Large Discovery Models: Towards Autonomous Scientific Discovery

Date	24 Sep 2026
Time	10:00 am - 11:00 am (HKT)
Venue	Tam Wing Fan Innovation Wing Two, G/F Run Run Shaw Building
Speaker	Prof. Jun Wang
Institution	University College London

Self Photos / Files - Prof. Jun Wang Seminar Poster

Title:

Schedule:

Date: 24^th September, 2026 (Thursday)

Time: 10 - 11 am (HKT)

Venue: Tam Wing Fan Innovation Wing Two, G/F Run Run Shaw Building

Speaker:

Prof. Jun Wang

University College London

Biography:

Professor Jun Wang is an internationally recognised AI scientist at UCL whose work connects foundational advances in machine learning and intelligent decision-making with adoption at commercial and scientific scale. He has published more than 300 papers and received eight best-paper awards across generative AI, reinforcement learning, multi-agent systems, LLM agents, recommender systems, information retrieval, and Bayesian optimisation. His pioneering contributions have helped shape both neural sequence generation and search-based reasoning for large language models. As co-founder of MediaGamma and an advisor to Genie AI, among other entrepreneurial roles, he has helped turn AI research into technologies used at scale. MediaGamma’s innovations underpinned a UCL REF2021 Impact Case Study and contributed to a successful acquisition. Publicly reported figures show that Genie AI has served more than 150,000 users and supported over 200,000 drafted documents, while Nozzle.ai’s platform analyses data covering more than 44 million Amazon customers. His optimisation algorithms are integrated into major AI ecosystems, including Ray Tune and Optuna, and have been applied at CERN and across drug discovery, autonomous laboratories, and materials design. He has also mentored more than 20 PhD researchers and postdoctoral fellows who have become academics, AI leaders, and technology entrepreneurs.

Abstract:

AI is starting to change science from “using computers to analyse data” to “actively helping scientific discoveries”. In drug discovery, chemistry, biology, and materials science, AI can already help the design of molecules, search for promising drugs, improve antibodies, plan experiments, and even control automated laboratories. Here, the idea of Large Discovery Models is introduced: AI systems that can read scientific knowledge, reason about it, remember past experience, learn from success and failure, and decide what to try next. Further analysis uncovers how AI agents use memory and reflection to improve over time without needing to retrain the whole model. These agents store useful past cases, experimental results, and reusable skills, then retrieve them when facing new problems. This allows them to learn continually during use, much like a scientist building experience over many projects. Inhouse developed Memento and Memento-Skills systems show how such memory-based agents can be connected to real industrial AI applications. The main message is that AI is becoming more than a tool for prediction: it is moving towards a new kind of scientific partner that can help generate ideas, plan actions, run experiments, learn from feedback, and accelerate discovery in areas such as chemistry, biology, and drug development.

- - ALL ARE WELCOME - -