TabQL: In-Context Q-Learning with Tabular Foundation Models

Qisai Liu, Zhanhong Jiang, Timilehin Ayanlade, Ashutosh Kumar Nirala, Yang Li, Aditya Balu, Soumik Sarkar · 2026

Read TabQL: In-Context Q-Learning with Tabular Foundation Models by Qisai Liu, Zhanhong Jiang, Timilehin Ayanlade, Ashutosh Kumar Nirala, Yang Li, Aditya Balu, Soumik Sarkar (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-2605-18979.txt.

Description

Qisai Liu and six colleagues replace the Q-network of deep Q-learning with a tabular foundation model that learns in context. TabQL represents Q-values through a sequence-to-sequence model over tabulated state-action-value triples, adapting rapidly from limited interaction. Zero- or few-shot inference comes from in-context updates, while warm-up with standard deep Q-learning bootstraps high-quality context. Theory shows the method interpolating between vanilla Q-learning and deep Q-learning, with Bellman updates amortized through in-context learning, and benchmarks confirm it.

Catalog record

CREATORQisai Liu, Zhanhong Jiang, Timilehin Ayanlade, Ashutosh Kumar Nirala, Yang Li, Aditya Balu, Soumik Sarkar
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE71446
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.18979

Browse the full library · About the Tower of Babel