Activation Map Compression through Tensor Decomposition for Deep Learning
This addresses the challenge of enabling efficient on-device training for resource-constrained embedded devices, representing an incremental improvement in activation compression techniques.
The paper tackles the memory bottleneck of storing activation maps during backpropagation in Edge AI by compressing them using tensor decomposition methods like SVD and HOSVD, achieving significant memory savings while maintaining learning performance and Pareto-superiority over state-of-the-art solutions.
Internet of Things and Deep Learning are synergetically and exponentially growing industrial fields with a massive call for their unification into a common framework called Edge AI. While on-device inference is a well-explored topic in recent research, backpropagation remains an open challenge due to its prohibitive computational and memory costs compared to the extreme resource constraints of embedded devices. Drawing on tensor decomposition research, we tackle the main bottleneck of backpropagation, namely the memory footprint of activation map storage. We investigate and compare the effects of activation compression using Singular Value Decomposition and its tensor variant, High-Order Singular Value Decomposition. The application of low-order decomposition results in considerable memory savings while preserving the features essential for learning, and also offers theoretical guarantees to convergence. Experimental results obtained on main-stream architectures and tasks demonstrate Pareto-superiority over other state-of-the-art solutions, in terms of the trade-off between generalization and memory footprint.