Evaluating Unsupervised Dimensionality Reduction Methods for Pretrained Sentence Embeddings

TL;DR

This paper evaluates unsupervised dimensionality reduction techniques, especially PCA, on pretrained sentence embeddings, showing significant size reduction with minimal or even improved downstream task performance.

Contribution

It demonstrates that simple unsupervised methods like PCA effectively reduce embedding dimensions by half without performance loss, and can sometimes enhance task results.

Findings

PCA reduces embedding size by nearly 50%

Dimensionality reduction maintains or improves downstream performance

Simple methods outperform complex ones in this context

Abstract

Sentence embeddings produced by Pretrained Language Models (PLMs) have received wide attention from the NLP community due to their superior performance when representing texts in numerous downstream applications. However, the high dimensionality of the sentence embeddings produced by PLMs is problematic when representing large numbers of sentences in memory- or compute-constrained devices. As a solution, we evaluate unsupervised dimensionality reduction methods to reduce the dimensionality of sentence embeddings produced by PLMs. Our experimental results show that simple methods such as Principal Component Analysis (PCA) can reduce the dimensionality of sentence embeddings by almost $50\%$, without incurring a significant loss in performance in multiple downstream tasks. Surprisingly, reducing the dimensionality further improves performance over the original high-dimensional versions for the sentence embeddings produced by some PLMs in some tasks.