IMRNNs: An Efficient Method for Interpretable Dense Retrieval via Embedding Modulation

Yash Saxena, Ankur Padia, Kalpa Gunaratna, Manas Gaur · 2026

Read IMRNNs: An Efficient Method for Interpretable Dense Retrieval via Embedding Modulation by Yash Saxena, Ankur Padia, Kalpa Gunaratna, Manas Gaur (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-20084.txt.

Description

Yash Saxena, Ankur Padia, Kalpa Gunaratna, and Manas Gaur make dense retrieval systems explain themselves. Dense retrievers embed queries and documents as vectors and match them by similarity. They power retrieval-augmented generation, yet their matching is opaque. How a query and document interact semantically stays hidden inside static embeddings.

Their method, IMRNNs, modulates embeddings so the interaction becomes interpretable, while staying efficient. Static embeddings cannot show which parts of a query matched which parts of a document. Modulation lets the two sides adapt to each other visibly. Understanding retrieval behaviour helps diagnose failures and improve design. The work targets a central trust problem in search and question answering. The flaw is not small. It hides why a match was made.

Catalog record

CREATORYash Saxena, Ankur Padia, Kalpa Gunaratna, Manas Gaur
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-20084

Browse the full library · About the Tower of Babel