ICLR 2026
Reforming the Mechanism: Editing Reasoning Patterns in LLMs with Circuit Reshaping
Fixes one specific reasoning mistake in an LLM without retraining the whole model or breaking its other skills.
ISBI 2026
BrainTAP: Brain Disorder Prediction with Adaptive Distill and Selective Prior Integration
Predicts brain disorders by teaching a model to fuse two very different kinds of brain-scan data.
WSDM 2026
MolEdit: Knowledge Editing for Multimodal Molecule Language Models
Updates a single outdated chemistry fact inside a molecule-language model without retraining it.
For the complete publication list, see my CV.