How amazingdd/ecc-snn works
Project type and purpose
• ECC-SNN is a research-oriented machine-learning pipeline implementing cost-effective edge-cloud collaboration for spiking neural networks.
• It provides Python scripts to prepare data, train and update edge/cloud models, and evaluate performance under different offloading thresholds.
File structure
• Top-level Python entry points:
• prepare.py (data & model initialization)
• update.py (task-by-task model updates)
• evaluation.py (performance measurement with thresholding)
• models/ directory: definitions of base networks (ResNet, VGG, ViT) and their spiking variants.
• scripts/ directory: utilities for I/O (imagenet_io.py), latency measurement, model inference, and a lightweight service wrapper.
• utils.py: shared helper functions.
README for additional insights
• The workflow is organized in two stages: Setup (prepare.py) and Running (update.py → evaluation.py).
• Edge-cloud collaboration is controlled by parameters (–cloud, –edge, –thr).
• Pretraining, distillation, and patience hyperparameters govern the training pipeline.
Elements
• Clear labels for each component (e.g., “Edge Spiking NN”, “Cloud ViT Service”)
• Dataset → Trainer
• Trainer → Model Checkpoints
• Edge Inference → Decision → [Local/Cloud] → Result
• Blue rectangles for compute modules (prepare/update/eval)
• Orange parallelograms for decision points