Multi-task Pruning via Filter Index Sharing: A Many-Objective Optimization Approach

Abstract State-of-the-art deep neural network plays an increasingly important role in artificial intelligence, while the huge number of parameters networks brings high memory cost and computational complexity. To solve this problem, filter pruning is widely used for compression acceleration. However, existing algorithms focus mainly on single model, few results are available to multi-task that capable multi-model promoting learning performance. By utilizing sharing technique, paper aimed establish a framework simultaneously merging filters networks. An optimization problem selecting solved by developing many-objective algorithm where three criteria adopted as objectives problem. With purpose keeping structure, index matrix introduced regulate information during training. The proposed quite flexible can be performed with either adaptive or pre-specified rates. Extensive experiments verify applicability superiority method both single-task pruning.