Spotlight Talks
Brief, strictly timed 5–8 minute introductions to participants’ research and the questions they would like to discuss.
Physics for AIWorkshop 2027
Exploring the principles, dynamics, and emergent behavior of modern AI. Two days of shared questions and open discussion.
About
Physics for AI Workshop 2027 brings together researchers and students working at the intersection of artificial intelligence, physics, and applied mathematics. We will explore how ideas from statistical physics, dynamical systems, optimization, and representation learning can help explain how modern AI systems learn and behave.
The workshop is built around shared scientific questions. Short research introductions, poster discussions, and open discussions will give participants time to exchange ideas, examine unresolved problems, and explore possible collaborations.
We aim to create a relaxed and welcoming environment where participants can think aloud, ask questions, and share work in progress. We value curiosity and thoughtful discussion, with the focus on ideas rather than affiliation or seniority.
Topics
- Learning objectives and physical principles
- Free energy, entropy regularization, variational principles, and energy-based formulations of learning.
- Generative models as dynamical systems
- Diffusion, flow models, stochastic processes, and the dynamics of generation.
- Representations and emergent behavior
- Latent-space structure, geometry, symmetry, and emergent phenomena in foundation models.
- Optimization and training dynamics
- Loss landscapes, implicit bias, phase transitions, and mechanisms of generalization.
These themes are starting points for discussion. We also welcome related questions that connect different perspectives or challenge existing assumptions.
Format
Poster Discussions
Face-to-face exchanges around ongoing work, preliminary ideas, and open problems.
Open Discussions
Extended discussions around shared scientific questions, with room for different viewpoints and emerging directions.
Informal Exchanges
Time to meet other participants, continue conversations, and explore research collaborations.
Program
The detailed program and discussion topics are being developed. We will share the agenda here as plans take shape.
- Dates
- –
- Venue
- Hong Kong Science Park · Specific meeting room to be announced
Organizers
- Shikai FangZhejiang University
- Ruoxi JiangFudan University
- Xinyu LiTsinghua University
- Shengchao LiuThe Chinese University of Hong Kong
- Weiyang LiuThe Chinese University of Hong Kong
- Yue SongTsinghua University
- Ting WangShanghai AI Laboratory
- Licheng WenShanghai AI Laboratory
Participation
Participation in the first edition is by invitation, with researchers encouraged to join alongside one or two students or collaborators. Registration details will be shared directly with invited participants.
We welcome active participation from everyone attending, including questions, preliminary ideas, and perspectives from different disciplines.
Contact
For questions about the workshop or participation arrangements, please contact the organizing team at phys4ai.org@gmail.com.
Learn more about the Physics for AI community.