Emergent Properties of Finetuned Language Representation Models
This provides insights into model interpretability and efficiency for NLP researchers, though it is incremental as it builds on existing analysis of over-parameterization.
The paper tackled the problem of understanding why large language models like BERT work well by analyzing the redundancy and information location in their output vectors, showing that the [CLS] embedding is highly redundant and can be compressed with minimal accuracy loss, especially for finetuned models, and that specific output dimensions alone yield competitive results.
Large, self-supervised transformer-based language representation models have recently received significant amounts of attention, and have produced state-of-the-art results across a variety of tasks simply by scaling up pre-training on larger and larger corpora. Such models usually produce high dimensional vectors, on top of which additional task-specific layers and architectural modifications are added to adapt them to specific downstream tasks. Though there exists ample evidence that such models work well, we aim to understand what happens when they work well. We analyze the redundancy and location of information contained in output vectors for one such language representation model -- BERT. We show empirical evidence that the [CLS] embedding in BERT contains highly redundant information, and can be compressed with minimal loss of accuracy, especially for finetuned models, dovetailing into open threads in the field about the role of over-parameterization in learning. We also shed light on the existence of specific output dimensions which alone give very competitive results when compared to using all dimensions of output vectors.