Robust Lottery Tickets for Pre-trained Language Models

TL;DR

This paper introduces a method to find robust lottery tickets in pre-trained language models by learning binary masks with adversarial training, significantly improving robustness without sacrificing accuracy.

Contribution

It proposes a novel binary mask learning approach with adversarial loss to identify robust subnetworks in PLMs, enhancing adversarial robustness.

Findings

Significant robustness improvement over previous methods

Effective binary mask learning with L0 regularization

Maintains high accuracy while improving robustness

Abstract

Recent works on Lottery Ticket Hypothesis have shown that pre-trained language models (PLMs) contain smaller matching subnetworks(winning tickets) which are capable of reaching accuracy comparable to the original models. However, these tickets are proved to be notrobust to adversarial examples, and even worse than their PLM counterparts. To address this problem, we propose a novel method based on learning binary weight masks to identify robust tickets hidden in the original PLMs. Since the loss is not differentiable for the binary mask, we assign the hard concrete distribution to the masks and encourage their sparsity using a smoothing approximation of L0 regularization.Furthermore, we design an adversarial loss objective to guide the search for robust tickets and ensure that the tickets perform well bothin accuracy and robustness. Experimental results show the significant improvement of the proposed method over previous work on adversarial robustness evaluation.