We have two papers accepted at NeurIPS 2026!

I am happy to share that COCOCO and VH-CBM have been accepted at NeurIPS 2026!

COCOCO: Concise and Logically Consistent Conformal Sets for Neuro-Symbolic Concept-Based Models

How can we express uncertainty about concepts and predictions while respecting logical constraints? COCOCO combines conformal prediction with neuro-symbolic reasoning to produce concise, logically consistent prediction sets with coverage guarantees.

Read the paper.
symbols concepts uncertainty

VH-CBM: Concepts Worth Having: Refining VLM-Guided Concept Bottleneck Models with Minimal Annotations

VH-CBM combines vision-language models with a small amount of human supervision to learn more accurate, better-calibrated concepts. A Gaussian process propagates expert annotations through the VLM’s embedding space, improving concept quality even with very limited annotations.

Read the paper.
concepts representations uncertainty

Many thanks to all my co-authors! You can find these and my other papers on the research page.