XLab
News from XLab.
XLab’s core mission: Micro-to-Macro Mechanistic AI — building AI systems that connect micro-scale mechanisms to macro-scale behaviors and outcomes, from parameter dynamics in foundation models to physicochemical mechanisms in molecular and biological systems. In fundamental AI, the lab develops parameter-efficient fine-tuning (PEFT) and multimodal foundation models that connect micro-scale parameter dynamics with emergent model behaviors. In AI4Science, the lab combines physicochemical mechanisms with 3D geometric deep learning to reveal molecular-level principles governing biological function, enabling more interpretable molecular discovery and precision medicine.
News
| Jul 31, 2026 | New paper published in Journal of Cheminformatics: How Well Can Off-the-Shelf LLMs Elucidate Molecular Structures from Mass Spectra Using Chain-of-Thought Reasoning? |
|---|---|
| Jun 24, 2026 | New preprint: Staying VIGILant: Mitigating Visual Laziness via Counterfactual Visual Alignment in MLLMs. Accepted to ECCV 2026. |
| Jun 09, 2026 | New preprint on bioRxiv: ExplainBind: Explainable Physicochemical Determinants of Protein-Ligand Binding via Non-Covalent Interactions |
| May 02, 2026 | New paper published in Genome Biology: EvoRMD: Integrating Biological Context and Evolutionary RNA Language Models for Interpretable Prediction of RNA Modifications. |
| Apr 04, 2026 | New preprint: Not All Directions Matter: Toward Structured and Task-Aware Low-Rank Adaptation. Paper accepted to ACL 2026. |
| Oct 03, 2025 | New preprint: HyperAdaLoRA: Accelerating LoRA Rank Allocation During Training via Hypernetworks without Sacrificing Performance, accepted to ACL Findings 2026. |