edgeml-dqn
edgeml-dqn is a Python implementation of adaptive edge offloading for image classification under rate limits. Developed under the paper "Adaptive Edge Offloading for Image Classification Under Rate Limit" from EMSOFT 2022, it extends prior work on real-time edge classification with token bucket constraints. The system uses a Deep Q-Network to learn optimal offloading policies between a weak on-device classifier and a stronger edge or cloud classifier, balancing inference accuracy with communication costs subject to token bucket rate constraints. Pre-computed offloading metrics for the ILSVRC validation set using VGG-style 16-layer and OFA 595MFlops classifier pairs are included. The repository provides scripts for training the DQN model, running simulations against baseline and lower bound policies, and Jupyter notebooks for visualizing results. Requirements include Python 3.7+ with Anaconda, numpy, TensorFlow 2, and matplotlib. The library supports configurable token bucket parameters and sequence types for