Compressing Sentence Representation with maximum Coding Rate Reduction
This addresses the need for efficient, space-saving models in natural language inference, though it is incremental as it builds on existing distillation and MCR2 methods.
The paper tackles the performance gap between large and small language models for sentence representation by augmenting a distilled Sentence-BERT model with a projection layer learned on the Maximum Coding Rate Reduction objective, achieving comparable results on semantic retrieval benchmarks.
In most natural language inference problems, sentence representation is needed for semantic retrieval tasks. In recent years, pre-trained large language models have been quite effective for computing such representations. These models produce high-dimensional sentence embeddings. An evident performance gap between large and small models exists in practice. Hence, due to space and time hardware limitations, there is a need to attain comparable results when using the smaller model, which is usually a distilled version of the large language model. In this paper, we assess the model distillation of the sentence representation model Sentence-BERT by augmenting the pre-trained distilled model with a projection layer additionally learned on the Maximum Coding Rate Reduction (MCR2)objective, a novel approach developed for general-purpose manifold clustering. We demonstrate that the new language model with reduced complexity and sentence embedding size can achieve comparable results on semantic retrieval benchmarks.