RL-Guided Pruning of CNNs Using Graph Embeddings

This paper presents a novel method for compressing Convolutional Neural Networks (CNNs) to enable efficient deployment on low-capacity devices. The proposed approach combines neural network pruning with reinforcement learning (RL) and graph embedding. Each network is represented as a computational graph, and Graph Convolutional Networks (GCNs) are utilized to learn graph-level embeddings that inform pruning decisions. By applying Proximal Policy Optimization (PPO), we automate the selection of layer-wise pruning ratios, eliminating the need for manual tuning. Experiments on ResNet-34 and VGG-19, trained on the CIFAR-10 dataset, demonstrate that our method achieves up to 80% compression while maintaining or improving model accuracy through post-pruning rewinding. We evaluated both structured and unstructured pruning strategies, analyzing the trade-offs in accuracy, FLOPs, parameter count, and inference time