Upload README.md with huggingface_hub

0ee136a verified 3 months ago

3.39 kB

language: en
license: apache-2.0
tags:
  - image-classification
  - green-ai
  - energy-efficiency
  - computer-vision
  - mobilevitv3
  - eden-framework
  - e2am
  - sustainable-ai
datasets:
  - imagenet
metrics:
  - accuracy
co2_eq_emissions:
  emissions: 0.9111
  unit: kg
  source: Estimated via CodeCarbon (grid factor 0.475 kg CO2e/kWh)
  hardware_used: NVIDIA GeForce GTX 1080 Ti
dataset_info:
  dataset_size: ~450,000 images – 300 classes (224 px)

EDEN-MobileViTv3-Custom-ImageNet300 — SOTA Optimized

Primary KPI: EAG (Energy-to-Accuracy Gradient) = 1.9376e-10 ΔAcc/ΔJoules

Abstract

This model is part of Project EDEN (Energy-Driven Evolution of Networks), implementing the E2AM (Energy Efficient Advanced Model) Framework. The goal is to shift AI benchmarking from pure accuracy to Green SOTA — maximizing predictive power per Joule consumed.

Applied Technique: Phase 2 – Progressive Unfreezing + AMP (E2AM SOTA)

Profiling Environment

Component	Specification
GPU	NVIDIA GeForce GTX 1080 Ti (11 GB VRAM, 250 W TDP)
CPU	Intel Xeon W-2125 (4 cores / 8 threads @ 4.00 GHz)
RAM	63.66 GB System RAM
OS	Windows 10
Dataset	Custom-ImageNet300 — ~450,000 images – 300 classes (224 px)

🟢 Green Delta Table

Comparing this model against the reference baseline (ResNet-50 equivalent)

Metric	ResNet50 Baseline	MobileViTv3 (EDEN)	Δ
Accuracy	0.9573	0.8850	`-7.24%`
Total Energy (J)	380,392,115	6,905,436	`98.18% saved`
CO₂ Emissions (kg)	50.1906	0.9111	—
EAG Score	—	1.9376e-10	ΔAcc/ΔJoules

A positive EAG means this model learns more per Joule than the baseline. A negative EAG indicates a trade-off where higher accuracy required more energy investment.

E2AM Algorithm — Applied Phases

Phase 1 – Zero-Overhead Initialization: Dataset pre-loaded into pinned System RAM to eliminate disk I/O power spikes.

Phase 2 – Progressive Unfreezing: Backbone frozen for the first E_unfreeze epochs (only the classification head trains). At E_unfreeze, all layers are unfrozen and the learning rate is decayed. Gradient accumulation over N micro-batches simulates large batch sizes without proportional VRAM cost, slashing power-draw spikes.

AMP (Automated Mixed Precision): torch.cuda.amp.autocast() halves GPU memory bandwidth, reducing energy per backward pass.

Sparse Regularisation: L1 penalty λ·Σ|W| applied to trainable weights, driving dead neurons to zero and enabling future pruning.

Training Statistics

Metric	Value
Final Accuracy	0.8850 (88.50%)
Total Energy Consumed	6,905,436 J (1.9182 kWh)
Training Time	10,333 s (2.87 hrs)
Estimated CO₂	0.9111 kg CO₂e
Training Log	`test1\eden_optimized_custom_imagenet_mobilevitv3.csv`

Cite This Research

If you use this model, please cite the EDEN / E2AM Framework:

@misc{eden2025,
  title     = {Project EDEN: Energy-Driven Evolution of Networks},
  author    = {EDEN Research Team},
  year      = {2025},
  note      = {Hugging Face Organization: ProjectEDEN},
  url       = {https://huggingface.co/Shanmuk4622}
}