id
string
sources
list
title
string
abstract
string
authors
list
categories
list
fields_of_study
list
published_date
timestamp[s]
url
string
pdf_url
string
arxiv_id
string
doi
string
citation_count
int64
influential_citation_count
int64
has_code
bool
code_url
string
venue
string
quality_score
float64
d5a610721942adb0a0524b47746b5e6b4a8104c5254edc54f5b43a8ab1959216
[ "arxiv", "semantic_scholar" ]
AW-MoE: All-Weather Mixture of Experts for Robust Multi-Modal 3D Object Detection
Robust 3D object detection under adverse weather conditions is crucial for autonomous driving. However, most existing methods simply combine all weather samples for training while overlooking data distribution discrepancies across different weather scenarios, leading to performance conflicts. To address this issue, we ...
[ "Hongwei Lin", "Xun Huang", "Chenglu Wen", "Cheng Wang" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2026-03-17T00:00:00
https://arxiv.org/abs/2603.16261
https://arxiv.org/pdf/2603.16261v1
2603.16261
10.48550/arXiv.2603.16261
1
0
true
https://github.com/windlinsherlock/AW-MoE
arXiv.org
0.7898
c3ec1871e482b75e9e21ecdacbcd120285394b8336817fa5153dfc3534331ea8
[ "arxiv", "semantic_scholar" ]
Knowledge Localization in Mixture-of-Experts LLMs Using Cross-Lingual Inconsistency
Modern LLMs continue to exhibit significant variance in behavior across languages, such as being able to recall factual information in some languages but not others. While typically studied as a problem to be mitigated, in this work, we propose leveraging this cross-lingual inconsistency as a tool for interpretability ...
[ "Lucas Bandarkar", "Alan Ansell", "Trevor Cohn" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2026-03-17T00:00:00
https://arxiv.org/abs/2603.17102
https://arxiv.org/pdf/2603.17102v1
2603.17102
10.48550/arXiv.2603.17102
3
1
false
null
arXiv.org
0.511
a46dfd39c829b4c978a1de6e3466e79636bfc759d4dddaeedc331a25666ba955
[ "arxiv", "semantic_scholar" ]
MoE-ACT: Scaling Multi-Task Bimanual Manipulation with Sparse Language-Conditioned Mixture-of-Experts Transformers
The ability of robots to handle multiple tasks under a unified policy is critical for deploying embodied intelligence in real-world household and industrial applications. However, out-of-distribution variation across tasks often causes severe task interference and negative transfer when training general robotic policie...
[ "Kangjun Guo", "Haichao Liu", "Yanji Sun", "Ruhan Zhao", "Jinni Zhou", "Jun Ma" ]
[ "cs.RO" ]
[ "Computer Science" ]
2026-03-16T00:00:00
https://arxiv.org/abs/2603.15265
https://arxiv.org/pdf/2603.15265v1
2603.15265
10.48550/arXiv.2603.15265
1
0
true
null
arXiv.org
0.788
943a64d8001b65702238b54306a6c7b8e336656930e75411ac05ae1dc913342c
[ "arxiv", "semantic_scholar" ]
Bridging Local and Global Knowledge: Cascaded Mixture-of-Experts Learning for Near-Shortest Path Routing
While deep learning models that leverage local features have demonstrated significant potential for near-optimal routing in dense Euclidean graphs, they struggle to generalize well in sparse networks where topological irregularities require broader structural awareness. To address this limitation, we train a Cascaded M...
[ "Yung-Fu Chen", "Anish Arora" ]
[ "cs.LG", "cs.NI" ]
[ "Computer Science" ]
2026-03-16T00:00:00
https://arxiv.org/abs/2603.15541
https://arxiv.org/pdf/2603.15541v1
2603.15541
10.48550/arXiv.2603.15541
0
0
false
null
arXiv.org
0.5099
bf0f16d8e660f4559def9b285d5c616764e7222fa47ee96e3ccdf52716efda04
[ "arxiv", "semantic_scholar" ]
Task-Conditioned Routing Signatures in Sparse Mixture-of-Experts Transformers
Sparse Mixture-of-Experts (MoE) architectures enable efficient scaling of large language models through conditional computation, yet the routing mechanisms responsible for expert selection remain poorly understood. In this work, we introduce routing signatures, a vector representation summarizing expert activation patt...
[ "Mynampati Sri Ranganadha Avinash" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-11T00:00:00
https://arxiv.org/abs/2603.11114
https://arxiv.org/pdf/2603.11114v1
2603.11114
10.48550/arXiv.2603.11114
0
0
false
null
arXiv.org
0.5042
23a28d13f48b04f0a383a3b8fab9edbfad6cd4900012e024d222abffee93f757
[ "arxiv", "semantic_scholar" ]
Variational Routing: A Scalable Bayesian Framework for Calibrated Mixture-of-Experts Transformers
Foundation models are increasingly being deployed in contexts where understanding the uncertainty of their outputs is critical to ensuring responsible deployment. While Bayesian methods offer a principled approach to uncertainty quantification, their computational overhead renders their use impractical for training or ...
[ "Albus Yizhuo Li", "Matthew Wicker" ]
[ "cs.LG", "cs.AI", "stat.ML" ]
[ "Computer Science", "Mathematics" ]
2026-03-10T00:00:00
https://arxiv.org/abs/2603.09453
https://arxiv.org/pdf/2603.09453v3
2603.09453
10.48550/arXiv.2603.09453
2
1
false
null
arXiv.org
0.503
e1f0fe191047317dca18254b9917abfe82eaff9ca8123b38100dbfa97f705e2d
[ "arxiv", "semantic_scholar" ]
LAR-MoE: Latent-Aligned Routing for Mixture of Experts in Robotic Imitation Learning
Imitation learning enables robots to acquire manipulation skills from demonstrations, yet deploying a policy across tasks with heterogeneous dynamics remains challenging, as models tend to average over distinct behavioral modes present in the demonstrations. Mixture-of-Experts (MoE) architectures address this by activa...
[ "Ariel Rodriguez", "Chenpan Li", "Lorenzo Mazza", "Rayan Younis", "Ortrun Hellig", "Sebastian Bodenstedt", "Martin Wagner", "Stefanie Speidel" ]
[ "cs.RO" ]
[ "Computer Science" ]
2026-03-09T00:00:00
https://arxiv.org/abs/2603.08476
https://arxiv.org/pdf/2603.08476v1
2603.08476
10.48550/arXiv.2603.08476
2
0
false
null
arXiv.org
0.5019
0100340ad49fd23fe58b9e0a96b3229672973e90dda2f753b21b72f518da7a91
[ "arxiv", "semantic_scholar" ]
Scaling Machine Learning Interatomic Potentials with Mixtures of Experts
Machine Learning Interatomic Potentials (MLIPs) enable accurate large-scale atomistic simulations, yet improving their expressive capacity efficiently remains challenging. Here we systematically develop Mixture-of-Experts (MoE) and Mixture-of-Linear-Experts (MoLE) architectures for MLIPs and analyze the effects of rout...
[ "Yuzhi Liu", "Duo Zhang", "Anyang Peng", "Weinan E", "Linfeng Zhang", "Han Wang" ]
[ "physics.chem-ph", "cs.LG", "physics.comp-ph" ]
[ "Computer Science", "Physics" ]
2026-03-09T00:00:00
https://arxiv.org/abs/2603.07977
https://arxiv.org/pdf/2603.07977v2
2603.07977
10.48550/arXiv.2603.07977
2
0
false
null
arXiv.org
0.5019
abaf72d9dcb8237f170e98af3b8e9308c2e965d7c8234d464348ebbfec8e115b
[ "arxiv", "semantic_scholar" ]
Speculating Experts Accelerates Inference for Mixture-of-Experts
Mixture-of-Experts (MoE) models have gained popularity as a means of scaling the capacity of large language models (LLMs) while maintaining sparse activations and reduced per-token compute. However, in memory-constrained inference settings, expert weights must be offloaded to CPU, creating a performance bottleneck from...
[ "Vivan Madan", "Prajwal Singhania", "Abhinav Bhatele", "Tom Goldstein", "Ashwinee Panda" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-09T00:00:00
https://arxiv.org/abs/2603.19289
https://arxiv.org/pdf/2603.19289v1
2603.19289
10.48550/arXiv.2603.19289
0
0
true
https://github.com/axonn-ai/yalis/tree/offload_prefetch
arXiv.org
0.7756
4b6300ee0dc9a714d53be5a8f22959b02d75ae8892ff1d79c5d22ccf884d7aa8
[ "arxiv", "semantic_scholar" ]
MoE Lens -- An Expert Is All You Need
Mixture of Experts (MoE) models enable parameter-efficient scaling through sparse expert activations, yet optimizing their inference and memory costs remains challenging due to limited understanding of their specialization behavior. We present a systematic analysis of expert specialization in MoEs through two complemen...
[ "Marmik Chaudhari", "Idhant Gulati", "Nishkal Hundia", "Pranav Karra", "Shivam Raval" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-03-06T00:00:00
https://arxiv.org/abs/2603.05806
https://arxiv.org/pdf/2603.05806v1
2603.05806
10.48550/arXiv.2603.05806
11
2
false
null
arXiv.org
0.4984
4b15603ac4292096a7583ef0684facc063bbba16a4b514e3cb3761923f6341d0
[ "arxiv", "semantic_scholar" ]
Sparse Crosscoders for diffing MoEs and Dense models
Mixture of Experts (MoE) achieve parameter-efficient scaling through sparse expert routing, yet their internal representations remain poorly understood compared to dense models. We present a systematic comparison of MoE and dense model internals using crosscoders, a variant of sparse autoencoders, that jointly models m...
[ "Marmik Chaudhari", "Nishkal Hundia", "Idhant Gulati" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-03-06T00:00:00
https://arxiv.org/abs/2603.05805
https://arxiv.org/pdf/2603.05805v1
2603.05805
10.48550/arXiv.2603.05805
0
0
false
null
arXiv.org
0.4984
14f31e1bf75fbd2fe91977feed130dc9682a85a38cb99a957c43c695db983471
[ "arxiv", "semantic_scholar" ]
MoEless: Efficient MoE LLM Serving via Serverless Computing
Large Language Models (LLMs) have become a cornerstone of AI, driving progress across diverse domains such as content creation, search and recommendation systems, and AI-assisted workflows. To alleviate extreme training costs and advancing model scales, Mixture-of-Experts (MoE) has become a popular backbone for modern ...
[ "Hanfei Yu", "Bei Ouyang", "Shwai He", "Ang Li", "Hao Wang" ]
[ "cs.DC", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2026-03-06T00:00:00
https://arxiv.org/abs/2603.06350
https://arxiv.org/pdf/2603.06350v1
2603.06350
10.48550/arXiv.2603.06350
0
0
true
null
arXiv.org
0.7703
5ec00039916aa4614b68a0a56d271285b5ab6d5c39cde89c38bb74fa8ca1b317
[ "arxiv", "semantic_scholar" ]
Mixture of Universal Experts: Scaling Virtual Width via Depth-Width Transformation
Mixture-of-Experts (MoE) decouples model capacity from per-token computation, yet their scalability remains limited by the physical dimensions of depth and width. To overcome this, we propose Mixture of Universal Experts (MOUE),a MoE generalization introducing a novel scaling dimension: Virtual Width. In general, MoUE ...
[ "Yilong Chen", "Naibin Gu", "Junyuan Shang", "Zhenyu Zhang", "Yuchen Feng", "Jiawei Sheng", "Tingwen Liu", "Shuohuan Wang", "Yu Sun", "Hua Wu", "Haifeng Wang" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-03-05T00:00:00
https://arxiv.org/abs/2603.04971
https://arxiv.org/pdf/2603.04971v1
2603.04971
10.48550/arXiv.2603.04971
3
1
false
null
arXiv.org
0.4973
54fb8a3bff1281d74187205935a92da17aaa4c4cbb204292856aa1c682d3b505
[ "arxiv", "semantic_scholar" ]
ECG-MoE: Mixture-of-Expert Electrocardiogram Foundation Model
Electrocardiography (ECG) analysis is crucial for cardiac diagnosis, yet existing foundation models often fail to capture the periodicity and diverse features required for varied clinical tasks. We propose ECG-MoE, a hybrid architecture that integrates multi-model temporal features with a cardiac period-aware expert mo...
[ "Yuhao Xu", "Xiaoda Wang", "Yi Wu", "Wei Jin", "Xiao Hu", "Carl Yang" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-03-04T00:00:00
https://arxiv.org/abs/2603.04589
https://arxiv.org/pdf/2603.04589v1
2603.04589
10.48550/arXiv.2603.04589
4
0
false
null
arXiv.org
0.4961
6aa880401d0e7ff960d95a3c16529fe191cbbfb9c62d35556712869aa302f50d
[ "arxiv", "semantic_scholar" ]
CoMoL: Efficient Mixture of LoRA Experts via Dynamic Core Space Merging
Large language models (LLMs) achieve remarkable performance on diverse downstream and domain-specific tasks via parameter-efficient fine-tuning (PEFT). However, existing PEFT methods, particularly MoE-LoRA architectures, suffer from limited parameter efficiency and coarse-grained adaptation due to the proliferation of ...
[ "Jie Cao", "Zhenxuan Fan", "Zhuonan Wang", "Tianwei Lin", "Ziyuan Zhao", "Rolan Yan", "Wenqiao Zhang", "Feifei Shao", "Hongwei Wang", "Jun Xiao", "Siliang Tang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-02-28T00:00:00
https://arxiv.org/abs/2603.00573
https://arxiv.org/pdf/2603.00573v2
2603.00573
10.48550/arXiv.2603.00573
2
0
false
null
arXiv.org
0.4916
c6782d8c86d1ce181bdd4433644e93baeaea622c64803a9e6e3a89dacc07b1c8
[ "arxiv", "semantic_scholar" ]
Modality-Guided Mixture of Graph Experts with Entropy-Triggered Routing for Multimodal Recommendation
Multimodal recommendation enhances ranking by integrating user-item interactions with item content, which is particularly effective under sparse feedback and long-tail distributions. However, multimodal signals are inherently heterogeneous and can conflict in specific contexts, making effective fusion both crucial and ...
[ "Ji Dai", "Quan Fang", "Dengsheng Cai" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-02-24T00:00:00
https://arxiv.org/abs/2602.20723
https://arxiv.org/pdf/2602.20723v2
2602.20723
10.48550/arXiv.2602.20723
0
0
false
null
arXiv.org
0.487
8d4e0d766aa7178ee3264d47cca931984e5394f867867e3616b7c70b6b3895e8
[ "arxiv", "semantic_scholar" ]
A Replicate-and-Quantize Strategy for Plug-and-Play Load Balancing of Sparse Mixture-of-Experts LLMs
Sparse Mixture-of-Experts (SMoE) architectures are increasingly used to scale large language models efficiently, delivering strong accuracy under fixed compute budgets. However, SMoE models often suffer from severe load imbalance across experts, where a small subset of experts receives most tokens while others are unde...
[ "Zijie Liu", "Jie Peng", "Jinhao Duan", "Zirui Liu", "Kaixiong Zhou", "Mingfu Liang", "Luke Simon", "Xi Liu", "Zhaozhuo Xu", "Tianlong Chen" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-02-23T00:00:00
https://arxiv.org/abs/2602.19938
https://arxiv.org/pdf/2602.19938v1
2602.19938
10.48550/arXiv.2602.19938
0
0
false
null
arXiv.org
0.4858
0af64980ae04fd1447fefb5fae02a9d597350b18863cee21b23ccbf401a0e7c3
[ "arxiv", "semantic_scholar" ]
Grouter: Decoupling Routing from Representation for Accelerated MoE Training
Traditional Mixture-of-Experts (MoE) training typically proceeds without any structural priors, effectively requiring the model to simultaneously train expert weights while searching for an optimal routing policy within a vast combinatorial space. This entanglement often leads to sluggish convergence and training insta...
[ "Yuqi Xu", "Rizhen Hu", "Zihan Liu", "Mou Sun", "Kun Yuan" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-02-22T00:00:00
https://arxiv.org/abs/2603.06626
https://arxiv.org/pdf/2603.06626v2
2603.06626
10.48550/arXiv.2603.06626
6
0
true
https://github.com/JimmyAwoe/Grouter
arXiv.org
0.7491
52b1b3ad5eb18edf5f47250fbebd6a09e8a5ecd0bbf3edf9d5616a3544fa68e2
[ "arxiv", "semantic_scholar" ]
Routing-Aware Explanations for Mixture of Experts Graph Models in Malware Detection
Mixture-of-Experts (MoE) offers flexible graph reasoning by combining multiple views of a graph through a learned router. We investigate routing-aware explanations for MoE graph models in malware detection using control flow graphs (CFGs). Our architecture builds diversity at two levels. At the node level, each layer c...
[ "Hossein Shokouhinejad", "Roozbeh Razavi-Far", "Griffin Higgins", "Ali. A Ghorbani" ]
[ "cs.CR", "cs.AI" ]
[ "Computer Science" ]
2026-02-22T00:00:00
https://arxiv.org/abs/2602.19025
https://arxiv.org/pdf/2602.19025v1
2602.19025
10.48550/arXiv.2602.19025
0
0
false
null
arXiv.org
0.4847
0cc3f2c90ec6f0e37139eea420a6776034cd71e82a5590278a8b89f3c812c940
[ "arxiv", "semantic_scholar" ]
Grassmannian Mixture-of-Experts: Concentration-Controlled Routing on Subspace Manifolds
Mixture-of-Experts models rely on learned routers to assign tokens to experts, yet standard softmax gating provides no principled mechanism to control the tradeoff between sparsity and utilization. We propose Grassmannian MoE (GrMoE), a routing framework that operates on the Grassmannian manifold of subspaces, where ga...
[ "Ibne Farabi Shihab", "Sanjeda Akter", "Anuj Sharma" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-02-19T00:00:00
https://arxiv.org/abs/2602.17798
https://arxiv.org/pdf/2602.17798v1
2602.17798
10.48550/arXiv.2602.17798
1
0
false
null
arXiv.org
0.4813
89825c64e037aa9c0e06a6cc615ab129a6ad01aa1d1ee180d5936b305e4572e0
[ "arxiv", "semantic_scholar" ]
Phase-Aware Mixture of Experts for Agentic Reinforcement Learning
Reinforcement learning (RL) has equipped LLM agents with a strong ability to solve complex tasks. However, existing RL methods normally use a \emph{single} policy network, causing \emph{simplicity bias} where simple tasks occupy most parameters and dominate gradient updates, leaving insufficient capacity for complex ta...
[ "Shengtian Yang", "Yu Li", "Shuo He", "Yewen Li", "Qingpeng Cai", "Peng Jiang", "Lei Feng" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-02-19T00:00:00
https://arxiv.org/abs/2602.17038
https://arxiv.org/pdf/2602.17038v3
2602.17038
10.48550/arXiv.2602.17038
1
0
false
null
arXiv.org
0.4813
f8028e71a8ea62bb19e7f8fc6c259b427149351e6c3097da0bd5378774eca8b0
[ "arxiv", "semantic_scholar" ]
MoE-Spec: Expert Budgeting for Efficient Speculative Decoding
Speculative decoding accelerates Large Language Model (LLM) inference by verifying multiple drafted tokens in parallel. However, for Mixture-of-Experts (MoE) models, this parallelism introduces a severe bottleneck: large draft trees activate many unique experts, significantly increasing memory pressure and diminishing ...
[ "Bradley McDanel", "Steven Li", "Sruthikesh Surineni", "Harshit Khaitan" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-02-17T00:00:00
https://arxiv.org/abs/2602.16052
https://arxiv.org/pdf/2602.16052v1
2602.16052
10.48550/arXiv.2602.16052
3
0
false
null
arXiv.org
0.479
ffd4c7f8a6af305c960bcd8fb055bb357281c2546865e7a3e06c5fe258f73f4e
[ "arxiv", "semantic_scholar" ]
DeepFusion: Accelerating MoE Training via Federated Knowledge Distillation from Heterogeneous Edge Devices
Recent Mixture-of-Experts (MoE)-based large language models (LLMs) such as Qwen-MoE and DeepSeek-MoE are transforming generative AI in natural language processing. However, these models require vast and diverse training data. Federated learning (FL) addresses this challenge by leveraging private data from heterogeneous...
[ "Songyuan Li", "Jia Hu", "Ahmed M. Abdelmoniem", "Geyong Min", "Haojun Huang", "Jiwei Huang" ]
[ "cs.LG", "cs.AI", "cs.MA" ]
[ "Computer Science" ]
2026-02-15T00:00:00
https://arxiv.org/abs/2602.14301
https://arxiv.org/pdf/2602.14301v1
2602.14301
10.48550/arXiv.2602.14301
0
0
false
null
arXiv.org
0.4767
a4033669b6fb0622d5a64aecb51d83cca0d243fe9d89ab1f32ccb3921bb41b36
[ "arxiv", "semantic_scholar" ]
Synergistic Intra- and Cross-Layer Regularization Losses for MoE Expert Specialization
Sparse Mixture-of-Experts (MoE) models scale Transformers efficiently but suffer from expert overlap -- redundant representations across experts and routing ambiguity, resulting in severely underutilized model capacity. While architectural solutions like DeepSeekMoE promote specialization, they require substantial stru...
[ "Rizhen Hu", "Yuan Cao", "Boao Kong", "Mou Sun", "Kun Yuan" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-02-15T00:00:00
https://arxiv.org/abs/2602.14159
https://arxiv.org/pdf/2602.14159v1
2602.14159
10.48550/arXiv.2602.14159
4
0
false
null
arXiv.org
0.4767
2b826d46e8df83d3bf0af41fa3cf437711e8ec7fd2a8dc1a33ebabf4b98d0add
[ "arxiv", "semantic_scholar" ]
Decoder-only Conformer with Modality-aware Sparse Mixtures of Experts for ASR
We present a decoder-only Conformer for automatic speech recognition (ASR) that processes speech and text in a single stack without external speech encoders or pretrained large language models (LLM). The model uses a modality-aware sparse mixture of experts (MoE): disjoint expert pools for speech and text with hard rou...
[ "Jaeyoung Lee", "Masato Mimura" ]
[ "eess.AS", "cs.AI", "cs.CL", "cs.SD" ]
[ "Computer Science", "Engineering" ]
2026-02-13T00:00:00
https://arxiv.org/abs/2602.12546
https://arxiv.org/pdf/2602.12546v1
2602.12546
10.48550/arXiv.2602.12546
0
0
false
null
IEEE International Conference on Acoustics, Speech, and Signal Processing
0.4744
9154470e2461ebc8c3ece58e9f6853e46ae6f663ddce4579433295fc9cbf351c
[ "arxiv", "semantic_scholar" ]
SD-MoE: Spectral Decomposition for Effective Expert Specialization
Mixture-of-Experts (MoE) architectures scale Large Language Models via expert specialization induced by conditional computation. In practice, however, expert specialization often fails: some experts become functionally similar, while others functioning as de facto shared experts, limiting the effective capacity and mod...
[ "Ruijun Huang", "Fang Dong", "Xin Zhang", "Hengjie Cao", "Zhendong Huang", "Anrui Chen", "Jixian Zhou", "Mengyi Chen", "Yifeng Yang", "Mingzhi Dong", "Yujiang Wang", "Jinlong Hou", "Qin Lv", "Robert P. Dick", "Yuan Cheng", "Fan Yang", "Tun Lu", "Chun Zhang", "Li Shang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-02-13T00:00:00
https://arxiv.org/abs/2602.12556
https://arxiv.org/pdf/2602.12556v1
2602.12556
10.48550/arXiv.2602.12556
1
0
false
null
arXiv.org
0.4744
20aed9382dfe0e743c409c346f4acbc869d5e9215670840d7e51b1309b860ec9
[ "arxiv", "semantic_scholar" ]
LAER-MoE: Load-Adaptive Expert Re-layout for Efficient Mixture-of-Experts Training
Expert parallelism is vital for effectively training Mixture-of-Experts (MoE) models, enabling different devices to host distinct experts, with each device processing different input data. However, during expert parallel training, dynamic routing results in significant load imbalance among experts: a handful of overloa...
[ "Xinyi Liu", "Yujie Wang", "Fangcheng Fu", "Xuefeng Xiao", "Huixia Li", "Jiashi Li", "Bin Cui" ]
[ "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2026-02-12T00:00:00
https://arxiv.org/abs/2602.11686
https://arxiv.org/pdf/2602.11686v1
2602.11686
10.1145/3779212.3790180
4
1
true
https://github.com/PKU-DAIR/Hetu-Galvatron/tree/laer-moe
International Conference on Architectural Support for Programming Languages and Operating Systems
0.7314
36c398ec794c67f78fd809c50f190fdbff0ebf64e88abda8ae46e6792ab7c8dc
[ "arxiv", "semantic_scholar" ]
MoEEdit: Efficient and Routing-Stable Knowledge Editing for Mixture-of-Experts LLMs
Knowledge editing (KE) enables precise modifications to factual content in large language models (LLMs). Existing KE methods are largely designed for dense architectures, limiting their applicability to the increasingly prevalent sparse Mixture-of-Experts (MoE) models that underpin modern scalable LLMs. Although MoEs o...
[ "Yupu Gu", "Rongzhe Wei", "Andy Zhu", "Pan Li" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-02-11T00:00:00
https://arxiv.org/abs/2602.10965
https://arxiv.org/pdf/2602.10965v1
2602.10965
10.48550/arXiv.2602.10965
3
2
false
null
arXiv.org
0.4721
1fb6d6c909f8ebc78dfe7c6fe2fe1cc0ee3274c077c15d1bec8d2377266074c1
[ "arxiv", "semantic_scholar" ]
Effective MoE-based LLM Compression by Exploiting Heterogeneous Inter-Group Experts Routing Frequency and Information Density
Mixture-of-Experts (MoE) based Large Language Models (LLMs) have achieved superior performance, yet the massive memory overhead caused by storing multiple expert networks severely hinders their practical deployment. Singular Value Decomposition (SVD)-based compression has emerged as a promising post-training technique;...
[ "Zhendong Mi", "Yixiao Chen", "Pu Zhao", "Xiaodong Yu", "Hao Wang", "Yanzhi Wang", "Shaoyi Huang" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-02-10T00:00:00
https://arxiv.org/abs/2602.09316
https://arxiv.org/pdf/2602.09316v2
2602.09316
10.48550/arXiv.2602.09316
2
0
false
null
arXiv.org
0.4709
2e7b0ef1600044e1f60a406991c40b6268f88b835954e40978969c56885ad0ba
[ "arxiv", "semantic_scholar" ]
Revealing the Challenges of Attention-FFN Disaggregation for Modern MoE Models and Hardware Systems
Deploying large-scale MoE models presents challenges in memory capacity and bandwidth for expert activation. While Attention-FFN Disaggregation (AFD) has emerged as a potential architecture to decouple compute and memory resources, its performance boundaries compared to standard large-scale Expert Parallelism (EP) rema...
[ "Guowei Liu", "Hongming Li", "Yaning Guo", "Yongxi Lyu", "Mo Zhou", "Yi Liu", "Zhaogeng Li", "Yanpeng Wang" ]
[ "cs.DC" ]
[ "Computer Science" ]
2026-02-10T00:00:00
https://arxiv.org/abs/2602.09721
https://arxiv.org/pdf/2602.09721v1
2602.09721
10.48550/arXiv.2602.09721
1
0
false
null
arXiv.org
0.4709
3012d2da9a1a77cef08711b197e58f965c8fbadfa74428e61f819b454d015430
[ "arxiv", "semantic_scholar" ]
Expert Divergence Learning for MoE-based Language Models
The Mixture-of-Experts (MoE) architecture is a powerful technique for scaling language models, yet it often suffers from expert homogenization, where experts learn redundant functionalities, thereby limiting MoE's full potential. To address this, we introduce Expert Divergence Learning, a novel pre-training strategy th...
[ "Jiaang Li", "Haibin Chen", "Langming Liu", "Yujin Yuan", "Yadao Wang", "Yizhen Zhang", "Chengting Yu", "Xin Tong", "Weidong Zhang", "Shilei Liu", "Wenbo Su", "Bo Zheng" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-02-10T00:00:00
https://arxiv.org/abs/2603.00054
https://arxiv.org/pdf/2603.00054v1
2603.00054
10.48550/arXiv.2603.00054
1
0
false
null
arXiv.org
0.4709
6101522e7988761580a40fde26fdccbc22a9107b2f1f89287f20c8985258497b
[ "arxiv", "semantic_scholar" ]
Sparse Models, Sparse Safety: Unsafe Routes in Mixture-of-Experts LLMs
By introducing routers to selectively activate experts in Transformer layers, the mixture-of-experts (MoE) architecture significantly reduces computational costs in large language models (LLMs) while maintaining competitive performance, especially for models with massive parameters. However, prior work has largely focu...
[ "Yukun Jiang", "Hai Huang", "Mingjie Li", "Yage Zhang", "Michael Backes", "Yang Zhang" ]
[ "cs.LG", "cs.AI", "cs.CR" ]
[ "Computer Science" ]
2026-02-09T00:00:00
https://arxiv.org/abs/2602.08621
https://arxiv.org/pdf/2602.08621v1
2602.08621
10.48550/arXiv.2602.08621
8
1
true
https://github.com/TrustAIRLab/UnsafeMoE
arXiv.org
0.726
60caa7ff888c963b5cb9940667dd86305a49741634dea6d384d35c6274e9c897
[ "arxiv", "semantic_scholar" ]
DirMoE: Dirichlet-routed Mixture of Experts
Mixture-of-Experts (MoE) models have demonstrated exceptional performance in large-scale language models. Existing routers typically rely on non-differentiable Top-$k$+Softmax, limiting their performance and scalability. We argue that two distinct decisions, which experts to activate and how to distribute expert contri...
[ "Amirhossein Vahidi", "Hesam Asadollahzadeh", "Navid Akhavan Attar", "Marie Moullet", "Kevin Ly", "Xingyi Yang", "Mohammad Lotfollahi" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-02-09T00:00:00
https://arxiv.org/abs/2602.09001
https://arxiv.org/pdf/2602.09001v1
2602.09001
10.48550/arXiv.2602.09001
1
0
false
null
arXiv.org
0.4698
7db41f588e40cc557fcdb29074932c6f148f31369063514015e78c0a230fd06a
[ "arxiv", "semantic_scholar" ]
TEAM: Temporal-Spatial Consistency Guided Expert Activation for MoE Diffusion Language Model Acceleration
Diffusion large language models (dLLMs) have recently gained significant attention due to their inherent support for parallel decoding. Building on this paradigm, Mixture-of-Experts (MoE) dLLMs with autoregressive (AR) initialization have further demonstrated strong performance competitive with mainstream AR models. Ho...
[ "Linye Wei", "Zixiang Luo", "Pingzhi Tang", "Meng Li" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-02-09T00:00:00
https://arxiv.org/abs/2602.08404
https://arxiv.org/pdf/2602.08404v2
2602.08404
10.48550/arXiv.2602.08404
5
1
true
https://github.com/PKU-SEC-Lab/TEAM-MoE-dLLM
arXiv.org
0.726
d977098d946caca03317032d4a8c37c79394480d07f46da2e4698dcfef4d32f1
[ "arxiv", "semantic_scholar" ]
Large Language Lobotomy: Jailbreaking Mixture-of-Experts via Expert Silencing
The rapid adoption of Mixture-of-Experts (MoE) architectures marks a major shift in the deployment of Large Language Models (LLMs). MoE LLMs improve scaling efficiency by activating only a small subset of parameters per token, but their routing structure introduces new safety attack surfaces. We find that safety-critic...
[ "Jona te Lintelo", "Lichao Wu", "Stjepan Picek" ]
[ "cs.CR" ]
[ "Computer Science" ]
2026-02-09T00:00:00
https://arxiv.org/abs/2602.08741
https://arxiv.org/pdf/2602.08741v1
2602.08741
10.48550/arXiv.2602.08741
2
1
true
null
arXiv.org
0.726
76e4678e878ebc9a1f93c1baa7465caf8f45c6d9ed480d1206c326446f340cc4
[ "arxiv", "semantic_scholar" ]
The Rise of Sparse Mixture-of-Experts: A Survey from Algorithmic Foundations to Decentralized Architectures and Vertical Domain Applications
The sparse Mixture of Experts(MoE) architecture has evolved as a powerful approach for scaling deep learning models to more parameters with comparable computation cost. As an important branch of large language model(LLM), MoE model only activate a subset of experts based on a routing network. This sparse conditional co...
[ "Dong Pan", "Bingtao Li", "Yongsheng Zheng", "Jiren Ma", "Victor Fei" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-02-08T00:00:00
https://arxiv.org/abs/2602.08019
https://arxiv.org/pdf/2602.08019v1
2602.08019
10.54097/bvpfjj49
0
0
false
null
Journal of Computer Science and Artificial Intelligence 5 (2025) 25-41
0.4686
684827ebfe9b201184a816182932dbea455e7ef5b18573427a6476e863f1f97b
[ "arxiv", "semantic_scholar" ]
SERE: Similarity-based Expert Re-routing for Efficient Batch Decoding in MoE Models
Mixture-of-Experts (MoE) architectures employ sparse activation to deliver faster training and inference with higher accuracy than dense LLMs. However, in production serving, MoE models require batch inference to optimize hardware efficiency, which may cause excessive expert activation and thus slow the memory-bound de...
[ "Juntong Wu", "Jialiang Cheng", "Fuyu Lv", "Ou Dan", "Li Yuan" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-02-07T00:00:00
https://arxiv.org/abs/2602.07616
https://arxiv.org/pdf/2602.07616v1
2602.07616
10.48550/arXiv.2602.07616
1
0
true
https://github.com/JL-Cheng/SERE
arXiv.org
0.7225
e37b1309e72586ee98e4ea2aeb8f671b8267a0e7fc932c34f5ba3d4f13dde1f0
[ "arxiv", "semantic_scholar" ]
MoSE: Mixture of Slimmable Experts for Efficient and Adaptive Language Models
Mixture-of-Experts (MoE) models scale large language models efficiently by sparsely activating experts, but once an expert is selected, it is executed fully. Hence, the trade-off between accuracy and computation in an MoE model typically exhibits large discontinuities. We propose Mixture of Slimmable Experts (MoSE), an...
[ "Nurbek Tastan", "Stefanos Laskaridis", "Karthik Nandakumar", "Samuel Horvath" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2026-02-05T00:00:00
https://arxiv.org/abs/2602.06154
https://arxiv.org/pdf/2602.06154v2
2602.06154
10.48550/arXiv.2602.06154
1
0
true
https://github.com/tnurbek/mose
arXiv.org
0.719
0787bcd3bf50d6f582a4f2c6ab48f81099f6ae88e89ea2d151f6d48efcf24f49
[ "arxiv", "semantic_scholar" ]
OmniMoE: An Efficient MoE by Orchestrating Atomic Experts at Scale
Mixture-of-Experts (MoE) architectures are evolving towards finer granularity to improve parameter efficiency. However, existing MoE designs face an inherent trade-off between the granularity of expert specialization and hardware execution efficiency. We propose OmniMoE, a system-algorithm co-designed framework that pu...
[ "Jingze Shi", "Zhangyang Peng", "Yizhang Zhu", "Yifan Wu", "Guang Liu", "Yuyu Luo" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2026-02-05T00:00:00
https://arxiv.org/abs/2602.05711
https://arxiv.org/pdf/2602.05711v1
2602.05711
10.48550/arXiv.2602.05711
0
0
true
https://github.com/flash-algo/omni-moe
arXiv.org
0.719
aece6c0cfe8674dbc23c441fd17884797dc83655b8233e0881aa0af0df7e2fdc
[ "arxiv", "semantic_scholar" ]
RASA: Routing-Aware Safety Alignment for Mixture-of-Experts Models
Mixture-of-Experts (MoE) language models introduce unique challenges for safety alignment due to their sparse routing mechanisms, which can enable degenerate optimization behaviors under standard full-parameter fine-tuning. In our preliminary experiments, we observe that naively applying full-parameter safety fine-tuni...
[ "Jiacheng Liang", "Yuhui Wang", "Tanqiu Jiang", "Ting Wang" ]
[ "cs.LG", "cs.AI", "cs.CR" ]
[ "Computer Science" ]
2026-02-04T00:00:00
https://arxiv.org/abs/2602.04448
https://arxiv.org/pdf/2602.04448v2
2602.04448
10.48550/arXiv.2602.04448
1
0
false
null
arXiv.org
0.4641
f449c475e42011401863c89cda0acda8ac033b189edb381f7c539e0c06df8832
[ "arxiv", "semantic_scholar" ]
Expert Selections In MoE Models Reveal (Almost) As Much As Text
We present a text-reconstruction attack on mixture-of-experts (MoE) language models that recovers tokens from expert selections alone. In MoE models, each token is routed to a subset of expert subnetworks; we show these routing decisions leak substantially more information than previously understood. Prior work using l...
[ "Amir Nuriyev", "Gabriel Kulp" ]
[ "cs.CL", "cs.CR" ]
[ "Computer Science" ]
2026-02-04T00:00:00
https://arxiv.org/abs/2602.04105
https://arxiv.org/pdf/2602.04105v3
2602.04105
10.48550/arXiv.2602.04105
3
1
false
null
arXiv.org
0.4641
a7387bd0d832dc92b6f08c7eab49fcfe50a4131b15e0a8adf40bfd38d158975c
[ "arxiv", "semantic_scholar" ]
DALI: A Workload-Aware Offloading Framework for Efficient MoE Inference on Local PCs
Mixture of Experts (MoE) architectures significantly enhance the capacity of LLMs without proportional increases in computation, but at the cost of a vast parameter size. Offloading MoE expert parameters to host memory and leveraging both CPU and GPU computation has recently emerged as a promising direction to support ...
[ "Zeyu Zhu", "Gang Li", "Peisong Wang", "Zitao Mo", "Minnan Pei", "Zhuoran Song", "Xiaoyao Liang", "Jian Cheng" ]
[ "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2026-02-03T00:00:00
https://arxiv.org/abs/2602.03495
https://arxiv.org/pdf/2602.03495v1
2602.03495
10.48550/arXiv.2602.03495
0
0
false
null
arXiv.org
0.4629
7072c3a0e8235fd9d9e83d75449d82d257ed84695edcb558b43eed02004e2fc7
[ "arxiv", "semantic_scholar" ]
Certain Head, Uncertain Tail: Expert-Sample for Test-Time Scaling in Fine-Grained MoE
Test-time scaling improves LLM performance by generating multiple candidate solutions, yet token-level sampling requires temperature tuning that trades off diversity against stability. Fine-grained MoE, featuring hundreds of well-trained experts per layer and multi-expert activation per token, offers an unexplored alte...
[ "Yuanteng Chen", "Peisong Wang", "Nanxin Zeng", "Yuantian Shao", "Shuang Qiu", "Gang Li", "Jing Liu", "Jian Cheng" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-02-02T00:00:00
https://arxiv.org/abs/2602.02443
https://arxiv.org/pdf/2602.02443v2
2602.02443
10.48550/arXiv.2602.02443
1
0
false
null
arXiv.org
0.4618
22f189e56be4b9d48a6cbdc6b44c485b6ae1c85229f52d687186bdcfc828a6d7
[ "arxiv", "semantic_scholar" ]
Dynamic Expert Sharing: Decoupling Memory from Parallelism in Mixture-of-Experts Diffusion LLMs
Among parallel decoding paradigms, diffusion large language models (dLLMs) have emerged as a promising candidate that balances generation quality and throughput. However, their integration with Mixture-of-Experts (MoE) architectures is constrained by an expert explosion: as the number of tokens generated in parallel in...
[ "Hao Mark Chen", "Zhiwen Mo", "Royson Lee", "Qianzhou Wang", "Da Li", "Shell Xu Hu", "Wayne Luk", "Timothy Hospedales", "Hongxiang Fan" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-01-31T00:00:00
https://arxiv.org/abs/2602.00879
https://arxiv.org/pdf/2602.00879v1
2602.00879
10.48550/arXiv.2602.00879
1
1
false
null
arXiv.org
0.4595
16a7898c1008a48fa4b56c0a8280ec753948998f80765541b523c44d5443c1ee
[ "arxiv", "semantic_scholar" ]
L2R: Low-Rank and Lipschitz-Controlled Routing for Mixture-of-Experts
Mixture-of-Experts (MoE) models scale neural networks by conditionally activating a small subset of experts, where the router plays a central role in determining expert specialization and overall model performance. However, many modern MoE systems still adopt linear routers in raw high-dimensional representation spaces...
[ "Minghao Yang", "Ren Togo", "Guang Li", "Takahiro Ogawa", "Miki Haseyama" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-01-29T00:00:00
https://arxiv.org/abs/2601.21349
https://arxiv.org/pdf/2601.21349v2
2601.21349
10.48550/arXiv.2601.21349
0
0
false
null
arXiv.org
0.4572
bbe46306bd85ad4b4ee4bf447514b2e8f0a0e9577759177f95cbb88a9d7683c7
[ "arxiv", "semantic_scholar" ]
BrainStack: Neuro-MoE with Functionally Guided Expert Routing for EEG-Based Language Decoding
Decoding linguistic information from electroencephalography (EEG) remains challenging due to the brain's distributed and nonlinear organization. We present BrainStack, a functionally guided neuro-mixture-of-experts (Neuro-MoE) framework that models the brain's modular functional architecture through anatomically partit...
[ "Ziyi Zhao", "Jinzhao Zhou", "Xiaowei Jiang", "Beining Cao", "Wenhao Ma", "Yang Shen", "Ren Li", "Yu-Kai Wang", "Chin-teng Lin" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-01-29T00:00:00
https://arxiv.org/abs/2601.21148
https://arxiv.org/pdf/2601.21148v1
2601.21148
10.48550/arXiv.2601.21148
0
0
false
null
arXiv.org
0.4572
0f8d0c64456ade09e8b431c207c62401c24df950d886db75485cb262cefa66c8
[ "arxiv", "semantic_scholar" ]
Seg-MoE: Multi-Resolution Segment-wise Mixture-of-Experts for Time Series Forecasting Transformers
Transformer-based models have recently made significant advances in accurate time-series forecasting, but even these architectures struggle to scale efficiently while capturing long-term temporal dynamics. Mixture-of-Experts (MoE) layers are a proven solution to scaling problems in natural language processing. However,...
[ "Evandro S. Ortigossa", "Eran Segal" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-01-29T00:00:00
https://arxiv.org/abs/2601.21641
https://arxiv.org/pdf/2601.21641v2
2601.21641
10.48550/arXiv.2601.21641
2
0
false
null
arXiv.org
0.4572
f54a138161fccab0d8f3dbcc8c3d5c4763f02c9d444e40798a750445e8705dfe
[ "arxiv", "semantic_scholar" ]
ShardMemo: Masked MoE Routing for Sharded Agentic LLM Memory
Agentic large language model (LLM) systems rely on external memory for long-horizon state and concurrent multi-agent execution, but centralized indexes and heuristic partitions become bottlenecks as memory volume and parallel access grow. We present ShardMemo, a budgeted tiered memory service with Tier A per-agent work...
[ "Yang Zhao", "Chengxiao Dai", "Yue Xiu", "Mengying Kou", "Yuliang Zheng", "Dusit Niyato" ]
[ "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-01-29T00:00:00
https://arxiv.org/abs/2601.21545
https://arxiv.org/pdf/2601.21545v1
2601.21545
10.48550/arXiv.2601.21545
0
0
false
null
arXiv.org
0.4572
5b9301d4ceb134c4f1c2f3c609e384ff79cbd7732b3bccc3d9498d01e918c44c
[ "arxiv", "semantic_scholar" ]
Dynamic Multi-Expert Projectors with Stabilized Routing for Multilingual Speech Recognition
Recent advances in LLM-based ASR connect frozen speech encoders with Large Language Models (LLMs) via lightweight projectors. While effective in monolingual settings, a single projector struggles to capture the diverse acoustic-to-semantic mappings required for multilingual ASR. To address this, we propose SMEAR-MoE, a...
[ "Isha Pandey", "Ashish Mittal", "Vartul Bahuguna", "Ganesh Ramakrishnan" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-01-27T00:00:00
https://arxiv.org/abs/2601.19451
https://arxiv.org/pdf/2601.19451v1
2601.19451
10.48550/arXiv.2601.19451
1
0
false
null
IEEE International Conference on Acoustics, Speech, and Signal Processing
0.4549
779454640ef85adfd6089ba2b7e2352f9cfebf42d4473e58d691729b7d996ed9
[ "arxiv", "semantic_scholar" ]
$\infty$-MoE: Generalizing Mixture of Experts to Infinite Experts
The Mixture of Experts (MoE) selects a few feed-forward networks (FFNs) per token, achieving an effective trade-off between computational cost and performance. In conventional MoE, each expert is treated as entirely independent, and experts are combined in a discrete space. As a result, when the number of experts incre...
[ "Shota Takashiro", "Takeshi Kojima", "Shohei Taniguchi", "Yusuke Iwasawa", "Yutaka Matsuo" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2026-01-25T00:00:00
https://arxiv.org/abs/2601.17680
https://arxiv.org/pdf/2601.17680v1
2601.17680
10.48550/arXiv.2601.17680
0
0
false
null
arXiv.org
0.4526
388ef9690bb443c6cac313ba44813aad4dd1684fafbb9c15d15846575b1ed745
[ "arxiv", "semantic_scholar" ]
Split-on-Share: Mixture of Sparse Experts for Task-Agnostic Continual Learning
Continual learning in Large Language Models (LLMs) is hindered by the plasticity-stability dilemma, where acquiring new capabilities often leads to catastrophic forgetting of previous knowledge. Existing methods typically treat parameters uniformly, failing to distinguish between specific task knowledge and shared capa...
[ "Fatema Siddika", "Md Anwar Hossen", "Tanwi Mallick", "Ali Jannesari" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-01-24T00:00:00
https://arxiv.org/abs/2601.17616
https://arxiv.org/pdf/2601.17616v2
2601.17616
10.48550/arXiv.2601.17616
0
0
false
null
arXiv.org
0.4515
c68300df7d856da59b5d9c50d52ce3e0d5d84025c170ed9902ac83e80f3a3a80
[ "arxiv", "semantic_scholar" ]
Mixture-of-Experts Models in Vision: Routing, Optimization, and Generalization
Mixture-of-Experts (MoE) architectures enable conditional computation by routing inputs to multiple expert subnetworks and are often motivated as a mechanism for scaling large language models. In this project, we instead study MoE behavior in an image classification setting, focusing on predictive performance, expert u...
[ "Adam Rokah", "Daniel Veress", "Caleb Caulk", "Sourav Sharan" ]
[ "cs.LG", "cs.CV" ]
[ "Computer Science" ]
2026-01-21T00:00:00
https://arxiv.org/abs/2601.15021
https://arxiv.org/pdf/2601.15021v1
2601.15021
10.48550/arXiv.2601.15021
0
0
true
https://github.com/moe-project-uu/mixture-of-experts-project
arXiv.org
0.6924
1e9b467e183ce8590e7f6265dbe578374760c88ef85f3ecb097a50ae6685088f
[ "arxiv", "semantic_scholar" ]
Understanding Multilingualism in Mixture-of-Experts LLMs: Routing Mechanism, Expert Specialization, and Layerwise Steering
Mixture-of-Experts (MoE) architectures have shown strong multilingual capabilities, yet the internal mechanisms underlying performance gains and cross-language differences remain insufficiently understood. In this work, we conduct a systematic analysis of MoE models, examining routing behavior and expert specialization...
[ "Yuxin Chen", "Zhengzhou Cai", "Xiangtian Ji", "Weixiang Zhao", "An Zhang", "Xiang Wang", "Tat-Seng Chua" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-01-20T00:00:00
https://arxiv.org/abs/2601.14050
https://arxiv.org/pdf/2601.14050v1
2601.14050
10.48550/arXiv.2601.14050
5
0
true
https://github.com/conctsai/Multilingualism-in-Mixture-of-Experts-LLMs
arXiv.org
0.6906
4a633722ab848010d8eddea89ed516e0ae257ab21164a41f334c5ccd38fb2c20
[ "arxiv", "semantic_scholar" ]
EMoE: Eigenbasis-Guided Routing for Mixture-of-Experts
The relentless scaling of deep learning models has led to unsustainable computational demands, positioning Mixture-of-Experts (MoE) architectures as a promising path towards greater efficiency. However, MoE models are plagued by two fundamental challenges: 1) a load imbalance problem known as the``rich get richer" phen...
[ "Anzhe Cheng", "Shukai Duan", "Shixuan Li", "Chenzhong Yin", "Mingxi Cheng", "Shahin Nazarian", "Paul Thompson", "Paul Bogdan" ]
[ "cs.LG", "cs.CV" ]
[ "Computer Science" ]
2026-01-17T00:00:00
https://arxiv.org/abs/2601.12137
https://arxiv.org/pdf/2601.12137v1
2601.12137
10.48550/arXiv.2601.12137
2
0
true
https://github.com/Belis0811/EMoE
IEEE International Conference on Acoustics, Speech, and Signal Processing
0.6853
1f6ee9eeb79bac95fe2c8633686439b3fe4c4f13ff649bee7d5fa3de9bc75cf6
[ "arxiv", "semantic_scholar" ]
MoST: Mixing Speech and Text with Modality-Aware Mixture of Experts
We present MoST (Mixture of Speech and Text), a novel multimodal large language model that seamlessly integrates speech and text processing through our proposed Modality-Aware Mixture of Experts (MAMoE) architecture. While current multimodal models typically process diverse modality representations with identical param...
[ "Yuxuan Lou", "Kai Yang", "Yang You" ]
[ "cs.CL", "cs.AI", "cs.LG", "cs.SD" ]
[ "Computer Science" ]
2026-01-15T00:00:00
https://arxiv.org/abs/2601.10272
https://arxiv.org/pdf/2601.10272v1
2601.10272
10.48550/arXiv.2601.10272
1
0
true
https://github.com/NUS-HPC-AI-Lab/MoST
arXiv.org
0.6818
fe13e99e36545b2df993d7bb01ea5b52cbdcceb71c6bb4c3b916acb1bb30027f
[ "arxiv", "semantic_scholar" ]
Horseshoe Mixtures-of-Experts (HS-MoE)
Horseshoe mixtures-of-experts (HS-MoE) models provide a Bayesian framework for sparse expert selection in mixture-of-experts architectures. We combine the horseshoe prior's adaptive global-local shrinkage with input-dependent gating, yielding data-adaptive sparsity in expert usage. Our primary methodological contributi...
[ "Nick Polson", "Vadim Sokolov" ]
[ "stat.ML", "cs.LG" ]
[ "Computer Science", "Mathematics" ]
2026-01-14T00:00:00
https://arxiv.org/abs/2601.09043
https://arxiv.org/pdf/2601.09043v1
2601.09043
10.48550/arXiv.2601.09043
0
0
false
null
arXiv.org
0.44
17f5d046f53c6df60b02b0ffd68fbc13e0a676372a171e85deddd52ab67ee18a
[ "arxiv", "semantic_scholar" ]
TAG-MoE: Task-Aware Gating for Unified Generative Mixture-of-Experts
Unified image generation and editing models suffer from severe task interference in dense diffusion transformers architectures, where a shared parameter space must compromise between conflicting objectives (e.g., local editing v.s. subject-driven generation). While the sparse Mixture-of-Experts (MoE) paradigm is a prom...
[ "Yu Xu", "Hongbin Yan", "Juan Cao", "Yiji Cheng", "Tiankai Hang", "Runze He", "Zijin Yin", "Shiyi Zhang", "Yuxin Zhang", "Jintao Li", "Chunyu Wang", "Qinglin Lu", "Tong-Yee Lee", "Fan Tang" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2026-01-12T00:00:00
https://arxiv.org/abs/2601.08881
https://arxiv.org/pdf/2601.08881v2
2601.08881
10.48550/arXiv.2601.08881
6
0
false
null
arXiv.org
0.4377
4df1244db8a0122524ad699eb57bc3ed7ace11d10ad3e529ee38d18abfbbfb74
[ "arxiv", "semantic_scholar" ]
MoE-DisCo:Low Economy Cost Training Mixture-of-Experts Models
Training large-scale Mixture-of-Experts (MoE) models typically requires high-memory, high-bandwidth GPUs (e.g., A100), and their high cost has become a major barrier to large-model training. In contrast, affordable hardware is low-cost but constrained by memory capacity and bandwidth, making it unsuitable for direct LL...
[ "Xin Ye", "Daning Cheng", "Boyang Zhang", "Yunquan Zhang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-01-11T00:00:00
https://arxiv.org/abs/2601.06857
https://arxiv.org/pdf/2601.06857v1
2601.06857
10.48550/arXiv.2601.06857
1
0
false
null
arXiv.org
0.4366
dbbb0d1c7a8cd5e006ab422b342f0ee25ef5c466bfcb5fc06b43e97c9ad4b6b4
[ "arxiv", "semantic_scholar" ]
Spectral Manifold Regularization for Stable and Modular Routing in Deep MoE Architectures
Mixture of Experts (MoE) architectures enable efficient scaling of neural networks but suffer from expert collapse, where routing converges to a few dominant experts. This reduces model capacity and causes catastrophic interference during adaptation. We propose the Spectrally-Regularized Mixture of Experts (SR-MoE), wh...
[ "Ibrahim Delibasoglu" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-01-07T00:00:00
https://arxiv.org/abs/2601.03889
https://arxiv.org/pdf/2601.03889v1
2601.03889
10.48550/arXiv.2601.03889
1
0
false
null
arXiv.org
0.432
e741de7adc9cf1ffd1ce278f7d31e6d9107befa4ac8dd31a0352a5162858de1f
[ "arxiv", "semantic_scholar" ]
Routing by Analogy: kNN-Augmented Expert Assignment for Mixture-of-Experts
Mixture-of-Experts (MoE) architectures scale large language models efficiently by employing a parametric ``router'' to dispatch tokens to a sparse subset of experts. Typically, this router is trained once and then frozen, rendering routing decisions brittle under distribution shifts. We address this limitation by intro...
[ "Boxuan Lyu", "Soichiro Murakami", "Hidetaka Kamigaito", "Peinan Zhang" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2026-01-05T00:00:00
https://arxiv.org/abs/2601.02144
https://arxiv.org/pdf/2601.02144v2
2601.02144
10.48550/arXiv.2601.02144
0
0
false
null
arXiv.org
0.4297
cac6ebee88d7f3efa51f26446aae913568067d99fb4b1b453755de172daa39d7
[ "arxiv", "semantic_scholar" ]
MambaFormer: Token-Level Guided Routing Mixture-of-Experts for Accurate and Efficient Clinical Assistance
The deployment of large language models (LLMs) in real-world clinical applications is constrained by the fundamental trade-off between computational cost and the efficiency of linear-time models. To address this, we propose an LLM-based MambaFormer hybrid Mixture-of-Experts (MoE) framework for efficient medical questio...
[ "Hamad Khan", "Saddam Hussain Khan" ]
[ "cs.CV", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-01-03T00:00:00
https://arxiv.org/abs/2601.01260
https://arxiv.org/pdf/2601.01260v1
2601.01260
10.48550/arXiv.2601.01260
1
0
false
null
arXiv.org
0.4274
1230e5d4bb15257effde3798385f7823755c1c82a72fa9012a509716055fcb86
[ "arxiv", "semantic_scholar" ]
HFedMoE: Resource-aware Heterogeneous Federated Learning with Mixture-of-Experts
While federated learning (FL) enables fine-tuning of large language models (LLMs) without compromising data privacy, the substantial size of an LLM renders on-device training impractical for resource-constrained clients, such as mobile devices. Thus, Mixture-of-Experts (MoE) models have emerged as a computation-efficie...
[ "Zihan Fang", "Zheng Lin", "Senkang Hu", "Yanan Ma", "Yihang Tao", "Yiqin Deng", "Xianhao Chen", "Yuguang Fang" ]
[ "cs.LG", "cs.AI", "cs.NI" ]
[ "Computer Science" ]
2026-01-02T00:00:00
https://arxiv.org/abs/2601.00583
https://arxiv.org/pdf/2601.00583v1
2601.00583
10.48550/arXiv.2601.00583
9
0
false
null
arXiv.org
0.4263
6401f98dfcb0f4abf7d136339cbcd8e00bed90606d5f420e1c3ba3cc6b279407
[ "arxiv", "semantic_scholar" ]
FLEX-MoE: Federated Mixture-of-Experts with Load-balanced Expert Assignment for Edge Computing
Mixture-of-Experts (MoE) models enable scalable neural networks through conditional computation, offering enhanced effectiveness and efficiency for next-generation wireless communications. However, deploying MoE with federated learning (FL) over wireless and IoT edge networks faces two critical challenges: 1) resource-...
[ "Boyang Zhang", "Xiaobing Chen", "Songyang Zhang", "Shuai Zhang", "Xiangwei Zhou", "Jian Zhang", "Mingxuan Sun" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-12-28T00:00:00
https://arxiv.org/abs/2512.23070
https://arxiv.org/pdf/2512.23070v2
2512.23070
10.48550/arXiv.2512.23070
1
0
false
null
arXiv.org
0.4205
9b1bb34ffe1bdd6a02ba7b6629865ee10596a4804d3517ea32ed10545c620fce
[ "arxiv", "semantic_scholar" ]
Text-Routed Sparse Mixture-of-Experts Model with Explanation and Temporal Alignment for Multi-Modal Sentiment Analysis
Human-interaction-involved applications underscore the need for Multi-modal Sentiment Analysis (MSA). Although many approaches have been proposed to address the subtle emotions in different modalities, the power of explanations and temporal alignments is still underexplored. Thus, this paper proposes the Text-routed sp...
[ "Dongning Rao", "Yunbiao Zeng", "Zhihua Jiang", "Jujian Lv" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-12-28T00:00:00
https://arxiv.org/abs/2512.22741
https://arxiv.org/pdf/2512.22741v1
2512.22741
10.48550/arXiv.2512.22741
0
0
false
null
arXiv.org
0.4205
dfcd97d4cfd46aa071450c96d6d51cc6fab58b715790856086319ab493078c03
[ "arxiv", "semantic_scholar" ]
Hybrid Quantum-Classical Mixture of Experts: Unlocking Topological Advantage via Interference-Based Routing
The Mixture-of-Experts (MoE) architecture has emerged as a powerful paradigm for scaling deep learning models, yet it is fundamentally limited by challenges such as expert imbalance and the computational complexity of classical routing mechanisms. This paper investigates the potential of Quantum Machine Learning (QML) ...
[ "Reda Heddad", "Lamiae Bouanane" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-12-25T00:00:00
https://arxiv.org/abs/2512.22296
https://arxiv.org/pdf/2512.22296v1
2512.22296
10.48550/arXiv.2512.22296
0
0
true
https://github.com/RH2004/QMOE
arXiv.org
0.6446
c6654c2b0914954b53a1d886d831eccbf8d75b06db012c8dc0e60601a4064f2d
[ "arxiv", "semantic_scholar" ]
Efficient MoE Inference with Fine-Grained Scheduling of Disaggregated Expert Parallelism
The mixture-of-experts (MoE) architecture scales model size with sublinear computational increase but suffers from memory-intensive inference due to KV caches and sparse expert activation. Recent disaggregated expert parallelism (DEP) distributes attention and experts to dedicated GPU groups but lacks support for share...
[ "Xinglin Pan", "Shaohuai Shi", "Wenxiang Lin", "Yuxin Wang", "Zhenheng Tang", "Wei Wang", "Xiaowen Chu" ]
[ "cs.DC", "cs.AI" ]
[ "Computer Science" ]
2025-12-25T00:00:00
https://arxiv.org/abs/2512.21487
https://arxiv.org/pdf/2512.21487v1
2512.21487
10.48550/arXiv.2512.21487
2
0
false
null
arXiv.org
0.4171
3023f80e009287d851b250be96d9c16ae2ba218b8fc62400bdd96d54591ab238
[ "arxiv", "semantic_scholar" ]
MoE-DiffuSeq: Enhancing Long-Document Diffusion Models with Sparse Attention and Mixture of Experts
We propose \textbf{MoE-DiffuSeq}, a diffusion-based framework for efficient long-form text generation that integrates sparse attention with a Mixture-of-Experts (MoE) architecture. Existing sequence diffusion models suffer from prohibitive computational and memory costs when scaling to long documents, largely due to de...
[ "Alexandros Christoforos", "Chadbourne Davis" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-12-23T00:00:00
https://arxiv.org/abs/2512.20604
https://arxiv.org/pdf/2512.20604v2
2512.20604
10.48550/arXiv.2512.20604
1
0
false
null
arXiv.org
0.4148
16d417c1ee13fdeba3ccf3a93950d9983f92e5e8334805d7bca977060dbd6585
[ "arxiv", "semantic_scholar" ]
How Many Experts Are Enough? Towards Optimal Semantic Specialization for Mixture-of-Experts
Finding the optimal configuration of Sparse Mixture-ofExperts (SMoE) that maximizes semantic differentiation among experts is essential for exploiting the full potential of MoE architectures. However, existing SMoE frameworks either heavily rely on hyperparameter tuning or overlook the importance of diversifying semant...
[ "Sumin Park", "Noseong Park" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-12-21T00:00:00
https://arxiv.org/abs/2512.19765
https://arxiv.org/pdf/2512.19765v1
2512.19765
10.48550/arXiv.2512.19765
0
0
false
null
AAAI Conference on Artificial Intelligence
0.4125
09cdadce27619dd6ee8ccadbe388198877ef454c248dcfb03fc4b5c390151476
[ "arxiv", "semantic_scholar" ]
Secret mixtures of experts inside your LLM
Despite being one of the earliest neural network layers, the Multilayer Perceptron (MLP) is arguably one of the least understood parts of the transformer architecture due to its dense computation and lack of easy visualization. This paper seeks to understand the MLP layers in dense LLM models by hypothesizing that thes...
[ "Enric Boix-Adsera" ]
[ "cs.LG", "cs.AI", "stat.ML" ]
[ "Computer Science", "Mathematics" ]
2025-12-20T00:00:00
https://arxiv.org/abs/2512.18452
https://arxiv.org/pdf/2512.18452v1
2512.18452
10.48550/arXiv.2512.18452
0
0
false
null
arXiv.org
0.4114
7ddffe35f92b69a66cea873119849100e8a0a6fbedc9385bf1c570b2a1bc9993
[ "arxiv", "semantic_scholar" ]
MoE Pathfinder: Trajectory-driven Expert Pruning
Mixture-of-experts (MoE) architectures used in large language models (LLMs) achieve state-of-the-art performance across diverse tasks yet face practical challenges such as deployment complexity and low activation efficiency. Expert pruning has thus emerged as a promising solution to reduce computational overhead and si...
[ "Xican Yang", "Yuanhe Tian", "Yan Song" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-12-20T00:00:00
https://arxiv.org/abs/2512.18425
https://arxiv.org/pdf/2512.18425v1
2512.18425
10.48550/arXiv.2512.18425
3
0
false
null
arXiv.org
0.4114
3c01cfba72f29b4b35903cb2a0a49c04903aaa98d824c34d8846d248b196f7f3
[ "arxiv", "semantic_scholar" ]
Janus: Disaggregating Attention and Experts for Scalable MoE Inference
Serving large Mixture-of-Experts (MoE) models is challenging because of their large memory footprints, heterogeneous resource demands, and highly dynamic inference workloads. Most existing MoE inference systems deploy the entire model as a monolithic unit, forcing attention and MoE layers to share the same resource con...
[ "Zhexiang Zhang", "Ye Wang", "Yumiao Zhao", "Jiayu Xiao", "Qianjing Yang", "Xiangyu Wang", "Jingzhe Jiang", "Qizhen Weng", "Ruichuan Chen", "Shaohuai Shi", "Adel N. Toosi", "Yin Chen", "Minchen Yu" ]
[ "cs.DC" ]
[ "Computer Science" ]
2025-12-15T00:00:00
https://arxiv.org/abs/2512.13525
https://arxiv.org/pdf/2512.13525v3
2512.13525
10.48550/arXiv.2512.13525
4
0
false
null
arXiv.org
0.4056
ace7e9297e5cc28c866ad86f9709a6ec76f2ef420d4813e716721c089e252336
[ "arxiv", "semantic_scholar" ]
Efficient MoE Serving in the Memory-Bound Regime: Balance Activated Experts, Not Tokens
Expert Parallelism (EP) permits Mixture of Experts (MoE) models to scale beyond a single GPU. To address load imbalance across GPUs in EP, existing approaches aim to balance the number of tokens each GPU processes. Surprisingly, we find that this objective degrades performance rather than improving it when processing i...
[ "Yanpeng Yu", "Haiyue Ma", "Krish Agarwal", "Nicolai Oswald", "Qijing Huang", "Hugo Linsenmaier", "Chunhui Mei", "Ritchie Zhao", "Ritika Borkar", "Bita Darvish Rouhani", "David Nellans", "Ronny Krashinsky", "Anurag Khandelwal" ]
[ "cs.DC", "cs.AR" ]
[ "Computer Science" ]
2025-12-10T00:00:00
https://arxiv.org/abs/2512.09277
https://arxiv.org/pdf/2512.09277v1
2512.09277
10.48550/arXiv.2512.09277
2
0
false
null
arXiv.org
0.3999
7732d4b8b3df31a35e92a6c4fb130f5b7f0611abfad2f274df4d5cf42e530ce7
[ "arxiv", "semantic_scholar" ]
Each Prompt Matters: Scaling Reinforcement Learning Without Wasting Rollouts on Hundred-Billion-Scale MoE
We present CompassMax-V3-Thinking, a hundred-billion-scale MoE reasoning model trained with a new RL framework built on one principle: each prompt must matter. Scaling RL to this size exposes critical inefficiencies-zero-variance prompts that waste rollouts, unstable importance sampling over long horizons, advantage in...
[ "Anxiang Zeng", "Haibo Zhang", "Hailing Zhang", "Kaixiang Mo", "Liang Yao", "Ling Hu", "Long Zhang", "Shuman Liu", "Shuyi Xie", "Yanshi Li", "Yizhang Chen", "Yuepeng Sheng", "Yuwei Huang", "Zhaochen Xu", "Zhiqiang Zhou", "Ziqin Liew" ]
[ "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-12-08T00:00:00
https://arxiv.org/abs/2512.07710
https://arxiv.org/pdf/2512.07710v1
2512.07710
10.48550/arXiv.2512.07710
1
0
false
null
arXiv.org
0.3976
931b940202eb676b0b18bf356c422479d8f0dfd2492f65f289effd7b39985506
[ "arxiv", "semantic_scholar" ]
Stable-MoE: Lyapunov-based Token Routing for Distributed Mixture-of-Experts Training over Edge Networks
The sparse activation mechanism of mixture of experts (MoE) model empowers edge intelligence with enhanced training efficiency and reduced computational resource consumption. However, traditional token routing in distributed MoE training faces significant challenges in resource-constrained edge networks characterized b...
[ "Long Shi", "Bingyan Ou", "Kang Wei", "Weihao Zhu", "Zhe Wang", "Zhiyong Chen" ]
[ "cs.DC" ]
[ "Computer Science" ]
2025-12-07T00:00:00
https://arxiv.org/abs/2512.06784
https://arxiv.org/pdf/2512.06784v3
2512.06784
10.48550/arXiv.2512.06784
2
0
false
null
IEEE Transactions on Vehicular Technology
0.3965
42d19b42514972a00fa86949fa099b03acadfa34c3560766855670384a939504
[ "arxiv", "semantic_scholar" ]
Statistic-Augmented, Decoupled MoE Routing and Aggregating in Autonomous Driving
Autonomous driving (AD) scenarios are inherently complex and diverse, posing significant challenges for a single deep learning model to effectively cover all possible conditions, such as varying weather, traffic densities, and road types. Large Model (LM)-Driven Mixture of Experts (MoE) paradigm offers a promising solu...
[ "Wei-Bin Kou", "Guangxu Zhu", "Jingreng Lei", "Chen Zhang", "Yik-Chung Wu", "Jianping Wang" ]
[ "cs.RO" ]
[ "Computer Science" ]
2025-12-07T00:00:00
https://arxiv.org/abs/2512.06664
https://arxiv.org/pdf/2512.06664v1
2512.06664
10.48550/arXiv.2512.06664
0
0
false
null
arXiv.org
0.3965
a9f36bdb1d4cfa210dd4df13f946e841fa50c1c1ef69b6da3a4abfdc2e40dc5a
[ "arxiv", "semantic_scholar" ]
OD-MoE: On-Demand Expert Loading for Cacheless Edge-Distributed MoE Inference
Mixture-of-Experts (MoE), while offering significant advantages as a Large Language Model (LLM) architecture, faces substantial challenges when deployed on low-cost edge devices with tight memory constraints. Expert offloading mitigates this issue by storing expert parameters in CPU memory and caching a subset of popul...
[ "Liujianfu Wang", "Yuyang Du", "Yuchen Pan", "Soung Chang Liew", "Jiacheng Liu", "Kexin Chen" ]
[ "cs.DC" ]
[ "Computer Science" ]
2025-12-03T00:00:00
https://arxiv.org/abs/2512.03927
https://arxiv.org/pdf/2512.03927v1
2512.03927
10.48550/arXiv.2512.03927
0
0
false
null
arXiv.org
0.3919
fc3d88b3aa1b79c80ba43ffcc0df462de85266d6e2e95830ecbdf7c225c5600c
[ "arxiv", "semantic_scholar" ]
A Theoretical Framework for Auxiliary-Loss-Free Load Balancing of Sparse Mixture-of-Experts in Large-Scale AI Models
In large-scale AI training, Sparse Mixture-of-Experts (s-MoE) layers enable scaling by activating only a small subset of experts per token. An operational challenge in this design is load balancing: routing tokens to minimize the number of idle experts, which is important for the efficient utilization of costly GPUs an...
[ "X. Y. Han", "Yuan Zhong" ]
[ "math.OC", "cs.AI", "cs.LG" ]
[ "Computer Science", "Mathematics" ]
2025-12-03T00:00:00
https://arxiv.org/abs/2512.03915
https://arxiv.org/pdf/2512.03915v3
2512.03915
10.48550/arXiv.2512.03915
1
0
false
null
arXiv.org
0.3919
6a258fcdaaaa9aebceb18605f4754bffb6c061c0b583ff44e9ddead2028437fd
[ "arxiv", "semantic_scholar" ]
Exploiting the Experts: Unauthorized Compression in MoE-LLMs
Mixture-of-Experts (MoE) architectures are increasingly adopted in large language models (LLMs) for their scalability and efficiency. However, their modular structure introduces a unique vulnerability: adversaries can attempt to compress or repurpose models by pruning experts and cheaply fine-tuning the remainder, effe...
[ "Pinaki Prasad Guha Neogi", "Ahmad Mohammadshirazi", "Dheeraj Kulshrestha", "Rajiv Ramnath" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-11-22T00:00:00
https://arxiv.org/abs/2511.19480
https://arxiv.org/pdf/2511.19480v1
2511.19480
10.48550/arXiv.2511.19480
1
0
false
null
arXiv.org
0.3793
099ecc68acddd44a69c80f723c4998e52e73c574ec28e8410fb24567aedf51dd
[ "arxiv", "semantic_scholar" ]
Dynamic Expert Quantization for Scalable Mixture-of-Experts Inference
Mixture-of-Experts (MoE) has become a practical architecture for scaling LLM capacity while keeping per-token compute modest, but deploying MoE models on a single, memory-limited GPU remains difficult because expert weights dominate the HBM footprint. Existing expert offloading and prefetching systems reduce the reside...
[ "Kexin Chu", "Dawei Xiang", "Zixu Shen", "Yiwei Yang", "Zecheng Liu", "Wei Zhang" ]
[ "cs.PF", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-11-19T00:00:00
https://arxiv.org/abs/2511.15015
https://arxiv.org/pdf/2511.15015v3
2511.15015
10.48550/arXiv.2511.15015
1
0
false
null
arXiv.org
0.3758
16420223fb875466df3924b1d6c115ca1527bb818e901a088e7fc754977eefaf
[ "arxiv", "semantic_scholar" ]
MoE-SpeQ: Speculative Quantized Decoding with Proactive Expert Prefetching and Offloading for Mixture-of-Experts
The immense memory requirements of state-of-the-art Mixture-of-Experts (MoE) models present a significant challenge for inference, often exceeding the capacity of a single accelerator. While offloading experts to host memory is a common solution, it introduces a severe I/O bottleneck over the PCIe bus, as the data-depe...
[ "Wenfeng Wang", "Jiacheng Liu", "Xiaofeng Hou", "Xinfeng Xia", "Peng Tang", "Mingxuan Zhang", "Chao Li", "Minyi Guo" ]
[ "cs.LG", "cs.DC" ]
[ "Computer Science" ]
2025-11-18T00:00:00
https://arxiv.org/abs/2511.14102
https://arxiv.org/pdf/2511.14102v1
2511.14102
10.48550/arXiv.2511.14102
3
0
false
null
arXiv.org
0.3747
b04dd1cb338035b28c2ef9c426691fed4c8186e58de0584182c5caa9436f8dc3
[ "arxiv", "semantic_scholar" ]
Orchestrating Heterogeneous Experts: A Scalable MoE Framework with Anisotropy-Preserving Fusion
In cross-border e-commerce, search relevance modeling faces the dual challenge of extreme linguistic diversity and fine-grained semantic nuances. Existing approaches typically rely on scaling up a single monolithic Large Language Model (LLM). However, our empirical analysis reveals that single models suffer from uneven...
[ "Ye Liu", "Xu Chen", "Wuji Chen", "Mang Li" ]
[ "cs.IR", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-11-18T00:00:00
https://arxiv.org/abs/2602.00003
https://arxiv.org/pdf/2602.00003v2
2602.00003
10.48550/arXiv.2602.00003
0
0
true
null
null
0.4428
fa3499c61e180b0cfbf393d18ccbda6cafb3678af006963e24d05b888da91ffc
[ "arxiv", "semantic_scholar" ]
YOLO Meets Mixture-of-Experts: Adaptive Expert Routing for Robust Object Detection
This paper presents a novel Mixture-of-Experts framework for object detection, incorporating adaptive routing among multiple YOLOv9-T experts to enable dynamic feature specialization and achieve higher mean Average Precision (mAP) and Average Recall (AR) compared to a single YOLOv9-T model.
[ "Ori Meiraz", "Sharon Shalev", "Avishai Weizman" ]
[ "cs.CV" ]
[ "Computer Science" ]
2025-11-17T00:00:00
https://arxiv.org/abs/2511.13344
https://arxiv.org/pdf/2511.13344v4
2511.13344
10.48550/arXiv.2511.13344
1
0
false
null
arXiv.org
0.3735
2344e6adea3b65e525574797d2910aedbd1c55cf494b0855dc8e89dbde8d1a72
[ "arxiv", "semantic_scholar" ]
Uni-MoE-2.0-Omni: Scaling Language-Centric Omnimodal Large Model with Advanced MoE, Training and Data
We present Uni-MoE 2.0 from the Lychee family. As a fully open-source omnimodal large model (OLM), it substantially advances Lychee's Uni-MoE series in language-centric multimodal understanding, reasoning, and generating. Based on the dense LLM, we build Uni-MoE-2.0-Omni from scratch through three core contributions: d...
[ "Yunxin Li", "Xinyu Chen", "Shenyuan Jiang", "Haoyuan Shi", "Zhenyu Liu", "Xuanyu Zhang", "Nanhao Deng", "Zhenran Xu", "Yicheng Ma", "Meishan Zhang", "Baotian Hu", "Min Zhang" ]
[ "cs.CL", "cs.AI", "cs.CV" ]
[ "Computer Science" ]
2025-11-16T00:00:00
https://arxiv.org/abs/2511.12609
https://arxiv.org/pdf/2511.12609v2
2511.12609
10.48550/arXiv.2511.12609
16
1
true
https://github.com/HITsz-TMG/Uni-MoE
arXiv.org
0.5755
6485d4fccd3870f6b77a4862a3fc46460656be6469fb736510be8f1d14c45fd4
[ "arxiv", "semantic_scholar" ]
SAC-MoE: Reinforcement Learning with Mixture-of-Experts for Control of Hybrid Dynamical Systems with Uncertainty
Hybrid dynamical systems result from the interaction of continuous-variable dynamics with discrete events and encompass various systems such as legged robots, vehicles and aircrafts. Challenges arise when the system's modes are characterized by unobservable (latent) parameters and the events that cause system dynamics ...
[ "Leroy D'Souza", "Akash Karthikeyan", "Yash Vardhan Pant", "Sebastian Fischmeister" ]
[ "cs.RO", "eess.SY" ]
[ "Computer Science", "Engineering" ]
2025-11-15T00:00:00
https://arxiv.org/abs/2511.12361
https://arxiv.org/pdf/2511.12361v1
2511.12361
10.48550/arXiv.2511.12361
1
0
false
null
arXiv.org
0.3713
99de0ddc4e855f1b80148a763ddc9ba2d36f1c12f302e77c4927e242f03afaa8
[ "arxiv", "semantic_scholar" ]
ERMoE: Eigen-Reparameterized Mixture-of-Experts for Stable Routing and Interpretable Specialization
Mixture-of-Experts (MoE) architectures expand model capacity by sparsely activating experts but face two core challenges: misalignment between router logits and each expert's internal structure leads to unstable routing and expert underutilization, and load imbalances create straggler bottlenecks. Standard solutions, s...
[ "Anzhe Cheng", "Shukai Duan", "Shixuan Li", "Chenzhong Yin", "Mingxi Cheng", "Heng Ping", "Tamoghna Chattopadhyay", "Sophia I Thomopoulos", "Shahin Nazarian", "Paul Thompson", "Paul Bogdan" ]
[ "cs.CV" ]
[ "Computer Science" ]
2025-11-14T00:00:00
https://arxiv.org/abs/2511.10971
https://arxiv.org/pdf/2511.10971v2
2511.10971
10.48550/arXiv.2511.10971
4
0
false
null
arXiv.org
0.3701
14ff08dceb360fea2d99af8aa5c1b61fe6e27848acc6d4a7db0b0b4b6546a8e3
[ "arxiv", "semantic_scholar" ]
Selective Sinkhorn Routing for Improved Sparse Mixture of Experts
Sparse Mixture-of-Experts (SMoE) models are scalable and computationally efficient, enabling large increases in model capacity with limited inference overhead. Existing SMoE methods often depend on auxiliary objectives, such as load-balancing loss and z-loss, or additional trainable components such as noisy gating. Whi...
[ "Duc Anh Nguyen", "Huu Binh Ta", "Nhuan Le Duc", "Tan Minh Nguyen", "Toan Tran" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-11-12T00:00:00
https://arxiv.org/abs/2511.08972
https://arxiv.org/pdf/2511.08972v2
2511.08972
10.48550/arXiv.2511.08972
0
0
false
null
arXiv.org
0.3678
d4196b1583d555be7e55d4a4ccdac140b71419be07d969a9ad8293c8be3b46ad
[ "arxiv", "semantic_scholar" ]
Routing Manifold Alignment Improves Generalization of Mixture-of-Experts LLMs
Sparse Mixture-of-Experts (MoE) have been widely adopted in recent large language models since it can efficiently scale up the model capability without increasing the inference cost. However, evaluations on broad downstream tasks reveal a consistent suboptimality of the routers in existing MoE LLMs, which results in a ...
[ "Zhongyang Li", "Ziyue Li", "Tianyi Zhou" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-11-10T00:00:00
https://arxiv.org/abs/2511.07419
https://arxiv.org/pdf/2511.07419v2
2511.07419
10.48550/arXiv.2511.07419
1
1
false
null
arXiv.org
0.3655
a0c9b39f2b7bfef42feb98fd79457ee6fc1b978f037ac0806b700b76b682956a
[ "arxiv", "semantic_scholar" ]
One Router to Route Them All: Homogeneous Expert Routing for Heterogeneous Graph Transformers
A common practice in heterogeneous graph neural networks (HGNNs) is to condition parameters on node/edge types, assuming types reflect semantic roles. However, this can cause overreliance on surface-level labels and impede cross-type knowledge transfer. We explore integrating Mixture-of-Experts (MoE) into HGNNs--a dire...
[ "Georgiy Shakirov", "Albert Arakelov" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-11-10T00:00:00
https://arxiv.org/abs/2511.07603
https://arxiv.org/pdf/2511.07603v2
2511.07603
10.48550/arXiv.2511.07603
0
0
false
null
arXiv.org
0.3655
153109bdb7c31da34c653bdb6361d7320c022a3dddba4cb7e020d603b93286b4
[ "arxiv", "semantic_scholar" ]
Route Experts by Sequence, not by Token
Mixture-of-Experts (MoE) architectures scale large language models (LLMs) by activating only a subset of experts per token, but the standard TopK routing assigns the same fixed number of experts to all tokens, ignoring their varying complexity. Prior adaptive routing methods introduce additional modules and hyperparame...
[ "Tiansheng Wen", "Yifei Wang", "Aosong Feng", "Long Ma", "Xinyang Liu", "Yifan Wang", "Lixuan Guo", "Bo Chen", "Stefanie Jegelka", "Chenyu You" ]
[ "cs.LG", "cs.AI", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2025-11-09T00:00:00
https://arxiv.org/abs/2511.06494
https://arxiv.org/pdf/2511.06494v2
2511.06494
10.48550/arXiv.2511.06494
2
0
true
https://github.com/Y-Research-SBU/SeqTopK
arXiv.org
0.5631
a1767610f9530aa739c3e30d09644cbae641460bcd894c781b3097743d06b0d4
[ "arxiv", "semantic_scholar" ]
PuzzleMoE: Efficient Compression of Large Mixture-of-Experts Models via Sparse Expert Merging and Bit-packed inference
Mixture-of-Experts (MoE) models have shown strong potential in scaling language models efficiently by activating only a small subset of experts per input. However, their widespread deployment remains limited due to the high memory overhead associated with storing all expert parameters, particularly as the number of exp...
[ "Yushu Zhao", "Zheng Wang", "Minjia Zhang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-11-06T00:00:00
https://arxiv.org/abs/2511.04805
https://arxiv.org/pdf/2511.04805v1
2511.04805
10.48550/arXiv.2511.04805
4
1
false
null
arXiv.org
0.3609
b07c6a5f8c0f15a8d3749076fad6b9d77ebc5aefd112cd27f987fdb5a8522ba2
[ "arxiv", "semantic_scholar" ]
Opportunistic Expert Activation: Batch-Aware Expert Routing for Faster Decode Without Retraining
An increasing number of LLMs employ Mixture-of-Experts (MoE) architectures where the feed-forward layer is replaced by a pool of experts and each token only activates a small subset of them. During autoregressive generation, these models often enter a memory-bound regime even for moderate batch sizes because the averag...
[ "Costin-Andrei Oncescu", "Qingyang Wu", "Wai Tong Chung", "Robert Wu", "Bryan Gopal", "Junxiong Wang", "Tri Dao", "Ben Athiwaratkun" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-11-04T00:00:00
https://arxiv.org/abs/2511.02237
https://arxiv.org/pdf/2511.02237v1
2511.02237
10.48550/arXiv.2511.02237
3
1
false
null
arXiv.org
0.3586
04fb65c141d981f69671f96f47d9f767667c35ae872dcc68574515535ae0d069
[ "arxiv", "semantic_scholar" ]
CryptoMoE: Privacy-Preserving and Scalable Mixture of Experts Inference via Balanced Expert Routing
Private large language model (LLM) inference based on cryptographic primitives offers a promising path towards privacy-preserving deep learning. However, existing frameworks only support dense LLMs like LLaMA-1 and struggle to scale to mixture-of-experts (MoE) architectures. The key challenge comes from securely evalua...
[ "Yifan Zhou", "Tianshi Xu", "Jue Hong", "Ye Wu", "Meng Li" ]
[ "cs.CR" ]
[ "Computer Science" ]
2025-11-03T00:00:00
https://arxiv.org/abs/2511.01197
https://arxiv.org/pdf/2511.01197v3
2511.01197
10.48550/arXiv.2511.01197
2
0
true
https://github.com/PKU-SEC-Lab/CryptoMoE
arXiv.org
0.5525
fa8c683ea312d5ff3518197b16240b02db46d57fa958b35c945a03f6f2715825
[ "arxiv", "semantic_scholar" ]
ExpertFlow: Adaptive Expert Scheduling and Memory Coordination for Efficient MoE Inference
The expansion of large language models is increasingly limited by the constrained memory capacity of modern GPUs. To mitigate this, Mixture-of-Experts (MoE) architectures activate only a small portion of parameters during inference, significantly lowering both memory demand and computational overhead. However, conventi...
[ "Zixu Shen", "Kexin Chu", "Yifan Zhang", "Dawei Xiang", "Runxin Wu", "Wei Zhang" ]
[ "cs.DC", "cs.AI", "cs.PF" ]
[ "Computer Science" ]
2025-10-30T00:00:00
https://arxiv.org/abs/2510.26730
https://arxiv.org/pdf/2510.26730v1
2510.26730
10.48550/arXiv.2510.26730
3
0
false
null
arXiv.org
0.3529
49f2fcad29ebe4a9e8ff55b43ae51f6205a7ce60e42e0e9f75ea3d7466434d97
[ "arxiv", "semantic_scholar" ]
Routing Matters in MoE: Scaling Diffusion Transformers with Explicit Routing Guidance
Mixture-of-Experts (MoE) has emerged as a powerful paradigm for scaling model capacity while preserving computational efficiency. Despite its notable success in large language models (LLMs), existing attempts to apply MoE to Diffusion Transformers (DiTs) have yielded limited gains. We attribute this gap to fundamental ...
[ "Yujie Wei", "Shiwei Zhang", "Hangjie Yuan", "Yujin Han", "Zhekai Chen", "Jiayu Wang", "Difan Zou", "Xihui Liu", "Yingya Zhang", "Yu Liu", "Hongming Shan" ]
[ "cs.CV" ]
[ "Computer Science" ]
2025-10-28T00:00:00
https://arxiv.org/abs/2510.24711
https://arxiv.org/pdf/2510.24711v2
2510.24711
10.48550/arXiv.2510.24711
16
2
true
https://github.com/ali-vilab/ProMoE
arXiv.org
0.5419
931a2a5dd445188983c1e9800df854cd09a854647a660f8c0a878f7d23b3e11c
[ "arxiv", "semantic_scholar" ]
MoE-GS: Mixture of Experts for Dynamic Gaussian Splatting
Recent advances in dynamic scene reconstruction have significantly benefited from 3D Gaussian Splatting, yet existing methods show inconsistent performance across diverse scenes, indicating no single approach effectively handles all dynamic challenges. To overcome these limitations, we propose Mixture of Experts for Dy...
[ "In-Hwan Jin", "Hyeongju Mun", "Joonsoo Kim", "Kugjin Yun", "Kyeongbo Kong" ]
[ "cs.CV" ]
[ "Computer Science" ]
2025-10-22T00:00:00
https://arxiv.org/abs/2510.19210
https://arxiv.org/pdf/2510.19210v2
2510.19210
10.48550/arXiv.2510.19210
1
0
false
null
arXiv.org
0.3438
cdb11ca280f47f85907cd0d9ff2b124ab9f20e26b1a2ea9531caaedfb5fad71f
[ "arxiv", "semantic_scholar" ]
MoE-Prism: Disentangling Monolithic Experts for Elastic MoE Services via Model-System Co-Designs
Mixture-of-Experts (MoE) models, the state-of-the-art in large-scale AI, achieve high quality by sparsely activating parameters. However, their reliance on routing between a few monolithic experts via a top-k mechanism creates a "quality cliff", offering only a few coarse-grained operating points. This inflexibility fo...
[ "Xinfeng Xia", "Jiacheng Liu", "Xiaofeng Hou", "Peng Tang", "Mingxuan Zhang", "Wenfeng Wang", "Chao Li" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2025-10-22T00:00:00
https://arxiv.org/abs/2510.19366
https://arxiv.org/pdf/2510.19366v1
2510.19366
10.48550/arXiv.2510.19366
1
0
false
null
arXiv.org
0.3438
af317171f31a701a31a7feb1cb798a10e005882cdb769546c0298617d98a88ff
[ "arxiv", "semantic_scholar" ]
ReXMoE: Reusing Experts with Minimal Overhead in Mixture-of-Experts
Mixture-of-Experts (MoE) architectures have emerged as a promising approach to scale Large Language Models (LLMs). MoE boosts the efficiency by activating a subset of experts per token. Recent works show that fine-grained experts substantially enriches the combinatorial flexibility of active experts and enhances model ...
[ "Zheyue Tan", "Zhiyuan Li", "Tao Yuan", "Dong Zhou", "Weilin Liu", "Yueqing Zhuang", "Yadong Li", "Guowei Niu", "Cheng Qin", "Zhuyu Yao", "Congyi Liu", "Haiyang Xu", "Boxun Li", "Guohao Dai", "Bo Zhao", "Yu Wang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-10-20T00:00:00
https://arxiv.org/abs/2510.17483
https://arxiv.org/pdf/2510.17483v1
2510.17483
10.48550/arXiv.2510.17483
3
0
false
null
arXiv.org
0.3415
b7a0950ed6e315642fe010f1a5099a99552cd3d4f94bd846b3d093276279a44a
[ "arxiv", "semantic_scholar" ]
L-MoE: End-to-End Training of a Lightweight Mixture of Low-Rank Adaptation Experts
The Mixture of Experts (MoE) architecture enables the scaling of Large Language Models (LLMs) to trillions of parameters by activating a sparse subset of weights for each input, maintaining constant computational cost during inference. Concurrently, Low-Rank Adaptation (LoRA) has emerged as a dominant technique for par...
[ "Shihao Ji", "Zihui Song" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-10-19T00:00:00
https://arxiv.org/abs/2510.17898
https://arxiv.org/pdf/2510.17898v2
2510.17898
10.48550/arXiv.2510.17898
2
0
false
null
arXiv.org
0.3403
480373c7fff3e3edf3a34ad8de55b0c904b647809359a23cfbc572bd745ebd68
[ "arxiv", "semantic_scholar" ]
Leave It to the Experts: Detecting Knowledge Distillation via MoE Expert Signatures
Knowledge Distillation (KD) accelerates training of large language models (LLMs) but poses intellectual property protection and LLM diversity risks. Existing KD detection methods based on self-identity or output similarity can be easily evaded through prompt engineering. We present a KD detection framework effective in...
[ "Pingzhi Li", "Morris Yu-Chao Huang", "Zhen Tan", "Qingquan Song", "Jie Peng", "Kai Zou", "Yu Cheng", "Kaidi Xu", "Tianlong Chen" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2025-10-19T00:00:00
https://arxiv.org/abs/2510.16968
https://arxiv.org/pdf/2510.16968v1
2510.16968
10.48550/arXiv.2510.16968
0
0
true
https://github.com/unites-lab/shadow-moe
arXiv.org
0.5259
e5da4ed5302283dbc91c0c7f4886d9d62b16c257ec299bab8bf7af43993aeda1
[ "arxiv", "semantic_scholar" ]
Input Domain Aware MoE: Decoupling Routing Decisions from Task Optimization in Mixture of Experts
Sparse Mixture of Experts (sMoE) has become a pivotal approach for scaling large vision-language models, offering substantial capacity while maintaining computational efficiency through dynamic, sparse activation of experts. However, existing routing mechanisms, typically based on similarity scoring, struggle to effect...
[ "Yongxiang Hua", "Haoyu Cao", "Zhou Tao", "Bocheng Li", "Zihao Wu", "Chaohu Liu", "Linli Xu" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-10-18T00:00:00
https://arxiv.org/abs/2510.16448
https://arxiv.org/pdf/2510.16448v1
2510.16448
10.1145/3746027.3755754
2
0
false
null
ACM Multimedia
0.3392