A Novel Architecture Slimming Method for Network Pruning and Knowledge Distillation

TL;DR

This paper introduces an automated architecture slimming method that optimizes layer configurations for network pruning and knowledge distillation, reducing manual effort and improving performance at similar compression rates.

Contribution

It formulates architecture determination as a PCA-based linear transformation, enabling automatic, effective layer-wise compression without extensive experimentation.

Findings

Significant performance gains over baselines at the same compression rate

Layer-wise compression rates align with known layer sensitivities

Method reduces human intervention in model compression processes

Abstract

Network pruning and knowledge distillation are two widely-known model compression methods that efficiently reduce computation cost and model size. A common problem in both pruning and distillation is to determine compressed architecture, i.e., the exact number of filters per layer and layer configuration, in order to preserve most of the original model capacity. In spite of the great advances in existing works, the determination of an excellent architecture still requires human interference or tremendous experimentations. In this paper, we propose an architecture slimming method that automates the layer configuration process. We start from the perspective that the capacity of the over-parameterized model can be largely preserved by finding the minimum number of filters preserving the maximum parameter variance per layer, resulting in a thin architecture. We formulate the determination of compressed architecture as a one-step orthogonal linear transformation, and integrate principle component analysis (PCA), where the variances of filters in the first several projections are maximized. We demonstrate the rationality of our analysis and the effectiveness of the proposed method through extensive experiments. In particular, we show that under the same overall compression rate, the compressed architecture determined by our method shows significant performance gain over baselines after pruning and distillation. Surprisingly, we find that the resulting layer-wise compression rates correspond to the layer sensitivities found by existing works through tremendous experimentations.