Contrastive adaptation of retrieval representations with an anchor geometry constraint
Received: 25 Oct 2026 Revised: 02 Nov 2026 Accepted: 08 Nov 2026
Published: 2026, vol. 30, issue 4, pp. 421–424
Abstract
AnchorBank corrects contrastive adaptation using relations to frozen anchors. We describe its controlled subspace and local projection properties and compare it with InfoNCE on four collections.
Keywords: information retrieval, contrastive learning, KL divergence, gradient projection.
BibTeX
@article{IS-Tarlinsky-Glavatsky2026,
author = {Tarlinsky, Igor Viktorovich and Glavatsky, Sergei Timofeevich},
title = {{Contrastive adaptation of retrieval representations with an anchor geometry constraint}},
journal = {Intelligent Systems. Theory and Applications},
year = {2026},
volume = {30},
number = {4},
pages = {421--424},
}
AMSBIB
\Bibitem{IS-Tarlinsky-Glavatsky2026}
\by I.\,V.~Tarlinsky, S.\,T.~Glavatsky
\paper Contrastive adaptation of retrieval representations with an anchor geometry constraint
\jour Intelligent Systems. Theory and Applications
\yr 2026
\vol 30
\issue 4
\pages 421--424
\lang In Russian
Published under
Creative Commons Attribution 4.0 International (CC BY 4.0)
RU