UPSCALE: Unconstrained Channel Pruning

TL;DR

UPSCALE introduces a novel channel reordering method that reduces inference latency and improves accuracy in pruned neural networks by removing constraints and optimizing channel order at export time.

Contribution

It proposes a generic algorithm to prune models without constraints, enhancing accuracy and speed, applicable to any pruning pattern.

Findings

Increases ImageNet accuracy by 2.1 points on average

Improves inference speed by up to 2x

Beneficial across multiple architectures like DenseNet, EfficientNetV2, ResNet

Abstract

As neural networks grow in size and complexity, inference speeds decline. To combat this, one of the most effective compression techniques -- channel pruning -- removes channels from weights. However, for multi-branch segments of a model, channel removal can introduce inference-time memory copies. In turn, these copies increase inference latency -- so much so that the pruned model can be slower than the unpruned model. As a workaround, pruners conventionally constrain certain channels to be pruned together. This fully eliminates memory copies but, as we show, significantly impairs accuracy. We now have a dilemma: Remove constraints but increase latency, or add constraints and impair accuracy. In response, our insight is to reorder channels at export time, (1) reducing latency by reducing memory copies and (2) improving accuracy by removing constraints. Using this insight, we design a generic algorithm UPSCALE to prune models with any pruning pattern. By removing constraints from existing pruners, we improve ImageNet accuracy for post-training pruned models by 2.1 points on average -- benefiting DenseNet (+16.9), EfficientNetV2 (+7.9), and ResNet (+6.2). Furthermore, by reordering channels, UPSCALE improves inference speeds by up to 2x over a baseline export.