Composite Code Sparse Autoencoders for first stage retrieval
This work addresses the need for scalable and memory-efficient retrieval systems in information retrieval, representing an incremental improvement over existing quantization and indexing techniques.
The paper tackles the problem of efficient similarity search for first-stage retrieval in information retrieval by proposing Composite Code Sparse Autoencoders (CCSA), which outperforms IVF with product quantization on the MSMARCO dataset and improves index size and memory usage for graph-based methods while maintaining recall and MRR.
We propose a Composite Code Sparse Autoencoder (CCSA) approach for Approximate Nearest Neighbor (ANN) search of document representations based on Siamese-BERT models. In Information Retrieval (IR), the ranking pipeline is generally decomposed in two stages: the first stage focus on retrieving a candidate set from the whole collection. The second stage re-ranks the candidate set by relying on more complex models. Recently, Siamese-BERT models have been used as first stage ranker to replace or complement the traditional bag-of-word models. However, indexing and searching a large document collection require efficient similarity search on dense vectors and this is why ANN techniques come into play. Since composite codes are naturally sparse, we first show how CCSA can learn efficient parallel inverted index thanks to an uniformity regularizer. Second, CCSA can be used as a binary quantization method and we propose to combine it with the recent graph based ANN techniques. Our experiments on MSMARCO dataset reveal that CCSA outperforms IVF with product quantization. Furthermore, CCSA binary quantization is beneficial for the index size, and memory usage for the graph-based HNSW method, while maintaining a good level of recall and MRR. Third, we compare with recent supervised quantization methods for image retrieval and find that CCSA is able to outperform them.