03736v2 | Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data | {
"content": "## Abstract\n\nAbstract Discrete diffusion models with absorbing processes have shown promise in language modeling. The key quantities to be estimated are the ratios between the marginal probabilities of two transitive states at all timesteps, called the concrete score. In this paper, we reveal that the... | [
{
"id": "HcWebFM4g9",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposes a reparameterization of the absorbing discrete diffusion, based upon the finding that the concrete score can be decomposed into a time-indepen... | {
"rating": "5;5;5;6;8",
"rating_avg": 5.8,
"confidence": "5;4;3;5;3",
"confidence_avg": 4,
"soundness": "3;3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;2;3;3",
"contribution_avg": 2.4,
"presentation": "3;2;2;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.819659"
} | {
"id": "A5CPxVmRFr",
"metareview": "This paper presents a reparameterization of absorbing discrete diffusion called RADD, which simplifies the modeling of the concrete score as a time-independent conditional probability of clean data. This reparameterization eliminates the need for time conditioning in the network... | {
"decision": "Accept (Poster)"
} |
sNntRFmn72 | 2410.10774v1 | Cavia: Camera-controllable Multi-view Video Diffusion with View-Integrated Attention | {
"content": "## Abstract\n\nAbstract In recent years there have been remarkable breakthroughs in image-to-video generation.\nHowever, the 3D consistency and camera controllability of generated frames have remained unsolved. Recent studies have attempted to incorporate camera control into the generation process, but ... | [
{
"id": "sWYi5bptg7",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 1,
"summary": "This paper presents a novel framework, Cavia, designed for generating multi-view videos with precise camera control. The authors introduce view-integrated attenti... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "5;4;3;4",
"confidence_avg": 4,
"soundness": "3;3;2;4",
"soundness_avg": 3,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;1;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.820601"
} | {
"id": "xnjo5xMVRp",
"metareview": "The paper receives mixed ratings from four reviewers. Most of the reviewers are leaning to reject the submission. They mainly have several concerns regarding this submission, including the insufficient novelty of embedding camera poses as a condition, the unconvincing results sh... | {
"decision": "Reject"
} |
sOAQY8hrAu | 2410.09009v1 | Semantic Score Distillation Sampling for Compositional Text-to-3D Generation | {
"content": "## Abstract\n\nAbstract Generating high-quality 3D assets from textual descriptions remains a pivotal challenge in computer graphics and vision research. Due to the scarcity of 3D data, state-of-the-art approaches utilize pre-trained 2D diffusion priors, optimized through Score Distillation Sampling (SD... | [
{
"id": "q2p8TpkX8N",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper introduces a score distillation-based learning framework for compositional text-to-3D generation. This framework consists of two components: (1) Progra... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "3;3;2;3",
"contribution_avg": 2.75,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.821279"
} | {
"id": "PThX0cZZra",
"metareview": "The paper in question introduces a novel mechanism, Semantic Score Distillation Sampling (SemanticSDS), for text-to-3D generation aimed at improving the compositional layout of objects in a 3D scene. The approach leverages program-aided layout planning and semantic maps for bett... | {
"decision": "Reject"
} |
sOQmgO0PTv | 2405.14600v1 | Discretization of continuous input spaces in the hippocampal autoencoder | {
"content": "## Abstract\n\nAbstract The hippocampus has been associated with both spatial cognition and episodic memory formation, but integrating these functions into a unified framework remains challenging. Here, we demonstrate that forming discrete memories of visual events in sparse autoencoder neurons can prod... | [
{
"id": "wjIqpJpSIh",
"initial_rating": 3,
"confidence": 4,
"soundness": 1,
"contribution": 2,
"presentation": 3,
"summary": "This paper models the formation of hippocampal-like spatial representations by training sparse autoencoders on visual images from 3D environments. By applying an ... | {
"rating": "3;3;5",
"rating_avg": 3.6666666666666665,
"confidence": "5;4;4",
"confidence_avg": 4.333333333333333,
"soundness": "1;1;3",
"soundness_avg": 1.6666666666666667,
"contribution": "3;2;2",
"contribution_avg": 2.3333333333333335,
"presentation": "4;3;2",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.821883"
} | {
"id": "zkEirw5Hzn",
"metareview": "This paper models the formation of hippocampal-like spatial representations by training sparse autoencoders on visual images from 3D environments. By applying an orthonormal activity regularization to encourage sparsity, the authors observe place cell-like behavior, where certai... | {
"decision": "Reject"
} |
sPuLtU32av | 2311.16086v1 | MAST: model-agnostic sparsified training | {
"content": "## Abstract\n\nAbstract We introduce a novel optimization problem formulation that departs from the conventional way of minimizing machine learning model loss as a black-box function. Unlike traditional formulations, the proposed approach explicitly incorporates an initially pre-trained model and random... | [
{
"id": "2jZikIURj0",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper introduces Model-Agnostic Sparsified Training (MAST), a novel optimization framework tailored for sparsification in model training. By defining a new o... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "1;2;3;3",
"contribution_avg": 2.25,
"presentation": "3;4;3;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.823210"
} | {
"id": "ArmakgZjyj",
"metareview": "This paper presents MAST, a new optimization framework for sparsification in model training. MAST offers a theoretically sound approach with potential applications in various machine learning scenarios. While the theoretical assumptions might be restrictive for some real-world p... | {
"decision": "Accept (Poster)"
} |
sR0xz6ZaH7 | 2410.18979v1 | PixelGaussian: Generalizable 3D Gaussian Reconstruction from Arbitrary Views | {
"content": "## Abstract\n\nAbstract We propose PixelGaussian , an efficient feed-forward framework for learning generalizable 3D Gaussian reconstruction from arbitrary views.\nMost existing methods rely on uniform pixel-wise Gaussian representations, which learn a fixed number of 3D Gaussians for each view and cann... | [
{
"id": "qsVmdrwUPB",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work introduces PixelGaussian, a feed-forward 3DGS model that can adaptively update the number of Gaussian via pruning and splitting. In particular, PixelGau... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "5;5;3;4",
"confidence_avg": 4.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.824630"
} | {
"id": "oZRdM7G1JJ",
"metareview": "This paper receives borderline ratings of 5,6,5,6. The AC look at the reviews, rebuttal and the discussions, and decide to reject the paper for now. The authors are encouraged to fix the issues and resubmit to future conference. The main concern raised by the reviewers is that t... | {
"decision": "Reject"
} |
sRIU6k2TcU | 2410.12361v1 | Proactive Agent: Shifting LLM Agents from Reactive Responses to Active Assistance | {
"content": "## Abstract\n\nAbstract Agents powered by large language models have shown remarkable abilities in solving complex tasks. However, most agent systems remain reactive, limiting their effectiveness in scenarios requiring foresight and autonomous decision-making. In this paper, we tackle the challenge of d... | [
{
"id": "MFtfQKP0S5",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper addresses the limitations of reactive agent systems by introducing a proactive approach to LLM agents. The authors develop a data-driven method where r... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.825398"
} | {
"id": "2bTIdSx3ZI",
"metareview": "(a) Summary: This paper proposes a novel data-driven formalization for developing a proactive agent that anticipates user needs and takes the initiative by suggesting tasks or providing information without explicit requests.\n(b) Strengths: The paper is generally well-written an... | {
"decision": "Accept (Poster)"
} |
sRaAt9OOnW | 2303.05978v2 | Uncovering Challenges of Solving the Continuous Gromov-Wasserstein Problem | {
"content": "## Abstract\n\nAbstract Recently, the Gromov-Wasserstein Optimal Transport (GWOT) problem has attracted the special attention of the ML community. In this problem, given two distributions supported on two (possibly different) spaces, one has to find the most isometric map between them. In the discrete v... | [
{
"id": "zWltBEFlKU",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper discusses the challenges of solving the Continuous Gromov-Wasserstein Optimal Transport (GWOT) problem and emphasizes the need for a reliable and genera... | {
"rating": "3;5;5;6;8",
"rating_avg": 5.4,
"confidence": "4;3;3;4;4",
"confidence_avg": 3.6,
"soundness": "3;2;2;3;4",
"soundness_avg": 2.8,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "3;2;3;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.826067"
} | {
"id": "aEjRihMSx6",
"metareview": "The paper addresses the continuous Gromov-Wasserstein Optimal Transport (GWOT) problem and proposes a benchmark alongside a NeuralGW solver to mitigate issues in existing methods. Despite the novelty of focusing on continuous GWOT, the reviewers highlighted several weaknesses th... | {
"decision": "Reject"
} |
sSRSKjLki6 | 2405.15684v1 | Prompt-Aware Adapter: Towards Learning Adaptive Visual Tokens for Multimodal Large Language Models | {
"content": "## Abstract\n\nAbstract To bridge the gap between vision and language modalities, Multimodal Large Language Models (MLLMs) usually learn an adapter that converts visual inputs to understandable tokens for Large Language Models (LLMs).\nHowever, most adapters generate consistent visual tokens, regardless... | [
{
"id": "rYR5WHBL1E",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper studies the impact of prompts on adapters, revealing that prompt-independent adapters still have deficiencies in visual understanding. To this end, thi... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;5;4;2",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;2;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.826879"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
sSWiZr8QU7 | 2410.17103v1 | A Hybrid Simulation of DNN-based Gray Box Models | {
"content": "## Abstract\n\nAbstract Simulation is vital for scientific and engineering disciplines, as it enables the prediction and design of physical systems. However, the computational challenges inherent to large-scale simulations often arise from complex device models featuring high degrees of nonlinearities o... | [
{
"id": "Bse5ght03a",
"initial_rating": 6,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "In this work, author try to augment simulator with DNN to learn hidden non-linear dynamics not captured by first principle solver. \nprior work explicitly define ... | {
"rating": "1;3;5;5;6",
"rating_avg": 4,
"confidence": "4;3;2;4;3",
"confidence_avg": 3.2,
"soundness": "1;2;2;3;2",
"soundness_avg": 2,
"contribution": "1;1;2;2;2",
"contribution_avg": 1.6,
"presentation": "1;2;2;2;2",
"presentation_avg": 1.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Desk Rejected Submission",
"venueid": "ICLR.cc/2025/Conference/Desk_Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.827980"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
sUvBTEYXGt | 2410.04974v2 | 6DGS: Enhanced Direction-Aware Gaussian Splatting for Volumetric Rendering | {
"content": "## Abstract\n\nAbstract Novel view synthesis has advanced significantly with the development of neural radiance fields (NeRF) and 3D Gaussian splatting (3DGS). However, achieving high quality without compromising real-time rendering remains challenging, particularly for physically-based ray tracing with... | [
{
"id": "qVtnyH6Lyx",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 3,
"summary": "This paper presents 6D Gaussian Splatting(6DGS), an advanced method for volumetric rendering that enhances traditional 3D Gaussian splatting (3DGS) by incorporati... | {
"rating": "6;6;8;8",
"rating_avg": 7,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "3;2;3;4",
"soundness_avg": 3,
"contribution": "2;3;3;4",
"contribution_avg": 3,
"presentation": "4;3;4;3",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.828827"
} | {
"id": "MzL1FkscuW",
"metareview": "This paper proposes a novel 3D Gaussians representation that can model view-dependent effects, and thus achieve better results on transparent or glossy materials. The proposed method achieves better performance compared to native 3DGS as well as N-D GS, the fact that it can be c... | {
"decision": "Accept (Poster)"
} |
sVBnGcbkkM | 2406.08472v2 | RILe: Reinforced Imitation Learning | {
"content": "## Abstract\n\nAbstract Reinforcement Learning has achieved significant success in generating complex behavior but often requires extensive reward function engineering. Adversarial variants of Imitation Learning and Inverse Reinforcement Learning offer an alternative by learning policies from expert dem... | [
{
"id": "79hUsw799M",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces ‘Reinforced Imitation Learning’, a new trainer-student framework. In this framework, the trainer receives feedback from a discriminator and ... | {
"rating": "3;3;3;8",
"rating_avg": 4.25,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;2;1;3",
"soundness_avg": 2,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.829458"
} | {
"id": "mbhhcMcMPP",
"metareview": "This paper proposes a novel method for adversarial imitation learning with three components: a discriminator that provides feedback to a trainer (an RL agent) which is optimized to reward the student (another RL agent). The paper argues that the additional trainer component can ... | {
"decision": "Reject"
} |
sYNWqQYJhz | 2406.10630v1 | Emerging Safety Attack and Defense in Federated Instruction Tuning of Large Language Models | {
"content": "## Abstract\n\nAbstract Federated learning (FL) enables multiple parties to collaboratively fine-tune an large language model (LLM) without the need of direct data sharing.\nIdeally, by training on decentralized data that is aligned with human preferences and safety principles, federated instruction tun... | [
{
"id": "9mWUUMMblu",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "The paper addresses the vulnerabilities in the safety alignment of large language models (LLMs) trained through federated instruction tuning (FedIT). It introduce... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;3;2",
"confidence_avg": 3.25,
"soundness": "1;4;2;3",
"soundness_avg": 2.5,
"contribution": "1;3;3;3",
"contribution_avg": 2.5,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.830532"
} | {
"id": "fqZs4o6VTf",
"metareview": "The paper addresses the vulnerabilities in the safety alignment of large language models (LLMs) trained through federated instruction tuning (FedIT). It introduces a novel safety attack method that allows malicious clients to compromise the safety of LLMs by using unaligned data... | {
"decision": "Accept (Poster)"
} |
sZGZJhaNSe | 2411.05735v1 | Aioli: A Unified Optimization Framework for Language Model Data Mixing | {
"content": "## Abstract\n\nAbstract Language model performance depends on identifying the optimal mixture of data groups to train on (e.g., law, code, math). Prior work has proposed a diverse set of methods to efficiently learn mixture proportions, ranging from fitting regression models over training runs to dynami... | [
{
"id": "lll3pUO2rG",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 4,
"presentation": 3,
"summary": "This paper targets the challenge of optimizing data mixture groups in training large language models. The authors claims that no existing method consistently outp... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"contribution": "3;3;3;4",
"contribution_avg": 3.25,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.831597"
} | {
"id": "425GRf5Bhb",
"metareview": "The paper introduces a framework that unifies various existing data mixing methods, providing a theoretical lens to analyze their strengths and weaknesses. The algorithm demonstrates consistent improvement over stratified sampling and robustness across different datasets. Data m... | {
"decision": "Accept (Poster)"
} |
sZJqKAVzKf | 2404.18134v1 | Learning Fairer Representations with FairVIC | {
"content": "## Abstract\n\nAbstract Mitigating bias in automated decision-making systems, specifically deep learning models, is a critical challenge in achieving fairness. This complexity stems from factors such as nuanced definitions of fairness, unique biases in each dataset, and the trade-off between fairness an... | [
{
"id": "6euE3JTNsN",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes FairVIC, a method to learn fair representations by adding terms for variance invariance and covariance, to remove any correlation between sensi... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;5;3;4",
"confidence_avg": 4,
"soundness": "1;2;2;2",
"soundness_avg": 1.75,
"contribution": "1;1;2;2",
"contribution_avg": 1.5,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.832481"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
saJkPzTmZz | 2406.15599v1 | Pareto-Optimal Learning from Preferences with Hidden Context | {
"content": "## Abstract\n\nAbstract Ensuring AI models align with human values is essential for their safety and functionality. Reinforcement learning from human feedback (RLHF) uses human preferences to achieve this alignment. However, preferences sourced from diverse populations can result in point estimates of h... | [
{
"id": "VDGOiefCN6",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 1,
"summary": "This paper introduces a reinforcement learning from human feedback with hidden context (RLHF-HC) framework called Pareto optimal preference learning (POPL). While... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "3;3;2;4",
"confidence_avg": 3,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;1;2;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.833216"
} | {
"id": "7LKBAJTPI5",
"metareview": "This paper proposes Pareto Optimal Preference Learning (POPL) to enable pluralistic alignment by framing discrepant group preferences as objectives with potential trade-offs, aiming for policies that are Pareto-optimal on the preference dataset.\n\nThe proposed Pareto-optimal so... | {
"decision": "Reject"
} |
sahQq2sH5x | 2407.01163v1 | Benchmarking Predictive Coding Networks -- Made Simple | {
"content": "## Abstract\n\nAbstract In this work, we tackle the problems of efficiency and scalability for predictive coding networks in machine learning. To do so, we first propose a library called PCX, whose focus lies on performance and simplicity, and provides a user-friendly, deep-learning oriented interface. ... | [
{
"id": "63hHkdvMT3",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 3,
"summary": "The authors undertake the very hard task of trying to streamline, standardise and robustify the scientific process in the subfield of Predictive Coding networks. ... | {
"rating": "5;6;8",
"rating_avg": 6.333333333333333,
"confidence": "3;4;4",
"confidence_avg": 3.6666666666666665,
"soundness": "1;3;4",
"soundness_avg": 2.6666666666666665,
"contribution": "3;3;4",
"contribution_avg": 3.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.66666666666666... | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.833935"
} | {
"id": "asAyKc8BsD",
"metareview": "The paper makes several contributions to predictive coding networks in machine learning in the form of a library to enable and facilitate comparisons, an attempt at unification of benchmarks, as well as rigorous empirical evaluation. These latter points address several identifie... | {
"decision": "Accept (Spotlight)"
} |
sb1HgVDLjN | 2410.11502v1 | Offline Model-Based Optimization by Learning to Rank | {
"content": "## Abstract\n\nAbstract Offline model-based optimization (MBO) aims to identify a design that maximizes a black-box function using only a fixed, pre-collected dataset of designs and their corresponding scores. This problem has garnered significant attention from both scientific and industrial domains. A... | [
{
"id": "Af13NM3Xsg",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "Offline model-based optimization (MBO) is concerned with the goal of maximizing an unknown objective function using an offline dataset. \n$\\min_{x \\in X} f(x)$ ... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "3;4;4",
"confidence_avg": 3.6666666666666665,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "1;2;3",
"contribution_avg": 2,
"presentation": "3;4;3",
"presentation_avg": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.834851"
} | {
"id": "nWsWpd3ANd",
"metareview": "The paper considers the problem of offline model based optimization. The key idea is to use learning to rank loss instead of mean squared error loss for training surrogate models. Although this point has already been discussed earlier in MatchOpt, the paper does more thorough an... | {
"decision": "Accept (Poster)"
} |
sbG8qhMjkZ | 2409.08469v2 | Improved Finite-Particle Convergence Rates for Stein Variational Gradient Descent | {
"content": "## Abstract\n\nAbstract. We provide finite-particle convergence rates for the Stein Variational Gradient Descent (SVGD) algorithm in the Kernelized Stein Discrepancy ( 𝖪𝖲𝖣 𝖪𝖲𝖣 \\mathsf{KSD} ) and Wasserstein-2 metrics. Our key insight is that the time derivative of the relative entropy between the... | [
{
"id": "ubtqw682wr",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 4,
"presentation": 3,
"summary": "This work provides a convergence analysis of the Stein Variational Gradient Descent (SVGD) algorithm in its full formulation, i.e., using finitely many particles ... | {
"rating": "6;8;8;8",
"rating_avg": 7.5,
"confidence": "3;4;5;3",
"confidence_avg": 3.75,
"soundness": "3;4;4;4",
"soundness_avg": 3.75,
"contribution": "3;4;4;4",
"contribution_avg": 3.75,
"presentation": "2;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.835838"
} | {
"id": "HWaU0DGYDY",
"metareview": "This work provides a convergence analysis of the Stein Variational Gradient Descent (SVGD) algorithm in its full formulation, i.e., using finitely many particles and in discrete time. Such a quantitative convergence proof was long sought after, ever since the algorithm was first... | {
"decision": "Accept (Oral)"
} |
scI9307PLG | 2405.15540v1 | Bundle Neural Network for message diffusion on graphs | {
"content": "## Abstract\n\nAbstract The dominant paradigm for learning on graph-structured data is message passing. Despite being a strong inductive bias, the local message passing mechanism suffers from pathological issues such as over-smoothing, over-squashing, and limited node-level expressivity.\nTo address the... | [
{
"id": "D3K6v8SW8c",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper extends previous work on graph neural networks equipped with cellular sheaves by introducing a more efficient way of computing the heat diffusion over ... | {
"rating": "6;6;6",
"rating_avg": 6,
"confidence": "3;2;4",
"confidence_avg": 3,
"soundness": "3;2;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;2;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.836725"
} | {
"id": "8O8DLBlg9U",
"metareview": "The manuscript proposes a novel graphical neural network architecture that sends messages on flat vector bundles. The reviewers unaminously supports the manuscript for acceptance. During the discussion phase, the authors also address most of the concerns by the reviewers. Theref... | {
"decision": "Accept (Spotlight)"
} |
scKAXgonmq | 2406.00384v1 | CapeX: Category-Agnostic Pose Estimation from Textual Point Explanation | {
"content": "## Abstract\n\nAbstract Conventional 2D pose estimation models are constrained by their design to specific object categories. This limits their applicability to predefined objects. To overcome these limitations, category-agnostic pose estimation (CAPE) emerged as a solution. CAPE aims to facilitate keyp... | [
{
"id": "IvIhbKbVCN",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes to employ textual information connected in a graph structure to tackle the problem of category-agnostic pose estimation (CAPE). The textural k... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;5;4;5",
"confidence_avg": 4.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.837403"
} | {
"id": "7GQbtpQydL",
"metareview": "This work proposes to make full use of text to enhance Category-Agnostic Pose Estimation (CAPE), and the method seems to be novel and interesting. The proposed scheme is reasonable and works for this task. Most of the reviewers pointed out that this work is novel and recommended... | {
"decision": "Accept (Poster)"
} |
se4vjm7h4E | 2410.01131v1 | nGPT: Normalized Transformer with Representation Learning on the Hypersphere | {
"content": "## Abstract\n\nAbstract We propose a novel neural network architecture, the normalized Transformer (nGPT) with representation learning on the hypersphere. In nGPT, all vectors forming the embeddings, MLP, attention matrices and hidden states are unit norm normalized. The input stream of tokens travels o... | [
{
"id": "HpXPHoyA4R",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 4,
"presentation": 4,
"summary": "Motivated by previous findings on the advantages of representation learning on a hypersphere, the authors propose a novel neural network architecture called the n... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;4",
"contribution_avg": 3,
"presentation": "2;4;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.838044"
} | {
"id": "M5oK5KvQlM",
"metareview": "This work proposes modifying the standard transformer architecture to normalize various aspects of the architecture (weight vectors, embeddings, etc) to lie on the unit sphere. The authors note that this allows for the matrix-vector multiplications to be interpreted as cosine-s... | {
"decision": "Accept (Poster)"
} |
semTHoVGsJ | 2410.05218v2 | Density estimation with LLMs: a geometric investigation of in-context learning trajectories | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) demonstrate remarkable emergent abilities to perform in-context learning across various tasks, including time series forecasting.\nThis work investigates LLMs’ ability to estimate probability density functions (PDFs) from data observed in-context;\nsu... | [
{
"id": "HhqlTw8fel",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper is looking at the classical problem of density estimation with the use of large language models. The question of interest here is how the large languag... | {
"rating": "3;6;6;8",
"rating_avg": 5.75,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "3;2;2;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.838812"
} | {
"id": "zNk7YBCegh",
"metareview": "This is an interesting paper that studies the capability of LLMs to perform density estimation when prompted with iid samples from a unknown distribution interpreting next token probabilities as an estimated PDF that can be interpreted as a kind of adaptive Gaussian estimation. ... | {
"decision": "Accept (Poster)"
} |
sfi2j1Ot6j | 2407.02263v4 | FreeCG: Free the Design Space of Clebsch-Gordan Transform for Machine Learning Force Fields | {
"content": "## Abstract\n\nAbstract Machine Learning Force Fields (MLFFs) are of great importance for chemistry, physics, materials science, and many other related fields. The Clebsch–Gordan Transform (CG transform) effectively encodes many-body interactions and is thus an important building block for many models o... | [
{
"id": "zx6kdJKQ47",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposes FreeCG, a method that implements the CG transform layer on the permutation-invariant abstract edges, which allows complete freedom in the desi... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;3;4;5",
"confidence_avg": 4,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "1;2;2;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.839595"
} | {
"id": "FniQdVtwBI",
"metareview": "This paper introduces a clever, highly expressive neural parameterization of force fields which obeys the requisite equivariances without increasing computational complexity. The new parameterization is then applied to a suite of challenging molecular benchmarks where it achiev... | {
"decision": "Accept (Poster)"
} |
sgHnfLX9Lt | 2310.13681v3 | Towards Realistic Mechanisms That Incentivize Federated Participation and Contribution | {
"content": "## Abstract\n\nAbstract Edge device participation in federating learning (FL) is typically studied through the lens of device-server communication ( e.g., device dropout) and assumes an undying desire from edge devices to participate in FL.\nAs a result, current FL frameworks are flawed when implemented... | [
{
"id": "0ZIGpqL4BT",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper introduces incentive mechanisms in federated learning (FL).",
"strengths": "REALFM incentivizes edge devices to participate in federated learning by... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "3;3;4;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;2",
"soundness_avg": 2.75,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.840378"
} | {
"id": "H5J316uaRJ",
"metareview": "a) Summary\n\nThis work introduces a novel federated learning mechanism designed to address the free-rider problem by incentivizing data contribution and device participation while maintaining individual rationality. It incorporates a non-linear relationship between model accura... | {
"decision": "Reject"
} |
sgaMYvGRG5 | 2406.11021v3 | $\alpha$-OCC: Uncertainty-Aware Camera-based 3D Semantic Occupancy Prediction | {
"content": "## Abstract\n\nAbstract In the realm of autonomous vehicle (AV) perception, comprehending 3D scenes is paramount for tasks such as planning and mapping. Camera-based 3D Semantic Occupancy Prediction (OCC) aims to infer scene geometry and semantics from limited observations. While it has gained popularit... | [
{
"id": "xTyQN1cxzx",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "In this paper, the authors focus on improving the Camera-based 3D semantic occupancy prediction (OCC) task by introducing the geometric the geometric uncertainty.... | {
"rating": "5;6;6;6;6",
"rating_avg": 5.8,
"confidence": "5;4;3;3;4",
"confidence_avg": 3.8,
"soundness": "3;3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;3;3",
"contribution_avg": 3,
"presentation": "3;2;3;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.841129"
} | {
"id": "kCXrkVvYTM",
"metareview": "This paper tackles semantic occupancy estimation from camera views. Given a camera view, the goal is to estimate the corresponding occupancy in a bounded, voxelized space. In this voxelized space, the network estimates the occupancy value for each voxel and a posterior over a (p... | {
"decision": "Reject"
} |
sgbI8Pxwie | 2410.11261v1 | Beyond Linear Approximations: A Novel Pruning Approach for Attention Matrix | {
"content": "## Abstract\n\nLarge Language Models (LLMs) have shown immense potential in enhancing various aspects of our daily lives, from conversational AI to search and AI assistants. However, their growing capabilities come at the cost of extremely large model sizes, making deployment on edge devices challenging... | [
{
"id": "jwlQ7P5PCH",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes a novel approach to pruning QK projection matrices by directly optimizing for an approximation of the attention matrix, bypassing traditional ... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "2;4;3;2",
"confidence_avg": 2.75,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.842352"
} | {
"id": "nqfpMjANlb",
"metareview": "This paper introduces a method for pruning query and key weight matrices in attention modules. The paper proposes a theoretical justification for the proposed scheme, with convergence guarantees. The method was initially evaluated on synthetic data, and as requested by the revie... | {
"decision": "Accept (Poster)"
} |
shqjOIK3SA | 2404.14309v2 | Towards Understanding the Robustness of Diffusion-Based Purification: A Stochastic Perspective | {
"content": "## Abstract\n\nAbstract Diffusion-Based Purification (DBP) has emerged as an effective defense mechanism against adversarial attacks. The efficacy of DBP has been attributed to the forward diffusion process, which narrows the distribution gap between clean and adversarial images through the addition of ... | [
{
"id": "MRHwrz5ECR",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 3,
"summary": "The author argues that the current effectiveness of the DBP method, attributed to the forward process of diffusion models, lacks sufficient empirical validation. ... | {
"rating": "6;6;6;8",
"rating_avg": 6.5,
"confidence": "5;4;4;3",
"confidence_avg": 4,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"contribution": "3;3;4;3",
"contribution_avg": 3.25,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.843662"
} | {
"id": "5zawN3wp9H",
"metareview": "This paper focuses on understanding diffusion-based purification which is an effective defense mechanism against adversarial attacks in recent literature. The paper identifies that the robustness provided by diffusion models is likely due to stochastic elements within the model ... | {
"decision": "Accept (Poster)"
} |
sicB10feCQ | 2410.04671v1 | CAR: Controllable Autoregressive Modeling for Visual Generation | {
"content": "## Abstract\n\nAbstract Controllable generation, which enables fine-grained control over generated outputs, has emerged as a critical focus in visual generative models.\nCurrently, there are two primary technical approaches in visual generation: diffusion models and autoregressive models. Diffusion mode... | [
{
"id": "PQz0nVDGFr",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The work proposes controllable autoregressive modeling (CAR), a conditional control module for VAR image generation. Following VAR, CAR adds control to pretrained... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;1;3;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.844977"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
sjWG7B8dvt | 2410.09102v1 | Instructional Segment Embedding: Improving LLM Safety with Instruction Hierarchy | {
"content": "## Abstract\n\nAbstract Large Language Models (LLMs) are susceptible to security and safety threats, such as prompt injection, prompt extraction, and harmful requests.\nOne major cause of these vulnerabilities is the lack of an instruction hierarchy.\nModern LLM architectures treat all inputs equally, f... | [
{
"id": "RETljdYa67",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 1,
"presentation": 3,
"summary": "To address robustness issues within Large Language Models (LLMs), authors introduce Instructional Segment Embedding (ISE), where they give the model hierarchical ... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "2;3;2;2",
"soundness_avg": 2.25,
"contribution": "1;2;3;3",
"contribution_avg": 2.25,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.845745"
} | {
"id": "Gz37JCQXoM",
"metareview": "The paper introduces Instructional Segment Embedding (ISE), a technique to enhance LLM safety by embedding instruction hierarchy information directly into the model architecture. ISE adds learnable embeddings to distinguish between different types of inputs (system messages, use... | {
"decision": "Accept (Poster)"
} |
skGSOcrIj7 | 2408.13885v1 | Neural Spacetimes for DAG Representation Learning | {
"content": "## Abstract\n\nAbstract We propose a class of trainable deep learning-based geometries called Neural Spacetimes (NSTs), which can universally represent nodes in weighted directed acyclic graphs (DAGs) as events in a spacetime manifold. While most works in the literature focus on undirected graph represe... | [
{
"id": "ZsDb2kjNwR",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents Neural SpaceTimes (NSTs), a class of trainable geometries that can encode nodes of weighted Directed Acyclic Graphs (DAGs) into a spacetime ma... | {
"rating": "5;6;6;8;8",
"rating_avg": 6.6,
"confidence": "4;2;3;3;3",
"confidence_avg": 3,
"soundness": "2;3;3;4;4",
"soundness_avg": 3.2,
"contribution": "2;3;3;3;4",
"contribution_avg": 3,
"presentation": "3;2;3;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.846824"
} | {
"id": "l1aZEtw5lS",
"metareview": "This paper introduces Neural SpaceTimes (NSTs), a novel framework for embedding directed acyclic graphs (DAGs) into continuous geometries that capture both edge weights and directionality. Unlike prior methods relying on fixed spacetime manifolds, NSTs dynamically learn the geom... | {
"decision": "Accept (Poster)"
} |
skHPtDnYGa | 2410.12329v1 | Understanding the Role of LLMs in Multimodal Evaluation Benchmarks | {
"content": "## Abstract\n\nAbstract The rapid advancement of Multimodal Large Language Models (MLLMs) has been accompanied by the development of various benchmarks to evaluate their capabilities.\nHowever, the true nature of these evaluations and the extent to which they assess multimodal reasoning versus merely le... | [
{
"id": "kBYf2BxkQW",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 1,
"presentation": 1,
"summary": "The paper proposes an examination of issues within multi-modal evaluation from two aspects:\n\n\t1.\tTo what extent answers rely solely on textual information\n\t... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "3;2;3;2",
"soundness_avg": 2.5,
"contribution": "2;2;2;1",
"contribution_avg": 1.75,
"presentation": "3;3;3;1",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.847880"
} | {
"id": "q7gvla4p2X",
"metareview": "Paper was reviewed by four expert reviewers and received 3 x marginally below the acceptance threshold and 1 x reject, not good enough ratings. Overall, all reviewers agree that the paper is lacking in a couple of core aspects, mainly: (1) the novelty is marginal, (2) it offers... | {
"decision": "Reject"
} |
sknUS8X9q0 | 2410.07456v1 | SAGE: Scalable Ground Truth Evaluations for Large Sparse Autoencoders | {
"content": "## Abstract\n\nAbstract A key challenge in interpretability is to decompose model activations into meaningful features. Sparse autoencoders (SAEs) have emerged as a promising tool for this task. However, a central problem in evaluating the quality of SAEs is the absence of ground truth features to serve... | [
{
"id": "ZHemDnv5Fj",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This work tries to scale up the feature evaluation problem for large sparse autoencoders (SAEs) in large language models, towards interpretable AI. The main contr... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "3;3;2;3",
"confidence_avg": 2.75,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "1;2;1;2",
"presentation_avg": 1.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.848607"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
slO3xTt4CG | 2410.02381v2 | MetaMetrics: Calibrating Metrics for Generation Tasks Using Human Preferences | {
"content": "## Abstract\n\nAbstract Understanding the quality of a performance evaluation metric is crucial for ensuring that model outputs align with human preferences. However, it remains unclear how well each metric captures the diverse aspects of these preferences, as metrics often excel in one particular area ... | [
{
"id": "FVqp6IifWe",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposed a method to align the automatic evaluation metrics with human preference evaluation, called MetaMetrics. Specifically, it's optimized to learn... | {
"rating": "3;5;6;6;6",
"rating_avg": 5.2,
"confidence": "4;3;3;4;3",
"confidence_avg": 3.4,
"soundness": "2;2;3;2;3",
"soundness_avg": 2.4,
"contribution": "2;2;3;2;3",
"contribution_avg": 2.4,
"presentation": "2;2;3;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.849301"
} | {
"id": "mEGDQKCGWL",
"metareview": "This paper proposed a metametrics that is more aligned with human preference on evaluating the generation tasks. The idea is quite simple, which can be summarized based on the rebuttal comment: \n1. Identify the downstream task, such as image captioning.\n2. Train MetaMetrics us... | {
"decision": "Accept (Poster)"
} |
smkspydzyN | 2406.14595v2 | Adversaries Can Misuse Combinations of Safe Models | {
"content": "## Abstract\n\nAbstract Developers try to evaluate whether an AI system can be misused by adversaries before releasing it; for example, they might test whether a model enables cyberoffense, user manipulation, or bioterrorism.\nIn this work, we show that individually testing models for misuse is inadequa... | [
{
"id": "LgOWSG2NY4",
"initial_rating": 3,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper studied adversaries to misuse combinations of AI models, even when each model is safe. The authors demonstrate that adversaries can decompose tasks in... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "5;3;4;4",
"confidence_avg": 4,
"soundness": "1;3;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.850080"
} | {
"id": "UPsMfFo4BJ",
"metareview": "The submission \"Adversaries Can Misuse Combinations of Safe Models\" provides the interesting observation that \"safety\" is not a monolithic tasks, but easily decomposed into subproblems. Reviewers all considered this an interesting observation, but from their reviews, had exp... | {
"decision": "Reject"
} |
sp9irsV1yq | 2410.16484v1 | Identifying Sub-networks in Neural Networks via Functionally Similar Representations | {
"content": "## Abstract\n\nAbstract Mechanistic interpretability aims to provide human-understandable insights into the inner workings of neural network models by examining their internals. Existing approaches typically require significant manual effort and prior knowledge, with strategies tailored to specific task... | [
{
"id": "zAZnQLGlO4",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper proposes a new metric based on the Gromov-Wasserstein distance to measure the similarity between layers in neural networks, which can be used for mecha... | {
"rating": "3;3;3;5;6",
"rating_avg": 4,
"confidence": "3;4;3;4;3",
"confidence_avg": 3.4,
"soundness": "2;1;2;3;3",
"soundness_avg": 2.2,
"contribution": "2;1;1;2;3",
"contribution_avg": 1.8,
"presentation": "2;2;2;3;2",
"presentation_avg": 2.2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.850823"
} | {
"id": "yNCDJLz4fp",
"metareview": "This paper introduces a method to automate mechanistic interpretability by identifying subnetworks that are functionally distinct/similar. This uses the GW distance, which allows matching representations across different intermediate representations. They perform experiments on ... | {
"decision": "Reject"
} |
sprjE7BTZR | 2410.14706v1 | Transformers are Efficient Compilers, Provably | {
"content": "## Abstract\n\nAbstract Transformer-based large language models (LLMs) have demonstrated surprisingly robust performance across a wide range of language-related tasks, including programming language understanding and generation. In this paper, we take the first steps towards a formal investigation of us... | [
{
"id": "R5GRZMpTRf",
"initial_rating": 3,
"confidence": 4,
"soundness": 1,
"contribution": 1,
"presentation": 1,
"summary": "The submission discusses a (set of proofs) that show that Transformers are able to perform some analysis tasks required during compilation with a bounded number o... | {
"rating": "1;3;5;8",
"rating_avg": 4.25,
"confidence": "3;4;4;3",
"confidence_avg": 3.5,
"soundness": "1;1;2;3",
"soundness_avg": 1.75,
"contribution": "1;1;4;3",
"contribution_avg": 2.25,
"presentation": "2;1;2;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.851819"
} | {
"id": "f1DVeb5SUN",
"metareview": "This paper proposes a novel framework for reasoning about whether transformers can represent standard compiler passes such as type inference. They provide a proof language called Cybertron to help formalize these proofs. They also demonstrate gaps between transformers and RNNs a... | {
"decision": "Reject"
} |
sr0My6yDNu | 2402.11196v2 | Maintaining Adversarial Robustness in Continuous Learning | {
"content": "## Abstract\n\nAbstract Adversarial robustness is essential for security and reliability of machine learning systems. However, adversarial robustness enhanced by defense algorithms is easily erased as the neural network’s weights update to learn new tasks. To address this vulnerability, it is essential ... | [
{
"id": "BBl5XiMfLv",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces Double Gradient Projection (DGP), a method designed to maintain adversarial robustness in continuous learning settings where neural networks ... | {
"rating": "1;3;3;6",
"rating_avg": 3.25,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "1;2;2;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.852984"
} | {
"id": "EZnuzQYSyr",
"metareview": "This work investigates maintaining adversarial robustness within the framework of continuous learning. During the rebuttal phase, reviewers raised several significant concerns. Firstly, there are practical applicability issues related to the proposed approach's high computationa... | {
"decision": "Reject"
} |
st7XqFgbAH | 2410.05434v1 | Better than Your Teacher: LLM Agents that learn from Privileged AI Feedback | {
"content": "## Abstract\n\nAbstract While large language models (LLMs) show impressive decision-making abilities, current methods lack a mechanism for automatic self-improvement from errors during task execution. We propose LEAP , an iterative fine-tuning framework that continually improves LLM agents using feedbac... | [
{
"id": "XqZc4ZLVNQ",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper proposes LEAP, an iterative fine-tuning framework that continually improves LLM agents using feedback from AI expert teachers. The key insight is to equ... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;2;3;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.854048"
} | {
"id": "ILkW6iVeKA",
"metareview": "This paper proposes an interesting framework to improve LLM agents by utilizing a LLM teacher with access to privileged information (the ground truth states that may be hidden in practice) to correct the reasoning and action trace of the student model, and iteratively fine-tune ... | {
"decision": "Accept (Poster)"
} |
suz4utPr9Y | 2406.06647v2 | How efficient is LLM-generated code? A rigorous & high-standard benchmark | {
"content": "## Abstract\n\nAbstract The emergence of large language models (LLMs) has significantly pushed the frontiers of program synthesis. Advancement of LLM-based program synthesis calls for a thorough evaluation of LLM-generated code. Most evaluation frameworks focus on the (functional) correctness of generat... | [
{
"id": "hgO12ZNVe0",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "The paper presents a systematic benchmark for efficiency of LLM generated code. The paper presents the eff@k metric, which captures efficiency of generated code (... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "3;3;4;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.856059"
} | {
"id": "T0kVkHZohJ",
"metareview": "The paper introduces a new metric (and Monte Carlo estimator) for the expected performance of a program sampled from a language model. They use this new theoretical advance to then do a very broad evaluation of language models, and a very narrow evaluation of programming problem... | {
"decision": "Accept (Poster)"
} |
svp1EBA6hA | 2406.12120v1 | Adding Conditional Control to Diffusion Models with Reinforcement Learning | {
"content": "## Abstract\n\nAbstract Diffusion models are powerful generative models that allow for precise control over the characteristics of the generated samples. While these diffusion models trained on large datasets have achieved success, there is often a need to introduce additional controls in downstream fin... | [
{
"id": "pBKgL5YHaQ",
"initial_rating": 5,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces an RL-based approach for integrating conditional control into pre-trained diffusion models. The primary contribution of the proposed algorit... | {
"rating": "5;5;5;8",
"rating_avg": 5.75,
"confidence": "2;3;2;4",
"confidence_avg": 2.75,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;2;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.856939"
} | {
"id": "UdQAkDX4ZV",
"metareview": "This paper outlines a way to fine-tune a diffusion model, allowing conditioning on additional information. The proposed method is pragmatic because it relies on standard implementations of fine-tuning, based on reinforcement learning. The authors demonstrate that the proposed m... | {
"decision": "Accept (Poster)"
} |
sw6Wpx2LGr | 2403.10492v3 | Mitigating Dialogue Hallucination for Large Vision Language Models via Adversarial Instruction Tuning | {
"content": "## Abstract\n\nAbstract Mitigating hallucinations of Large Vision Language Models (LVLMs) is crucial to enhance their reliability for general-purpose assistants.\nThis paper shows that such hallucinations of LVLMs can be significantly exacerbated by preceding user-system dialogues.\nTo precisely measure... | [
{
"id": "2JaiyZQMtF",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper addresses the issue of dialogue hallucination in LVLMs, where preceding dialogues can lead to unreliable outputs. \n\nIt introduces a benchmark dataset... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "3;2;4;3",
"confidence_avg": 3,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.857778"
} | {
"id": "2H4aw6Lx02",
"metareview": "### Summary:\nThis paper proposes a method to mitigate dialogue hallucination in Large Vision Language Models (LVLMs) through:\n1. A benchmark called EvalDial that extends existing datasets with adversarial dialogues\n2. An Adversarial Instruction Tuning (AIT) approach to make L... | {
"decision": "Reject"
} |
swWF948IiC | 2406.07687v1 | Adversarial Machine Unlearning | {
"content": "## Abstract\n\nAbstract This paper focuses on the challenge of machine unlearning, aiming to remove\nthe influence of specific training data on machine learning models.\nTraditionally, the development of unlearning algorithms runs parallel with that of membership inference attacks (MIA), a type of priva... | [
{
"id": "rlq6ApCX7M",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper formulates machine unlearning as a Stackelberg Game, where the model owner seeks to unlearn a forget set while preserving the model's utility, and the ... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;3;5;4",
"confidence_avg": 4,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.858475"
} | {
"id": "BTiRp137YR",
"metareview": "This paper proposes a game-theoretic framework that integrates MIAs into the design of unlearning algorithms, which is novel and interesting to model the unlearning problem as a Stackelberg game in which an unlearner strives to unlearn specific training data from a model, while ... | {
"decision": "Accept (Poster)"
} |
sx2jXZuhIx | 2407.00367v1 | SVG: 3D Stereoscopic Video Generation via Denoising Frame Matrix | {
"content": "## Abstract\n\nAbstract Video generation models have demonstrated great capabilities of producing impressive monocular videos, however, the generation of 3D stereoscopic video remains under-explored. We propose a pose-free and training-free approach for generating 3D stereoscopic videos using an off-the... | [
{
"id": "nxCxsR2yCB",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper aims to synthesize 3D stereoscopic videos using an off-the-shelf monocular video generation model without finetuning. Concretely, the authors leverage ... | {
"rating": "5;6;8",
"rating_avg": 6.333333333333333,
"confidence": "5;4;3",
"confidence_avg": 4,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "1;3;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.859186"
} | {
"id": "Wz47MXsqnd",
"metareview": "Summary: This paper presents a method for synthesizing 3D stereoscopic videos using a pre-trained monocular video generation model. The basic idea is to create a warped version of a video using estimated video depth and inpaint it using a video generation model. The core contrib... | {
"decision": "Accept (Poster)"
} |
sy1lbQxj9J | 2404.18444v2 | U-Nets as Belief Propagation: Efficient Classification, Denoising, and Diffusion in Generative Hierarchical Models | {
"content": "## Abstract\n\nAbstract U-Nets are among the most widely used architectures in computer vision, renowned for their exceptional performance in applications such as image segmentation, denoising, and diffusion modeling. However, a theoretical explanation of the U-Net architecture design has not yet been f... | [
{
"id": "mhVKM8Knra",
"initial_rating": 5,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper examines the widely-used U-Net architecture under the scope of generative hierarchical models. The authors present a theoretical framework that treats t... | {
"rating": "5;5;6;6;8",
"rating_avg": 6,
"confidence": "2;2;2;1;4",
"confidence_avg": 2.2,
"soundness": "2;3;3;3;4",
"soundness_avg": 3,
"contribution": "2;2;3;3;3",
"contribution_avg": 2.6,
"presentation": "3;2;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.860564"
} | {
"id": "olFj8cPffi",
"metareview": "This work examines graphical models and the U-Net architecture in simplified form in order to theoretically relate the two. In particular, or a restricted setting, the U-Net architecture and computation can be connected with belief propagation and approximation bounds are derive... | {
"decision": "Accept (Poster)"
} |
syMZF5fc8y | 2410.21043v1 | Disentangled and Self-Explainable Node Representation Learning | {
"content": "## Abstract\n\nAbstract Node representations, or embeddings, are low-dimensional vectors that capture node properties, typically learned through unsupervised structural similarity objectives or supervised tasks. While recent efforts have focused on explaining graph model decisions, the interpretability ... | [
{
"id": "0gbCMd1hcK",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces **DISENE (Disentangled and Self-Explainable Node Embedding)**, a novel framework designed to generate self-explainable and disentangled node... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;2;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.862082"
} | {
"id": "cvoJpguAdm",
"metareview": "This paper proposes a new framework named DISENE (Disentangled and Self-Explainable Node Embedding) for generating self-explainable and disentangled node embeddings in graph-based learning tasks. Reviewers agreed that this paper brings an interesting perspective by considering n... | {
"decision": "Reject"
} |
syThiTmWWm | 2410.07137v1 | Cheating Automatic LLM Benchmarks: Null Models Achieve High Win Rates | {
"content": "## Abstract\n\nAbstract Automatic LLM benchmarks, such as AlpacaEval 2.0, Arena-Hard-Auto, and MT-Bench, have become popular for evaluating language models due to their cost-effectiveness and scalability compared to human evaluation. Achieving high win rates on these benchmarks can significantly boost t... | [
{
"id": "qQV6SH8BF4",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "This paper investigates vulnerabilities in automatic LLM benchmarks (like AlpacaEval 2.0, Arena-Hard-Auto, and MT-Bench) by demonstrating that even a \"null model... | {
"rating": "5;6;6;6;6;8;8;8",
"rating_avg": 6.625,
"confidence": "3;3;3;4;3;3;4;4",
"confidence_avg": 3.375,
"soundness": "3;4;3;4;3;3;4;4",
"soundness_avg": 3.5,
"contribution": "2;4;3;3;2;3;4;4",
"contribution_avg": 3.125,
"presentation": "4;3;2;4;3;3;4;4",
"presentation_avg": 3.375
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.862872"
} | {
"id": "5YEZaYnbNE",
"metareview": "This paper shows that a constant response can game certain automatic LLM benchmarks. This contribution will make many in the community rethink LLM benchmarking which is topical right now. The reviewers unanimously vote for acceptance, and I agree. Reviewers raised several wea... | {
"decision": "Accept (Oral)"
} |
t1LfiWCYux | 2408.11748v3 | Understanding Depth and Height Perception in Large Visual-Language Models | {
"content": "## Abstract\n\nAbstract Geometric understanding is crucial for navigating and interacting with our environment. While large Vision Language Models (VLMs) demonstrate impressive capabilities, deploying them in real-world scenarios necessitates a comparable geometric understanding in visual perception. In... | [
{
"id": "EPAVNMMkcb",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This work proposes a benchmark to evaluate large Vision and Language Models (VLMs) on their capabilities in perceiving and understanding height and depth from sin... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;4;5;4",
"confidence_avg": 4.25,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;1;2;2",
"contribution_avg": 1.75,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.863588"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
t21RmVmJrT | 2410.08414v1 | Understanding the Interplay between Parametric and Contextual Knowledge for Large Language Models | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) encode vast amounts of knowledge during pre-training (parametric knowledge, or PK ) and can further be enhanced by incorporating contextual knowledge ( CK ).\nCan LLMs effectively integrate their internal PK with external CK to solve complex problems?... | [
{
"id": "BIUK7SbG7t",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper analyzes interactions between LLM parametric knowledge and contextual knowledge. It finds that parametric knowledge often gets suppressed when contextua... | {
"rating": "3;3;5;6;6",
"rating_avg": 4.6,
"confidence": "4;4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;2;3;3",
"soundness_avg": 2.4,
"contribution": "1;2;2;2;3",
"contribution_avg": 2,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.864337"
} | {
"id": "zMCr4WuERP",
"metareview": "This paper investigates the interactions between parametric knowledge (PK) and contextual knowledge (CK) in large language models (LLMs) through the introduction of EchoQA, a benchmark that categorizes PK-CK relationships into four types: Supportive, Complementary, Conflicting, ... | {
"decision": "Reject"
} |
t3rdi80xCz | 2405.18100v1 | A Pontryagin Perspective on Reinforcement Learning | {
"content": "## Abstract\n\nAbstract Reinforcement learning has traditionally focused on learning state-dependent policies to solve optimal control problems in a closed-loop fashion.\nIn this work, we introduce the paradigm of open-loop reinforcement learning where a fixed action sequence is learned instead.\nWe pre... | [
{
"id": "8DyH63YSch",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The authors present a new framework for incorporating control theoretic methods into reinforcement learning (RL). They propose the use of the Pontryagin’s princip... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.865144"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
t5FD4QTDTu | 2410.08421v1 | Generalizable autoregressive modeling of time series through functional narratives | {
"content": "## Abstract\n\nAbstract Time series data are inherently functions of time, yet current transformers often learn time series by modeling them as mere concatenations of time periods, overlooking their functional properties. In this work, we propose a novel objective for transformers that learn time series... | [
{
"id": "gNAK1EN8oT",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents Narratives of Time Series (NoTS), a transformer-based framework that reinterprets time series data as temporal functions, capturing both local... | {
"rating": "3;3;5;5;8",
"rating_avg": 4.8,
"confidence": "4;3;4;3;3",
"confidence_avg": 3.4,
"soundness": "3;2;2;2;3",
"soundness_avg": 2.4,
"contribution": "3;2;2;2;3",
"contribution_avg": 2.4,
"presentation": "4;2;2;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.865944"
} | {
"id": "LZ6PohGKp6",
"metareview": "The paper proposes NoTS, a transformer-based method for pretraining time series data by reinterpreting them as sequences of functions at different levels of detail. This is achieved through \"degradation operators\" that progressively smooth the time series. NoTS uses an autoreg... | {
"decision": "Reject"
} |
t717joHHSc | 2406.02536v2 | Mitigate Position Bias in Large Language Models via Scaling a Single Dimension | {
"content": "## Abstract\n\nAbstract Large Language Models (LLMs) are increasingly applied in various real-world scenarios due to their excellent generalization capabilities and robust generative abilities. However, they exhibit position bias, also known as \"lost in the middle\", a phenomenon that is especially pro... | [
{
"id": "IqmZcR4Zk8",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This work analyzes the role of attention mechanism in position biases of LLMs and further proposes a novel method to mitigate position bias. The main findings in... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "3;3;4;4",
"confidence_avg": 3.5,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.866731"
} | {
"id": "nyHbEpmYIi",
"metareview": "The paper proposes an approach to identify positional hidden states (defined as dimensions that change monotonically wrt position) and manipulate their scale to address lost in the middle phenomenon. Overall, the reviewers raised several issues, including (1) flawed experiment s... | {
"decision": "Reject"
} |
t7P5BUKcYv | 2410.07348v1 | MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts | {
"content": "## Abstract\n\nAbstract In this work, we aim to simultaneously enhance the effectiveness and efficiency of Mixture-of-Experts (MoE) methods. To achieve this, we propose MoE++, a general and heterogeneous MoE framework that integrates both Feed-Forward Network (FFN) and zero-computation experts. Specific... | [
{
"id": "dgHhgbm85Y",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces the MoE++, which adds zero-computation experts to enhance the efficiency of computation.\nIt utilizes zero-computation experts to minimize o... | {
"rating": "5;6;8",
"rating_avg": 6.333333333333333,
"confidence": "4;3;5",
"confidence_avg": 4,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "3;3;4",
"presentation_avg": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.867417"
} | {
"id": "aqjGQtvbDj",
"metareview": "This paper proposes MoE++, a heterogeneous MoE framework that enhances both the efficiency and effectiveness of MoE methods by integrating FFN and zero-computation experts (zero, copy, and constant experts). This design reduces computing overhead by dynamically assigning simpler... | {
"decision": "Accept (Oral)"
} |
t8ctvylFn7 | 2405.15454v1 | Linearly Controlled Language Generation with Performative Guarantees | {
"content": "## Abstract\n\nAbstract The increasing prevalence of Large Language Models (LMs) in critical applications\nhighlights the need for controlled language generation strategies that are not only computationally efficient but that also enjoy performance guarantees.\nTo achieve this, we use a common model of ... | [
{
"id": "19Z6ZVcPbZ",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposed an interesting, gradient-free method to control the language generation. The authors introduce a method called LiSeCo, where they first train l... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.868123"
} | {
"id": "1IXNicV652",
"metareview": "This paper proposes a method for controllable text generation. The approach first trains a linear probe to classify attributes from hidden states, and then derives a closed-form solution for perturbing those states to ensure that the predicted attribute probabilities are within ... | {
"decision": "Reject"
} |
t8qcGXaepr | 2410.07819v1 | Uncovering Overfitting in Large Language Model Editing | {
"content": "## Abstract\n\nAbstract Knowledge editing has been proposed as an effective method for updating and correcting the internal knowledge of Large Language Models (LLMs). However, existing editing methods often struggle with complex tasks, such as multi-hop reasoning. In this paper, we identify and investig... | [
{
"id": "MrJcVdf1NF",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "In this paper, the authors focus on investigating parameter-modifying knowledge editing methods and find that existing parameter-modifying knowledge editing appro... | {
"rating": "6;8;8",
"rating_avg": 7.333333333333333,
"confidence": "3;4;3",
"confidence_avg": 3.3333333333333335,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "4;3;3",
"contribution_avg": 3.3333333333333335,
"presentation": "4;4;3",
"presentation_avg": 3.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.868884"
} | {
"id": "stlPcG6EuK",
"metareview": "This paper investigates parameter-modifying knowledge editing methods and identifies that existing approaches exhibit overfitting. Specifically, when questions related to the edited subject and relation arise, the model often responds with the edited object, even if it is not a... | {
"decision": "Accept (Spotlight)"
} |
t9JUTS9ADL | 2409.08264v2 | Windows Agent Arena: Evaluating Multi-Modal OS Agents at Scale | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) show remarkable potential to act as computer agents, enhancing human productivity and software accessibility in multi-modal tasks that require planning and reasoning.\nHowever, measuring agent performance in realistic environments remains a challenge ... | [
{
"id": "DPi1ANYI79",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces WindowsAgentArena, a comprehensive benchmarking environment tailored for evaluating multi-modal agents in Windows OS settings. This platform ... | {
"rating": "5;5;5;6;6",
"rating_avg": 5.4,
"confidence": "4;4;4;2;4",
"confidence_avg": 3.6,
"soundness": "2;3;3;3;3",
"soundness_avg": 2.8,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "2;4;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.869687"
} | {
"id": "e7NnHqyn27",
"metareview": "This paper presents a Windows-centric extension to OSWorld, with 150 Windows-specific tasks. On top of that, it provides architecture to run these tasks in parallel, and also on relatively cheap CPU-only hardware + (what I think is Azure-specific?) cloud calls to e.g. OpenAI mo... | {
"decision": "Reject"
} |
t9l63huPRt | 2312.12540v4 | Lightning-Fast Image Inversion and Editing for Text-to-Image Diffusion Models | {
"content": "## Abstract\n\nAbstract Diffusion inversion is the problem of taking an image and a text prompt that describes it and finding a noise latent that would generate the exact same image.\nMost current deterministic inversion techniques operate by approximately solving an implicit equation and may converge s... | [
{
"id": "5HhjIisKEh",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper addresses diffusion inversion for fast and accurate inversion within text-to-image diffusion models. The authors propose Guided Newton-Raphson Inversio... | {
"rating": "3;6;6;8",
"rating_avg": 5.75,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "3;4;3;3",
"soundness_avg": 3.25,
"contribution": "1;4;3;3",
"contribution_avg": 2.75,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.870770"
} | {
"id": "GgvcmCxlIy",
"metareview": "This paper proposed an image inverse and editing method based on the Newton-Raphson (NR) algorithm to find root of the implicit equation of inversion for diffusion model. It designed a fixed-point iteration approach enabling faster convergence, guided towards the in-distributio... | {
"decision": "Accept (Poster)"
} |
t9lS1lX9FQ | 2405.16435v2 | Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning | {
"content": "## Abstract\n\nAbstract We present a novel end-to-end framework that generates highly compact (typically 6-15 dimensions), discrete (int4 type), and interpretable node representations—termed node identifiers (node IDs)—to tackle inference challenges on large-scale graphs. By employing vector quantizatio... | [
{
"id": "9TOtuyAyvP",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a vector quantization for GNN representations created from different \"depths\" of the network at each node. Its similar to other recently pr... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "4;4;4",
"confidence_avg": 4,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.871655"
} | {
"id": "D1SMxQGbQe",
"metareview": "The paper introduces a vector quantization (VQ) method for GNN representations, particularly leveraging node embeddings from varying depths of the network. The method is distinct from recent works (e.g., VQGraph) in not relying on a reconstruction loss. The authors provide exten... | {
"decision": "Accept (Poster)"
} |
tAGmxz1TUi | 2410.11439v1 | A Simple Approach to Unifying Diffusion-based Conditional Generation | {
"content": "## Abstract\n\nAbstract Recent progress in image generation has sparked research into controlling these models through condition signals, with various methods addressing specific challenges in conditional generation. Instead of proposing another specialized technique, we introduce a simple, unified fram... | [
{
"id": "OSHfHmb9Fc",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 2,
"summary": "The paper presents a diffusion model-based conditional image generation method as an add-on to current latent diffusion model (e.g., SD1.5). The model consists of... | {
"rating": "5;6;6;6;6",
"rating_avg": 5.8,
"confidence": "4;3;2;4;4",
"confidence_avg": 3.4,
"soundness": "3;3;3;3;4",
"soundness_avg": 3.2,
"contribution": "3;2;3;3;3",
"contribution_avg": 2.8,
"presentation": "3;3;3;3;2",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.872437"
} | {
"id": "IxBU9Kt4da",
"metareview": "This paper introduces a framework for handling diverse conditional generation tasks by learning a joint distribution over correlated image pairs with diffusion models. Parallel model branches employ joint cross-attention blocks such that branches can attend to each other and ena... | {
"decision": "Accept (Poster)"
} |
tBZK9BI2GZ | 2411.07722v1 | Is Cognition consistent with Perception? Assessing and Mitigating Multimodal Knowledge Conflicts in Document Understanding | {
"content": "## Abstract\n\nAbstract Multimodal large language models (MLLMs) have shown impressive capabilities in document understanding, a rapidly growing research area with significant industrial demand in recent years.\nAs a multimodal task, document understanding requires models to possess both perceptual and ... | [
{
"id": "84rJ9K8VyR",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper explores the capability of large multimodal models in understanding document images. The authors highlight that current multimodal large models often p... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "2;2;3;2",
"soundness_avg": 2.25,
"contribution": "2;3;2;2",
"contribution_avg": 2.25,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.873125"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
tCYdsuQgZZ | 2407.05061v1 | Test-time Contrastive Concepts for Open-World Semantic Segmentation | {
"content": "## Abstract\n\nAbstract Recent VLMs, pre-trained on large amounts of image-text pairs to align both modalities, have opened the way to open-vocabulary semantic segmentation. Given an arbitrary set of textual queries, image regions are assigned the closest query in feature space. However, the usual setup... | [
{
"id": "geuIF4KIse",
"initial_rating": 6,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper discussed the 'background' concept in open-world semantic segmentation and proposed a method to suppress background with contrastive concepts when segme... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "5;4;5",
"confidence_avg": 4.666666666666667,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "2;3;3",
"presentation_avg": 2.666666666666666... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.873956"
} | {
"id": "nBUeJpJyMf",
"metareview": "This paper introduces a test-time method for open-vocabulary semantic segmentation (OVSS) that suppresses false positive segments using contrastive concepts (CC) when segmenting a specific category in test time. The authors propose generating contrastive concepts at test time us... | {
"decision": "Reject"
} |
tCfvktlrHI | 2406.18872v1 | Efficacy of Language Model Self-Play in Non-Zero-Sum Games | {
"content": "## Abstract\n\nAbstract Game-playing agents like AlphaGo have achieved superhuman performance through self-play, which is theoretically guaranteed to yield optimal policies in competitive games. However, most language tasks are partially or fully cooperative, so it is an open question whether techniques... | [
{
"id": "a1szwS0h12",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This work proposes to study how LLMs learn in competitive games. They choose to investigate using mostly GPT-3.5 and the game Deal or No Deal (DoND) where two pla... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "3;2;3;4",
"soundness_avg": 3,
"contribution": "2;3;3;4",
"contribution_avg": 3,
"presentation": "3;2;4;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.875011"
} | {
"id": "OW3eyK945J",
"metareview": "This paper explores the application of self-play training to language models in non-zero-sum settings through experiments in the Deal or No Deal negotiation game.\n\nThe reviewers agree that the proposed self-play training on large language models is a novel approach. However, t... | {
"decision": "Reject"
} |
tDIL7UXmSS | 2405.13351v1 | Quantum (Inspired) $D^2$-sampling with Applications | {
"content": "## Abstract\n\nAbstract D 2 superscript 𝐷 2 D^{2} italic_D start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT -sampling is a fundamental component of sampling-based clustering algorithms such as k 𝑘 k italic_k -means++.\nGiven a dataset V ⊂ ℝ d 𝑉 superscript ℝ 𝑑 V\\subset\\mathbb{R}^{d} italic_V ⊂ blackboa... | [
{
"id": "85k3oaiUba",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper provides a quantum algorithm as well as a dequantized (i.e quantum inspired classical) version of it for the problem of $D^2$ sampling. This type of sa... | {
"rating": "6;6;8",
"rating_avg": 6.666666666666667,
"confidence": "2;3;3",
"confidence_avg": 2.6666666666666665,
"soundness": "3;4;3",
"soundness_avg": 3.3333333333333335,
"contribution": "3;3;4",
"contribution_avg": 3.3333333333333335,
"presentation": "3;3;4",
"presentation_avg": 3.33333333333333... | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.876116"
} | {
"id": "WHS4OnCxIi",
"metareview": "This paper gave a quantum algorithm for D^2-sampling in clustering algorithms. The main contribution is on the theoretical side, in the sense that the proposed quantum algorithm has provable speedup over existing classical algorithms. Numerical experiments also demonstrated the ... | {
"decision": "Accept (Poster)"
} |
tFDTHA3odg | 2312.07062v2 | ThinkBot: Embodied Instruction Following with Thought Chain Reasoning | {
"content": "## Abstract\n\nAbstract Embodied Instruction Following (EIF) requires agents to complete human instruction by interacting objects in complicated surrounding environments. Conventional methods directly consider the sparse human instruction to generate action plans for agents, which usually fail to achiev... | [
{
"id": "LkTCewMyTz",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces ThinkBot, an agent designed for Embodied Instruction Following (EIF) that effectively executes human instructions in complex environments by ... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;2;4;4",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.878440"
} | {
"id": "RJKsHveBO5",
"metareview": "The paper proposes ThinkBot, a framework for embodied instruction following (EIF) that uses chain-of-thought-like reasoning to follow instructions in which some of the actions may not be explicitly conveyed. The \"instruction completer\" uses a large language model (LLM) to infe... | {
"decision": "Accept (Poster)"
} |
tFV5GrWOGm | 2410.08368v1 | ElasticTok: Adaptive Tokenization for Image and Video | {
"content": "## Abstract\n\nAbstract Efficient video tokenization remains a key bottleneck in learning general purpose vision models that are capable of processing long video sequences. Prevailing approaches are restricted to encoding videos to a fixed number of tokens, where too few tokens will result in overly los... | [
{
"id": "Wiat8Y8cK5",
"initial_rating": 6,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 1,
"summary": "- The paper presents an approach to dynamically assign different number of tokens per different image/video depending on the input complexity.\n- During training,... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "3;3;3;5",
"confidence_avg": 3.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;3;1",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.878993"
} | {
"id": "KiH4HmfIMs",
"metareview": "The paper proposes an adaptive approach to encoding frames into tokens to substantially reduce the number of tokens to be processed in downstream tasks. \n\nReviewers appreciated the novel application of adaptive tokenization to auto-encoding images and videos. They considered t... | {
"decision": "Accept (Poster)"
} |
tFwEsrx1hm | 2407.06071v1 | From Loops to Oops: Fallback Behaviors of Language Models Under Uncertainty | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) often exhibit undesirable behaviors, such as hallucinations and sequence repetitions.\nWe propose to view these behaviors as fallbacks that models exhibit under uncertainty, and investigate the connection between them.\nWe categorize fallback behavior... | [
{
"id": "nbrYgidbvT",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 4,
"summary": "This paper looks at “fallback” behaviors in large language models in the face of epistemic uncertainty, specifically focusing on repetitive text, degenerate phras... | {
"rating": "5;5;5;8",
"rating_avg": 5.75,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;2;4;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.879731"
} | {
"id": "RNB6Qt68BY",
"metareview": "The paper investigates fallback behaviours in large language models (LLMs) under epistemic uncertainty, categorizing them into sequence repetitions, degenerate text, and hallucinations, and proposes a \"fallback hierarchy\" that connects these behaviours. Experiments are perform... | {
"decision": "Reject"
} |
tH12wjcuXx | 2410.00759v1 | Targeted synthetic data generation for tabular data via hardness characterization | {
"content": "## Abstract\n\nAbstract Synthetic data generation has been proven successful in improving model performance and robustness in the context of scarce or low-quality data.\nUsing the data valuation framework to statistically identify beneficial and detrimental observations, we introduce a novel augmentatio... | [
{
"id": "ArktuqDFSJ",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes a novel synthetic data augmentation approach focused on enhancing binary classification models, particularly on tabular data. It introduces a t... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "5;3;2;3",
"confidence_avg": 3.25,
"soundness": "2;1;2;3",
"soundness_avg": 2,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "4;2;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.880566"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
tJE9WeqHEI | 2405.08707v1 | Beyond Scaling Laws: Understanding Transformer Performance with Associative Memory | {
"content": "## Abstract\n\nAbstract Increasing the size of a Transformer model does not always lead to enhanced performance. This phenomenon cannot be explained by the empirical scaling laws. Furthermore, improved generalization ability occurs as the model memorizes the training samples.\nWe present a theoretical f... | [
{
"id": "eVXlwE3vxh",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces a theoretical framework for studying *causal* Transformer models and their scaling properties from the perspective of Associative Memories. ... | {
"rating": "3;5;8;8",
"rating_avg": 6,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "1;4;4;4",
"soundness_avg": 3.25,
"contribution": "1;3;3;3",
"contribution_avg": 2.5,
"presentation": "1;2;3;4",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.881209"
} | {
"id": "b4We6xiDFo",
"metareview": "The authors present a mostly theoretical analysis of how performance in Transformers scales with model size, based on an approximation of Transformers as a sequence of energy-based models. The reviewers agreed the analysis was interesting and novel, but two of the reviewers took... | {
"decision": "Reject"
} |
tJHDw8XfeC | 2410.17215v2 | MiniPLM: Knowledge Distillation for Pre-training Language Models | {
"content": "## Abstract\n\nAbstract Knowledge distillation (KD) is widely used to train small, high-performing student language models (LMs) using large teacher LMs.\nWhile effective in fine-tuning, KD during pre-training faces challenges in efficiency, flexibility, and effectiveness.\nExisting methods either incur... | [
{
"id": "TIArboBBcv",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper studies the field of language model pre-training by proposing an efficient KD framework. MINIPLM’s use of offline inference and Difference Sampling stan... | {
"rating": "5;6;6;6;8",
"rating_avg": 6.2,
"confidence": "3;3;3;4;3",
"confidence_avg": 3.2,
"soundness": "3;3;3;3;4",
"soundness_avg": 3.2,
"contribution": "2;3;3;3;3",
"contribution_avg": 2.8,
"presentation": "3;4;3;3;4",
"presentation_avg": 3.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.882017"
} | {
"id": "JEIiSTkjOP",
"metareview": "The paper introduces MINIPLM, a method for simplifying large language models. MINIPLM uses ``Difference Sampling'' to find difficult training examples where a large teacher model and a smaller reference model give different answers. When trained under the same conditions, MINIPL... | {
"decision": "Accept (Poster)"
} |
tJoS2d0Onf | 2407.17470v1 | SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistency | {
"content": "## Abstract\n\nAbstract We present Stable Video 4D (SV4D) — a latent video diffusion model for multi-frame and multi-view consistent dynamic 3D content generation.\nUnlike previous methods that rely on separately trained generative models for video generation and novel view synthesis, we design a unifie... | [
{
"id": "WiA2fYnx57",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "Stable Video 4D (SV4D) introduces a unified latent video diffusion model for generating multi-frame and multi-view consistent 4D content. Unlike previous methods ... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "4;4;4;5",
"confidence_avg": 4.25,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.882828"
} | {
"id": "VbG9qZSEzc",
"metareview": "This paper receives ratings of 5,6,8,6. The AC follows the suggestions of the reviewers to accept the paper. The paper is well-written and is easy to follow. The proposed method of mix-sampling strategy, and the use of view attention and frame attention blocks are effective in s... | {
"decision": "Accept (Poster)"
} |
tKnPtyDt6H | 2410.05952v1 | Active Evaluation Acquisition for Efficient LLM Benchmarking | {
"content": "## Abstract\n\nAbstract As large language models (LLMs) become increasingly versatile, numerous large scale benchmarks have been developed to thoroughly assess their capabilities. These benchmarks typically consist of diverse datasets and prompts to evaluate different aspects of LLM performance. However... | [
{
"id": "pkAMeDqWUe",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposes the Adaptive Evaluation Acquisition (AEA) approach which is basically an adaptive subset selection of prompts for a model so that the generali... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;3;3;4",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.883439"
} | {
"id": "IFdx3m5kf6",
"metareview": "The paper studies strategies to improve evaluation efficiency by actively selecting a subset of prompts to evaluate and then predicting the performance of the rest of the prompts. The paper empirically demonstrates that the proposed approach can reduce the number of evaluation p... | {
"decision": "Reject"
} |
tL8dpJmECp | 2405.13977v3 | Improving Fairness and Mitigating MADness in Generative Models | {
"content": "## Abstract\n\nAbstract We propose autophagy p enalized l ikelihood e stimation (PLE), an unbiased alternative to maximum likelihood estimation (MLE) which is more fair and less susceptible to model autophagy disorder (MADness). Model autophagy refers to models trained on their own output; PLE ensures t... | [
{
"id": "UaTC6qwIHs",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 1,
"summary": "The paper proposes penalized autophogy estimation (PLE), which is a method that impproves fairness and reduce MADness of generative models.\nThe key idea is to ma... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "3;3;2;3",
"confidence_avg": 2.75,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;2;1",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.884258"
} | {
"id": "XLQ4Fhrtel",
"metareview": "The paper aims to address the problem of representation bias in generative models, which is the models’ decreased performance when trained using their own synthetic outputs. The main idea is to use a regularization to make sure that model parameters are stable during training. T... | {
"decision": "Reject"
} |
tLFWU6izoA | 2407.20171v4 | Diffusion Feedback Helps CLIP See Better | {
"content": "## Abstract\n\nAbstract Contrastive Language-Image Pre-training (CLIP), which excels at abstracting open-world representations across domains and modalities, has become a foundation for a variety of vision and multimodal tasks.\nHowever, recent studies reveal that CLIP has severe visual shortcomings, su... | [
{
"id": "kAM6inLj8a",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper aims to improve some CLIP's shortcoming cases, e.g., understanding the quantity,color,structure of images, since CLIP is trained by focusing on the high... | {
"rating": "3;5;5;5;8",
"rating_avg": 5.2,
"confidence": "3;5;4;4;3",
"confidence_avg": 3.8,
"soundness": "2;3;2;3;3",
"soundness_avg": 2.6,
"contribution": "3;3;2;3;3",
"contribution_avg": 2.8,
"presentation": "1;4;2;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.884972"
} | {
"id": "g8GD9xV2Xb",
"metareview": "This paper introduces DIVA, a post-training technique for CLIP-like models to overcome some of their shortcomings using a self-supervised diffusion approach that is trained from images only, using the CLIP feature to condition the diffusion model. This approach improves performa... | {
"decision": "Accept (Poster)"
} |
tNvCSw8ONp | 2409.18857v1 | Mitigating Selection Bias with Node Pruning and Auxiliary Options | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) often show unwarranted preference for certain choice options when responding to multiple-choice questions, posing significant reliability concerns in LLM-automated systems.\nTo mitigate this selection bias problem, previous solutions utilized debiasin... | [
{
"id": "MS23Yj3QAQ",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The authors find that large language models exhibit a selection bias in question answering (QA) tasks, where the choice of options influences their outputs. To ad... | {
"rating": "1;5;5;6",
"rating_avg": 4.25,
"confidence": "4;3;3;4",
"confidence_avg": 3.5,
"soundness": "1;2;3;3",
"soundness_avg": 2.25,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.885785"
} | {
"id": "0Eu2pwAcNF",
"metareview": "The paper shows certain value in terms of innovation and experimental verification, but the deficiencies in experimental design and analysis, as well as writing and expression, have affected the quality of the paper, which suggests a reject.\nThe authors need to further improve ... | {
"decision": "Reject"
} |
tNxr38vfYR | 2410.14072v1 | Efficient Vision-Language Models by Summarizing Visual Tokens into Compact Registers | {
"content": "## Abstract\n\nAbstract Recent advancements in vision-language models (VLMs) have expanded their potential for real-world applications, enabling these models to perform complex reasoning on images.\nIn the widely used fully autoregressive transformer-based models like LLaVA, projected visual tokens are ... | [
{
"id": "ag6zNxpDHk",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces Victor (Visual Compact Token Registers), a method for reducing the number of visual tokens in vision-language models to improve efficiency w... | {
"rating": "3;3;5;6;8",
"rating_avg": 5,
"confidence": "4;4;4;3;4",
"confidence_avg": 3.8,
"soundness": "3;2;2;3;4",
"soundness_avg": 2.8,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "2;2;3;3;4",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.886482"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
tPNHOoZFl9 | 2407.10490v2 | Learning Dynamics of LLM Finetuning | {
"content": "## Abstract\n\nAbstract Learning dynamics, which describes how the learning of specific training examples influences the model’s prediction of other examples,\ngive us a powerful tool for understanding the behavior of deep learning systems.\nWe study the learning dynamics of large language models during... | [
{
"id": "QiHWk89hT3",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "This work studies the learning dynamics of LLM finetuning by analyzing the way in which likelihoods assigned by the model to different completions of training pro... | {
"rating": "6;8;8;8",
"rating_avg": 7.5,
"confidence": "3;3;3;4",
"confidence_avg": 3.25,
"soundness": "3;3;4;3",
"soundness_avg": 3.25,
"contribution": "2;3;4;4",
"contribution_avg": 3.25,
"presentation": "3;4;4;4",
"presentation_avg": 3.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.887562"
} | {
"id": "Nke9RODsuH",
"metareview": "### Summary of the paper\nThe paper analyzes the learning dynamics of LLM finetuning by examining how model-assigned likelihoods to different prompt responses evolve during training. Key findings and contributions include:\n- Development of an analytical framework to describe sa... | {
"decision": "Accept (Oral)"
} |
tQ1PmLfPBL | 2408.07547v1 | PeriodWave: Multi-Period Flow Matching for High-Fidelity Waveform Generation | {
"content": "## Abstract\n\nAbstract Recently, universal waveform generation tasks have been investigated conditioned on various out-of-distribution scenarios. Although GAN-based methods have shown their strength in fast waveform generation, they are vulnerable to train-inference mismatch scenarios such as two-stage... | [
{
"id": "kSWbF0vDf6",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper studies the task of speech and audio waveform generation using the flow-matching modeling paradigm. The authors proposed a multi-period approach to gen... | {
"rating": "3;5;6;6;6",
"rating_avg": 5.2,
"confidence": "4;3;3;2;4",
"confidence_avg": 3.2,
"soundness": "3;3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;3;3",
"contribution_avg": 2.8,
"presentation": "3;2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.888878"
} | {
"id": "39grtCd5sO",
"metareview": "> This paper studies the task of speech and audio waveform generation using the flow-matching modeling paradigm. The authors proposed a multi-period approach to generation artifacts in high-frequencies, conditioned on either mel-spectrogram and discrete representation extracted ... | {
"decision": "Accept (Poster)"
} |
tQyh0gnfqW | 2410.01500v1 | Discrete Diffusion Schrödinger Bridge Matching for Graph Transformation | {
"content": "## Abstract\n\nAbstract Transporting between arbitrary distributions is a fundamental goal in generative modeling.\nRecently proposed diffusion bridge models provide a potential solution, but they rely on a joint distribution that is difficult to obtain in practice.\nFurthermore, formulations based on c... | [
{
"id": "s2U6098Ipr",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work presents Discrete Diffusion Schrödinger Bridge Matching (DDSBM), a framework that adapts continuous-time Markov chains (CTMCs) to address the Schrödinge... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "3;4;2",
"confidence_avg": 3,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "2;2;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.889892"
} | {
"id": "vHDBXHLoUv",
"metareview": "The authors introduce Discrete Diffusion Schrödinger Bridge Matching (DDSBM), a new method for transforming data in discrete spaces like graphs using continuous-time Markov chains. DDSBM improves key properties of molecules with minimal changes to their structure, preserving oth... | {
"decision": "Accept (Poster)"
} |
tR2qSmSOQ3 | 2310.19805v4 | Improving Offline-to-Online Reinforcement Learning with Q Conditioned State Entropy Exploration | {
"content": "## Abstract\n\nAbstract Studying how to fine-tune offline reinforcement learning (RL) pre-trained policy is profoundly significant for enhancing the sample efficiency of RL algorithms. However, directly fine-tuning pre-trained policies often results in sub-optimal performance. This is primarily due to t... | [
{
"id": "a7ZCTkGj5t",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes a new reinforcement learning method called Q-conditioned State Entropy Maximization (QCSE) which aims to improve the performance of offline-to... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;1;3;3",
"soundness_avg": 2.25,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "3;2;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.890769"
} | {
"id": "mMRypzdOwb",
"metareview": "The paper proposes a novel intrinsic motivation quantity for exploration in offline-to-online RL. While this is an important topic, and the paper is mostly well written and motivated, some concerns were raised regarding the mathematical clarity, and indeed correctness, of the pr... | {
"decision": "Reject"
} |
tRNKe2Vgqt | 2406.08407v3 | MMWorld: Towards Multi-discipline Multi-faceted World Model Evaluation in Videos | {
"content": "## Abstract\n\nAbstract Multimodal Language Language Models (MLLMs) demonstrate the emerging abilities of \"world models\"—interpreting and reasoning about complex real-world dynamics.\nTo assess these abilities, we posit videos are the ideal medium, as they encapsulate rich representations of real-worl... | [
{
"id": "72RBbUFnQ0",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces MMWorld, a new benchmark for evaluating multimodal large language models' (MLLMs) ability to understand and reason about video content acros... | {
"rating": "5;5;5;6;6",
"rating_avg": 5.4,
"confidence": "4;4;4;4;4",
"confidence_avg": 4,
"soundness": "3;3;2;3;3",
"soundness_avg": 2.8,
"contribution": "2;3;2;3;3",
"contribution_avg": 2.6,
"presentation": "3;2;2;2;4",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.891635"
} | {
"id": "kBQTnN7Mt2",
"metareview": "This paper proposes a world model evaluation in videos. All the reviews are positive in the final review. The raised concerns have been addressed in the authors' responses and acknowledged by the reviewers. Overall, the AC has checked all the files and stands on the reviewers' s... | {
"decision": "Accept (Poster)"
} |
tSmkYZ8vU7 | 2410.06593v1 | Towards Natural Image Matting in the Wild via Real-Scenario Prior | {
"content": "## Abstract\n\nAbstract Recent approaches attempt to adapt powerful interactive segmentation models, such as SAM, to interactive matting and fine-tune the models based on synthetic matting datasets. However, models trained on synthetic data fail to generalize to complex and occlusion scenes. We address ... | [
{
"id": "m7Nq3Dfzu6",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper attempts to address the poor generalization of SAM-based interactive image matting in reality. To this end, a COCO-Matting dataset is collected and ann... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "5;4;5;4",
"confidence_avg": 4.5,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;2",
"contribution_avg": 2.25,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.892351"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
tTDUrseRRU | 2410.03051v1 | AuroraCap: Efficient, Performant Video Detailed Captioning and a New Benchmark | {
"content": "## Abstract\n\nAbstract Video detailed captioning is a key task which aims to generate comprehensive and coherent textual descriptions of video content, benefiting both video understanding and generation. In this paper, we propose AuroraCap , a video captioner based on a large multimodal model. We follo... | [
{
"id": "0ukflQZIFC",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 4,
"presentation": 3,
"summary": "Towards the video detailed captioning task, this paper conducts comprehensive research including the introduction of a new model called AuroraCap, a novel dataset... | {
"rating": "5;6;6;6;8",
"rating_avg": 6.2,
"confidence": "3;4;4;4;4",
"confidence_avg": 3.8,
"soundness": "3;3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;3;4",
"contribution_avg": 3.2,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.893190"
} | {
"id": "44ppRisSWw",
"metareview": "This paper introduces a new benchmark and evaluation metric for video detailed captioning, alongside a strong baseline model AuroraCap with token merging strategy. Reviewers acknowledged the clear presentation and contributions of the proposed benchmark, metric, and extensive ex... | {
"decision": "Accept (Poster)"
} |
tTPHgb0EtV | 2409.01586v3 | Booster: Tackling Harmful Fine-tuning for Large Language Models via Attenuating Harmful Perturbation | {
"content": "## Abstract\n\nAbstract Harmful fine-tuning issue (Qi et al., 2023 ) poses serious safety concerns for Large language models’ fine-tuning-as-a-service. While existing defenses have been proposed to mitigate the issue, their performances are still far away from satisfactory, and the root cause of the pro... | [
{
"id": "etRsqFX3GK",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "In this paper, the authors propose a method to alleviate the influence of attacking fine-tuning for breaking the LLM alignment. Specifically, the authors add a re... | {
"rating": "3;8;8;8",
"rating_avg": 6.75,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "1;3;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.894129"
} | {
"id": "hN5ON7TBQE",
"metareview": "This paper proposes an alignment-stage method, Booster, to defend against harmful fine-tuning attack by adding a loss regularizer in the alignment stage's optimization. Empirical results show that Booster can effectively reduce the harmful score of the fine-tuned models while m... | {
"decision": "Accept (Oral)"
} |
tVRVE0OAyb | 2402.01943v2 | Precedence-Constrained Winter Value for Effective Graph Data Valuation | {
"content": "## Abstract\n\nAbstract Data valuation is essential for quantifying data’s worth, aiding in assessing data quality and determining fair compensation.\nWhile existing data valuation methods have proven effective in evaluating the value of Euclidean data, they face limitations when applied to the increasi... | [
{
"id": "jFLzPLh4in",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper tackles the problem of modeling and computing the value of data in learning, when graphs are the data modality. The work is the first to consider this ... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "5;5;3;3",
"confidence_avg": 4,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.894939"
} | {
"id": "gUyphh4jNL",
"metareview": "The paper presents an algorithm to assess the utility of a data entity when the data is modeled as a graph. The paper introduces the idea of Precedence-Constrained Winter (PC-Winter) Value, which transfers graph data valuation into a cooperative game theory exercise. The reviewe... | {
"decision": "Accept (Poster)"
} |
tVfvKrboVY | 2409.06679v1 | E2LLM: Encoder Elongated Large Language Models for Long-Context Understanding and Reasoning | {
"content": "## Abstract\n\nAbstract In the realm of Large Language Models (LLMs), the ability to process long contexts is increasingly crucial for tasks such as multi-round dialogues, code generation, and document summarization. This paper addresses the challenges of enhancing the long-context performance, reducing... | [
{
"id": "IvvkTHofZb",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 1,
"summary": "The paper presents E2LLM, an architecture that addresses long-context processing in LLMs through a chunk-based approach. \n\nAt its core, E2LLM employs a pretrain... | {
"rating": "5;5;5;5;5",
"rating_avg": 5,
"confidence": "3;3;4;4;3",
"confidence_avg": 3.4,
"soundness": "3;2;3;2;3",
"soundness_avg": 2.6,
"contribution": "2;3;2;3;3",
"contribution_avg": 2.6,
"presentation": "3;3;4;2;1",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.895679"
} | {
"id": "ADA1zbVpjA",
"metareview": "This paper proposes a sub-quadratic transformer intended for generating text conditional on some long input context. The architecture, resemblant of the legacy encoder-decoder models, first compresses the context by chunking and applying a local encoder and temporal pooling. The... | {
"decision": "Reject"
} |
tZCqSVncRf | 2410.09542v1 | MIRAGE: Evaluating and Explaining Inductive Reasoning Process in Language Models | {
"content": "## Abstract\n\nAbstract Inductive reasoning is an essential capability for large language models (LLMs) to achieve higher intelligence, which requires the model to generalize rules from observed facts and then apply them to unseen examples. We present Mirage , a synthetic dataset that addresses the limi... | [
{
"id": "x69zRbdQfT",
"initial_rating": 8,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper evaluates and analyzes LLM's ability for inductive reasoning. Compared to previous works in inductive reasoning, it adopts a new setting, which is to c... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "2;2;4;3",
"confidence_avg": 2.75,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;2;4;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.896439"
} | {
"id": "Urz8XSECAe",
"metareview": "The paper claims that LLMs perform better when asked to generalize from facts to neighbor facts (dubbed neighbor-based reasoning) than when asked to learn rules from the available facts, and thereafter use these rules (rule-based reasoning). \nA dataset MIRAGE is built to invest... | {
"decision": "Accept (Poster)"
} |
tZDhrhUOcs | 2410.00168v1 | SSR: Alignment-Aware Modality Connector for Speech Language Models | {
"content": "## Abstract\n\nAbstract Fusing speech into pre-trained language model (SpeechLM) usually suffers from inefficient encoding of long-form speech and catastrophic forgetting of pre-trained text modality. We propose SSR-Connector (Segmented Speech Representation Connector) for better modality fusion. Levera... | [
{
"id": "Kw0K8UTmK1",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper proposes a new architecture with the corresponding loss design to connect speech embeddings to LLM. Thorough experiments are conducted to demonstrate t... | {
"rating": "3;3;6;6",
"rating_avg": 4.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;2;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.897260"
} | {
"id": "P3g7AwoL24",
"metareview": "The paper introduces SSR-Connector, an approach designed to integrate speech data into pre-trained LLMs. It achieves this by aligning and compressing speech representations to match text embeddings, enhancing the fusion of speech and text modalities. SSR-Connector employs a two-... | {
"decision": "Reject"
} |
tZiMLgsHMu | 2411.00142v1 | JudgeRank: Leveraging Large Language Models for Reasoning-Intensive Reranking | {
"content": "## Abstract\n\nAbstract Accurate document retrieval is crucial for the success of retrieval-augmented generation (RAG) applications, including open-domain question answering and code completion.\nWhile large language models (LLMs) have been employed as dense encoders or listwise rerankers in RAG systems... | [
{
"id": "iPGJEXVAJK",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 4,
"summary": "This paper introduces JudgeRank, a new method for LLM-based pointwise passage reranking. The method diverges from existing methods in that they apply LLMs to do p... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;2",
"contribution_avg": 2.25,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.898014"
} | {
"id": "pyR7XLqkEh",
"metareview": "This paper introduces JudgeRank, a novel LLM-based method for pointwise passage reranking. It employs a three-step process: query decomposition to identify core issues, query-focused document summarization, and combining these outputs for relevance scoring. Tested on BRIGHT and ... | {
"decision": "Reject"
} |
tZozeR3VV7 | 2410.01264v1 | Backdooring Vision-Language Models with Out-Of-Distribution Data | {
"content": "## Abstract\n\nAbstract The emergence of Vision-Language Models (VLMs) represents a significant advancement in integrating computer vision with Large Language Models (LLMs) to generate detailed text descriptions from visual inputs. Despite their growing importance, the security of VLMs, particularly aga... | [
{
"id": "rt1jzT0k2Z",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper studies backdoor attacks on vision-language models (VLLMs) under an out-of-distribution setting, where the attacker has access only to a new, out-of-di... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "4;3;3",
"confidence_avg": 3.3333333333333335,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.898614"
} | {
"id": "Bzy6llI31i",
"metareview": "This paper proposes that backdoor poisoning attacks on VLMs should be performed using out-of-distribution (OOD) samples, as the attacker may not have access to the training data distribution. In this context, the attacker still has access to the trained VLM and can fine-tune the... | {
"decision": "Accept (Poster)"
} |
taHwqSrbrb | 2410.03456v2 | Dynamic Diffusion Transformer | {
"content": "## Abstract\n\nAbstract Diffusion Transformer (DiT), an emerging diffusion model for image generation, has demonstrated superior performance but suffers from substantial computational costs. Our investigations reveal that these costs stem from the static inference paradigm, which inevitably introduces r... | [
{
"id": "rzTv5OV4vj",
"initial_rating": 3,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces the Dynamic Diffusion Transformer (DyDiT), an architecture that dynamically allocates FLOPs to the most demanding areas based on varying tim... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.899299"
} | {
"id": "3i6anZcYra",
"metareview": "The paper introduces DyDiT, a method aimed at enhancing computational efficiency in diffusion models by dynamically adjusting resources during image generation. The proposed approach leverages mechanisms like TDW and SDT to reduce redundant computations while maintaining competi... | {
"decision": "Accept (Poster)"
} |
tbx3u2oZAu | 2406.00944v2 | A Theory for Token-Level Harmonization in Retrieval-Augmented Generation | {
"content": "## Abstract\n\nAbstract Retrieval-augmented generation (RAG) utilizes retrieved texts to enhance large language models (LLMs). Studies show that while RAG provides valuable external information (benefit), it may also mislead LLMs (detriment) with noisy or incorrect retrieved texts. Although many existin... | [
{
"id": "gmc4gDwZiB",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "In RAG settings, retrieved passages may contradict the parametric memory of an LLM (obtained during the LLM training process). The paper explores how to reason an... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "4;3;4",
"confidence_avg": 3.6666666666666665,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "2;2;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.900161"
} | {
"id": "OWx0IKUoC0",
"metareview": "This paper provides a theory to describe the trade-off between the value of external knowledge and its potential risk of misleading LLMs in next token prediction of RAG, and proposes a practical method to make pure LLM and RAG collaborate at token level. This paper connects theo... | {
"decision": "Accept (Poster)"
} |
tc90LV0yRL | 2408.08926v2 | Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models | {
"content": "## Abstract\n\nAbstract Language Model (LM) agents for cybersecurity that are capable of autonomously identifying vulnerabilities and executing exploits have the potential to cause real-world impact. Policymakers, model providers, and other researchers in the AI and cybersecurity communities are interes... | [
{
"id": "5rigdRhVDk",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The authors introduce a framework (CyBench) which allows to\n\n- specify cybersec tasks (focused on CTF, hence exploitation of vulnerabilities / cyberoffense) and... | {
"rating": "8;8;8",
"rating_avg": 8,
"confidence": "4;3;4",
"confidence_avg": 3.6666666666666665,
"soundness": "4;3;3",
"soundness_avg": 3.3333333333333335,
"contribution": "4;4;3",
"contribution_avg": 3.6666666666666665,
"presentation": "3;4;3",
"presentation_avg": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.901170"
} | {
"id": "BFYxk5wQ41",
"metareview": "There is consensus among the reviewers that this is a strong benchmark paper that is likely to positively impact the community working on understanding the capabilities of AI agents. The reviewers agree that this paper is worthy of recognition at ICLR, and I concur.\n\nThe paper... | {
"decision": "Accept (Oral)"
} |
tccML2tDd4 | 2410.01225v2 | Perceptual Piercing: Human Visual Cue-Based Object Detection in Low Visibility Conditions | {
"content": "## Abstract\n\nAbstract This study proposes a novel deep learning framework inspired by atmospheric scattering and human visual cortex mechanisms to enhance object detection under poor visibility scenarios such as fog, smoke, and haze. These conditions pose significant challenges for object recognition,... | [
{
"id": "qHFU7F6wjH",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 1,
"presentation": 2,
"summary": "In this paper, the authors propose a deep learning framework for foggy scenes object detection inspired by atmospheric scattering and human visual cortex mechanis... | {
"rating": "1;3;3",
"rating_avg": 2.3333333333333335,
"confidence": "5;4;3",
"confidence_avg": 4,
"soundness": "1;1;2",
"soundness_avg": 1.3333333333333333,
"contribution": "1;1;1",
"contribution_avg": 1,
"presentation": "1;3;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.901874"
} | {
"id": "Jnh6WxGcMn",
"metareview": "This study proposes a deep learning framework inspired by atmospheric scattering and human visual cortex mechanisms to enhance object detection under poor visibility scenarios. A lightweight object-detection model is applied first, followed by a spatial attention-based dehazing ... | {
"decision": "Reject"
} |
tcdbBbHHPo | 2405.21040v1 | Direct Alignment of Language Models via Quality-Aware Self-Refinement | {
"content": "## Abstract\n\nAbstract Reinforcement Learning from Human Feedback (RLHF) has been commonly used to align the behaviors of Large Language Models (LLMs) with human preferences. Recently, a popular alternative is Direct Policy Optimization (DPO), which replaces an LLM-based reward model with the policy it... | [
{
"id": "ybuyuudgIQ",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "Recently, offline preferences-based RL methods have gained popularity for doing RLHF alignment without the need for online exploration or training a reward model.... | {
"rating": "3;5;5",
"rating_avg": 4.333333333333333,
"confidence": "3;4;4",
"confidence_avg": 3.6666666666666665,
"soundness": "2;2;2",
"soundness_avg": 2,
"contribution": "2;2;2",
"contribution_avg": 2,
"presentation": "2;3;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.902509"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
tciQfO8S8j | 2410.15287v1 | Training Language Models to Critique with Multi-Agent Feedback | {
"content": "## Abstract\n\nAbstract Critique ability, a meta-cognitive capability of humans, presents significant challenges for LLMs to improve.\nRecent works primarily rely on supervised fine-tuning (SFT) using critiques generated by a single LLM like GPT-4.\nHowever, these model-generated critiques often exhibit... | [
{
"id": "ljJxU5iUtI",
"initial_rating": 8,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes a critique data generation pipeline called MultiCritique to enhance the critique capabilities of language models. The process is divided into ... | {
"rating": "5;5;6;6;6;8",
"rating_avg": 6,
"confidence": "3;4;3;3;3;2",
"confidence_avg": 3,
"soundness": "2;2;3;3;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;2;3;3;2;2",
"contribution_avg": 2.3333333333333335,
"presentation": "3;2;3;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:03.903208"
} | {
"id": "puWmB8RMLq",
"metareview": "After carefully considering the six expert reviews and the subsequent author-reviewer discussions, I recommend rejecting this submission. While the paper presents an interesting approach to improving language models' critique abilities through multi-agent feedback, several conce... | {
"decision": "Reject"
} |
tcsZt9ZNKD | 2406.04093v1 | Scaling and evaluating sparse autoencoders | {
"content": "## Abstract\n\nAbstract Sparse autoencoders provide a promising unsupervised approach for extracting interpretable features from a language model by reconstructing activations from a sparse bottleneck layer. Since language models learn many concepts, autoencoders need to be very large to recover all rel... | [
{
"id": "YXtmcN5e0x",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes the use of a k-sparse auto encoder model to control sparsity, simplifying tuning and improving the reconstruction-sparsity frontier. The paper... | {
"rating": "3;8;8;10;10",
"rating_avg": 7.8,
"confidence": "3;4;4;4;4",
"confidence_avg": 3.8,
"soundness": "3;3;4;4;4",
"soundness_avg": 3.6,
"contribution": "3;4;4;4;4",
"contribution_avg": 3.8,
"presentation": "3;4;3;3;4",
"presentation_avg": 3.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:03.904204"
} | {
"id": "mTYg3CGVIK",
"metareview": "The paper proposes scaled sparse autoencoders with top-k activation to directly control the number of active latent variables and thereby improve interpretability. The proposed approach bares significant benefits in terms of reconstruction ability, non-dead latents and sparsity,... | {
"decision": "Accept (Oral)"
} |
td5nvlhJdc | 2408.10455v4 | IDEA: Enhancing the Rule Learning Ability of Large Language Model Agent through Induction, Deduction, and Abduction | {
"content": "## Abstract\n\nAbstract While large language models (LLMs) have been thoroughly evaluated for deductive and inductive reasoning, their proficiency in abductive reasoning and holistic rule learning in interactive environments remains less explored. We introduce RULEARN , a novel benchmark specifically de... | [
{
"id": "z8OxKNI75Y",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This works proposes RULEARN, a benchmark consists of interactive puzzles that are designed to assess the rulelearning abilities of LLM agents in interactive setti... | {
"rating": "3;3;5;6;6",
"rating_avg": 4.6,
"confidence": "4;4;5;4;4",
"confidence_avg": 4.2,
"soundness": "2;2;3;3;3",
"soundness_avg": 2.6,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:03.905220"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.