Text Representation Distillation via Information Bottleneck Principle

Yanzhao Zhang, Dingkun Long, Zehan Li, Pengjun Xie

Submission Type: Regular Long Paper

Submission Track: Information Retrieval and Text Mining

Submission Track 2: Efficient Methods for NLP

Keywords: knowledge distillation, text representation, text retrieval, language model

TL;DR: A novel Knowledge Distillation method inspired by the Information Bottleneck is proposed to reduce the performance loss during the distillation process of large language models into smaller text representation models.

Abstract: Pre-trained language models (PLMs) have recently shown great success in text representation field. However, the high computational cost and high-dimensional representation of PLMs pose significant challenges for practical applications. To make models more accessible, an effective method is to distill large models into smaller representation models. In order to relieve the issue of performance degradation after distillation, we propose a novel Knowledge Distillation method called \textbf{IBKD}. This approach is motivated by the Information Bottleneck principle and aims to maximize the mutual information between the final representation of the teacher and student model, while simultaneously reducing the mutual information between the student model's representation and the input data. This enables the student model to preserve important learned information while avoiding unnecessary information, thus reducing the risk of over-fitting. Empirical studies on two main downstream applications of text representation (Semantic Textual Similarity and Dense Retrieval tasks) demonstrate the effectiveness of our proposed approach.

Submission Number: 2676