The Lottery Ticket Hypothesis for Pre-trained BERT Networks

TL;DR

This paper investigates whether sparse, trainable subnetworks exist within pre-trained BERT models, finding such subnetworks at initialization that can transfer across tasks, highlighting the relevance of the lottery ticket hypothesis in NLP.

Contribution

The study demonstrates the existence of trainable, transferable subnetworks in pre-trained BERT models at initialization, extending the lottery ticket hypothesis to large-scale NLP models.

Findings

Matching subnetworks exist at 40-90% sparsity in BERT.

Subnetworks on the pre-training task transfer universally.

Subnetworks on other tasks transfer limited or not at all.

Abstract

In natural language processing (NLP), enormous pre-trained models like BERT have become the standard starting point for training on a range of downstream tasks, and similar trends are emerging in other areas of deep learning. In parallel, work on the lottery ticket hypothesis has shown that models for NLP and computer vision contain smaller matching subnetworks capable of training in isolation to full accuracy and transferring to other tasks. In this work, we combine these observations to assess whether such trainable, transferrable subnetworks exist in pre-trained BERT models. For a range of downstream tasks, we indeed find matching subnetworks at 40% to 90% sparsity. We find these subnetworks at (pre-trained) initialization, a deviation from prior NLP research where they emerge only after some amount of training. Subnetworks found on the masked language modeling task (the same task used to pre-train the model) transfer universally; those found on other tasks transfer in a limited fashion if at all. As large-scale pre-training becomes an increasingly central paradigm in deep learning, our results demonstrate that the main lottery ticket observations remain relevant in this context. Codes available at https://github.com/VITA-Group/BERT-Tickets.