Unveiling Key Aspects of Fine-Tuning in Sentence Embeddings: A Representation Rank Analysis

TL;DR

This paper analyzes the effects of fine-tuning on sentence embeddings using representation rank, revealing key dynamics and proposing a rank reduction strategy to improve performance and stability.

Contribution

It introduces a novel analysis of fine-tuning phases via representation rank and proposes a rank reduction method to enhance sentence embedding training.

Findings

Representation rank peaks define distinct fine-tuning phases.

Dynamics of alignment, uniformity, and linguistic abilities change across phases.

Rank reduction improves performance and stability of state-of-the-art methods.

Abstract

The latest advancements in unsupervised learning of sentence embeddings predominantly involve employing contrastive learning-based (CL-based) fine-tuning over pre-trained language models. In this study, we analyze the latest sentence embedding methods by adopting representation rank as the primary tool of analysis. We first define Phase 1 and Phase 2 of fine-tuning based on when representation rank peaks. Utilizing these phases, we conduct a thorough analysis and obtain essential findings across key aspects, including alignment and uniformity, linguistic abilities, and correlation between performance and rank. For instance, we find that the dynamics of the key aspects can undergo significant changes as fine-tuning transitions from Phase 1 to Phase 2. Based on these findings, we experiment with a rank reduction (RR) strategy that facilitates rapid and stable fine-tuning of the latest CL-based methods. Through empirical investigations, we showcase the efficacy of RR in enhancing the performance and stability of five state-of-the-art sentence embedding methods.