English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Dense retrieval overcome the lexical gap and has shown great success inad-hoc information retrieval (IR). Despite their success, dense retrievers areexpensive to serve across practical use cases. For use cases requiring tosearch from millions of documents, the dense index becomes bulky and requireshigh memory usage for storing the index. More recently, learning-to-hash (LTH)techniques, for e.g., BPR and JPQ, produce binary document vectors, therebyreducing the memory requirement to efficiently store the dense index. LTHtechniques are supervised and finetune the retriever using a ranking loss. Theyoutperform their counterparts, i.e., traditional out-of-the-box vectorcompression techniques such as PCA or PQ. A missing piece from prior work isthat existing techniques have been evaluated only in-domain, i.e., on a singledataset such as MS MARCO. In our work, we evaluate LTH and vector compressiontechniques for improving the downstream zero-shot retrieval accuracy of theTAS-B dense retriever while maintaining efficiency at inference. Our resultsdemonstrate that, unlike prior work, LTH strategies when applied naively canunderperform the zero-shot TAS-B dense retriever on average by up to 14%nDCG@10 on the BEIR benchmark. To solve this limitation, in our work, wepropose an easy yet effective solution of injecting domain adaptation withexisting supervised LTH techniques. We experiment with two well-knownunsupervised domain adaptation techniques: GenQ and GPL. Our domain adaptationinjection technique can improve the downstream zero-shot retrievaleffectiveness for both BPR and JPQ variants of the TAS-B model by on average11.5% and 8.2% nDCG@10 while both maintaining 32$\times$ memory efficiency and14$\times$ and 2$\times$ speedup respectively in CPU retrieval latency on BEIR.All our code, models, and data are publicly available athttps://github.com/thakur-nandan/income.

Injecting Domain Adaptation with Learning-to-hash for Effective and Efficient Zero-shot Dense Retrieval