AutoPruner: An End-to-End Trainable Filter Pruning Method for Efficient Deep Model Inference

TL;DR

AutoPruner introduces an end-to-end trainable filter pruning method that automatically selects less important filters during training, leading to more efficient deep models with better accuracy than previous methods.

Contribution

It proposes a novel AutoPruner layer that jointly trains filter importance and prunes filters, integrating channel selection into the training process for improved efficiency.

Findings

AutoPruner outperforms state-of-the-art pruning algorithms.

The method maintains higher accuracy after pruning.

Mini-batch pooling and binarization are crucial for success.

Abstract

Channel pruning is an important family of methods to speed up deep model's inference. Previous filter pruning algorithms regard channel pruning and model fine-tuning as two independent steps. This paper argues that combining them into a single end-to-end trainable system will lead to better results. We propose an efficient channel selection layer, namely AutoPruner, to find less important filters automatically in a joint training manner. Our AutoPruner takes previous activation responses as an input and generates a true binary index code for pruning. Hence, all the filters corresponding to zero index values can be removed safely after training. We empirically demonstrate that the gradient information of this channel selection layer is also helpful for the whole model training. By gradually erasing several weak filters, we can prevent an excessive drop in model accuracy. Compared with previous state-of-the-art pruning algorithms (including training from scratch), AutoPruner achieves significantly better performance. Furthermore, ablation experiments show that the proposed novel mini-batch pooling and binarization operations are vital for the success of filter pruning.