← smac.pub home

The Matryoshka Hypencoder

pdf bibtex short conference paper

Authors: Majd Alkawaas, Sean MacAvaney

Appeared in: 49th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2026)

Links/IDs:
DOI 10.1145/3805712.3809980 DBLP conf/sigir/AlkawaasM26 arXiv 2607.17457 Google Scholar 7wWfoDgAAAAJ:VOx2b1Wkg3QC Enlighten 383305 smac.pub sigir2026-hypencoder

Abstract:

The Hypencoder is a recently-proposed retrieval approach that encodes queries as shallow neural networks ("Q-Nets") that estimate relevance over pre-computed document embeddings. Inspired by Matryoshka Representation Learning, we show that the Hypencoder can be extended to support multiple sizes of Q-Nets, allowing trade-offs between effectiveness and efficiency when deployed. We find that this "Matryoshka Hypencoder" achieves comparable in-domain effectiveness with approximately 7x fewer active parameters in-domain and half as many active parameters out-of-domain, which corresponds to a 1.6-3.4x increase in scoring throughput. This work paves the way for practical deployment of Hypencoders.

BibTeX @inproceedings{alkawaas:sigir2026-hypencoder, author = {Alkawaas, Majd and MacAvaney, Sean}, title = {The Matryoshka Hypencoder}, booktitle = {49th International ACM SIGIR Conference on Research and Development in Information Retrieval}, year = {2026}, url = {https://arxiv.org/abs/2607.17457}, doi = {10.1145/3805712.3809980} }