Lottery Pools: Winning More by Interpolating Tickets without Increasing Training or Inference Cost

TL;DR

This paper introduces Lottery Pools, a simple interpolation method over subnetworks from lottery ticket pruning, significantly improving their performance without extra training or inference costs across multiple datasets and architectures.

Contribution

We propose Lottery Pools, an interpolation-based ensemble method that enhances sparse subnetworks from lottery ticket pruning without additional training or inference overhead.

Findings

Lottery Pools outperform original lottery tickets on CIFAR-10/100 and ImageNet.

Sparse subnetworks surpass pre-trained dense models in accuracy.

Method improves robustness in out-of-distribution scenarios.

Abstract

Lottery tickets (LTs) is able to discover accurate and sparse subnetworks that could be trained in isolation to match the performance of dense networks. Ensemble, in parallel, is one of the oldest time-proven tricks in machine learning to improve performance by combining the output of multiple independent models. However, the benefits of ensemble in the context of LTs will be diluted since ensemble does not directly lead to stronger sparse subnetworks, but leverages their predictions for a better decision. In this work, we first observe that directly averaging the weights of the adjacent learned subnetworks significantly boosts the performance of LTs. Encouraged by this observation, we further propose an alternative way to perform an 'ensemble' over the subnetworks identified by iterative magnitude pruning via a simple interpolating strategy. We call our method Lottery Pools. In contrast to the naive ensemble which brings no performance gains to each single subnetwork, Lottery Pools yields much stronger sparse subnetworks than the original LTs without requiring any extra training or inference cost. Across various modern architectures on CIFAR-10/100 and ImageNet, we show that our method achieves significant performance gains in both, in-distribution and out-of-distribution scenarios. Impressively, evaluated with VGG-16 and ResNet-18, the produced sparse subnetworks outperform the original LTs by up to 1.88% on CIFAR-100 and 2.36% on CIFAR-100-C; the resulting dense network surpasses the pre-trained dense-model up to 2.22% on CIFAR-100 and 2.38% on CIFAR-100-C.