Belief-Aware VLM Model for Human-like Reasoning

Anshul Nayak*, Shahil Shaik*, Yue Wang · 2026

Read Belief-Aware VLM Model for Human-like Reasoning by Anshul Nayak*, Shahil Shaik*, Yue Wang (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-2604-09686.txt.

Description

Nayak, Shaik, and Wang give vision-language models a way to track belief for human intent inference. Traditional intent models lean on visible states and generalize poorly across tasks and changing environments. Current vision-language systems lack ways to track belief over long horizons. Their framework approximates belief with a vector memory that retrieves relevant multimodal context for reasoning. Decisions are refined with a reinforcement-learning policy over the model's latent space. Tests on public question-answering sets such as HD-EPIC show consistent gains over zero-shot baselines.

Catalog record

CREATORAnshul Nayak*, Shahil Shaik*, Yue Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26130
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2604.09686

Browse the full library · About the Tower of Babel