Inducing Semi-Structured Sparsity by Masking for Efficient Model Inference in Convolutional Networks
This provides a practical acceleration technique for convolutional networks in vision and foundation models, with incremental improvements to existing sparsity methods.
The paper tackles the problem of accelerating convolutional models for efficient inference by proposing a method to learn semi-structured sparsity patterns through masking, achieving more than two-fold speedup without performance loss while keeping the original model weights unchanged for easy updates.
The crucial role of convolutional models, both as standalone vision models and backbones in foundation models, necessitates effective acceleration techniques. This paper proposes a novel method to learn semi-structured sparsity patterns for convolution kernels in the form of maskings enabling the utilization of readily available hardware accelerations. The approach accelerates convolutional models more than two-fold during inference without decreasing model performance. At the same time, the original model weights and structure remain unchanged keeping the model thus easily updatable. Beyond the immediate practical use, the effect of maskings on prediction is easily quantifiable. Therefore, guarantees on model predictions under maskings are derived showing stability bounds for learned maskings even after updating the original underlying model.