Reproduce “Beyond Relevance: Bayesian Evidence Acquisition for Agentic Whole-Slide Image Reasoning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Bryan Wong, Xun Xu, Huazhu Fu, Nancy F. Chen, Mun Yong Yi

The BEACON framework introduces a Bayesian approach to whole-slide image (WSI) reasoning, acquiring diagnostically informative patches by maximizing expected information gain to efficiently reduce uncertainty. It improves zero-shot performance by up to 4.87% over prior agentic methods and reduces computational costs while operating as a training-free, plug-and-play solution.

Key insights

  • Semantic relevance is often an insufficient criterion for selecting diagnostically informative evidence; EIG provides a more effective measure by targeting uncertainty reduction.
  • Bayesian evidence acquisition consistently outperforms both random sampling and relevance-based retrieval strategies under identical acquisition budgets.
  • The framework's effectiveness is attributable to its core Bayesian strategy rather than specific model architectures, demonstrating broad compatibility across different VLM and MLLM backbones.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.