The Hong Kong Institute of Science & Innovation, the Chinese Academy of Sciences (HKISI, CAS) and City University of Hong Kong (CityUHK), co-hosted the International Summit of AI for Science was held on 21 November. The inaugural summit brought together over 300 participants, including researchers and industry professionals from around the world at CityUHK to exchange ideas on AI for Science, an emerging interdisciplinary field.

At the summit, Prof Ivor Tsang, Director of A*STAR Centre for Frontier AI Research (A*STAR CFAR), delivered a keynote talk titled “SciPilot: A Preference-Guided Framework for Scientific Discovery”, where he delved into the integration of data curation, representation learning, computational modelling, and expert feedback into a unified iterative workflow.

Abstract:

Scientific discovery in fields such as materials design and RNA 3D structure prediction is often constrained by limited datasets and the high cost of experimental validation. Identifying viable materials or drug candidates can take decades of iterative trial-and-error, requiring a balance of computation, laboratory work, and expert insight. In this talk, we introduce SciPilot, a preference-guided framework that integrates data curation, representation learning, computational modelling, and expert feedback into a unified iterative workflow. The framework organises heterogeneous scientific data and generates structured representations using generative models for in-silico simulation and optimisation. These representations support systematic search to propose candidate materials or molecular structures. Domain experts then evaluate and rank the candidates, providing preference guidance that are encoded and used to refine the search space and modelling strategies. By incorporating expert judgment directly into the computational loop, SciPilot accelerates discovery, reduces unnecessary exploration, and improves interpretability, offering a scalable and adaptive approach beyond purely data-driven or computational methods.

> More on the International Summit of AI for Science.

Photo Credit: CityUHK