Scalable Attentive Sentence-Pair Modeling via Distilled Sentence Embedding

TL;DR

This paper introduces Distilled Sentence Embedding (DSE), a fast and effective sentence-pair modeling method that uses knowledge distillation from cross-attentive models to produce high-quality sentence embeddings for large-scale natural language understanding tasks.

Contribution

The paper presents DSE, a novel knowledge distillation approach that creates efficient sentence embeddings from cross-attentive models, enabling scalable sentence-pair similarity computations.

Findings

DSE outperforms several sentence embedding methods on GLUE tasks.

DSE accelerates similarity computation by several orders of magnitude.

DSE achieves state-of-the-art results on universal sentence representation benchmarks.

Abstract

Recent state-of-the-art natural language understanding models, such as BERT and XLNet, score a pair of sentences (A and B) using multiple cross-attention operations - a process in which each word in sentence A attends to all words in sentence B and vice versa. As a result, computing the similarity between a query sentence and a set of candidate sentences, requires the propagation of all query-candidate sentence-pairs throughout a stack of cross-attention layers. This exhaustive process becomes computationally prohibitive when the number of candidate sentences is large. In contrast, sentence embedding techniques learn a sentence-to-vector mapping and compute the similarity between the sentence vectors via simple elementary operations. In this paper, we introduce Distilled Sentence Embedding (DSE) - a model that is based on knowledge distillation from cross-attentive models, focusing on sentence-pair tasks. The outline of DSE is as follows: Given a cross-attentive teacher model (e.g. a fine-tuned BERT), we train a sentence embedding based student model to reconstruct the sentence-pair scores obtained by the teacher model. We empirically demonstrate the effectiveness of DSE on five GLUE sentence-pair tasks. DSE significantly outperforms several ELMO variants and other sentence embedding methods, while accelerating computation of the query-candidate sentence-pairs similarities by several orders of magnitude, with an average relative degradation of 4.6% compared to BERT. Furthermore, we show that DSE produces sentence embeddings that reach state-of-the-art performance on universal sentence representation benchmarks. Our code is made publicly available at https://github.com/microsoft/Distilled-Sentence-Embedding.