Scholar
Core Francisco Park
Google Scholar ID: RfXjPuEAAAAJ
Harvard University
Citations & Impact
Citations
384
H-index
10
i10-index
10
Publications
20
Co-authors
54
list available
Publications
Resume (English only)
Academic Achievements
- - SmartEM: machine-learning guided electron microscopy (Accepted at Nature Methods)
- - Decomposing Elements of Problem Solving: What 'Math' Does RL Teach? (arXiv)
- - New News: System-2 Fine-tuning for Robust Integration of New Knowledge (arXiv)
- - Competition Dynamics Shape Algorithmic Phases of In-Context Learning (ICLR 2025 Spotlight)
- - ICLR: In-Context Learning of Representations (ICLR 2025)
- - Emergence of Hidden Capabilities: Exploring Learning Dynamics in Concept Space (NeurIPS 2024 Spotlight)
- - 3D Reconstruction of Dark Matter Fields with Diffusion Models: Towards Application to Galaxy Surveys (ICML 2024 Workshop: AI for Science)
Research Experience
- CBS-NTT Postdoctoral Fellow at Harvard
- - Conducted extensive research during his Ph.D. at Harvard Physics, including theoretical and practical projects
- - Researched the potential of AI systems for long-horizon research, particularly through RL to improve mathematical abilities
- - Demonstrated that LLMs could internalize fundamentally new knowledge but only via spending test-time compute
Education
- Ph.D. in Physics from Harvard University, where he explored a remarkably diverse set of research projects, shaping an interdisciplinary approach to science. Theoretical side, investigated fundamental capabilities of AI systems—understanding how models develop in-context learning abilities, learn the structure of data at test time, and learn to manipulate abstract concepts. Tackled some data analysis challenges across multiple domains, such as reconstructing 3D dark matter density fields and creating dust extinction maps for cosmological surveys. Also built practical AI systems, like an interactive deep learning system to track individual neurons in 3D videos of freely moving C. elegans, or a deep learning system that guides electron microscopes in real-time.
Background
- Long-term research goal is to enable AI-driven scientific discovery across multiple fields in science. Current research focuses on:
- - Science of AI: Designing controllable data/environments to systematically understand fundamental working principles of AI
- - AI for Open-ended Science: Investigating what could enable AI systems to conduct long-horizon research that goes beyond existing paradigms to make fundamental discoveries
- Personal research philosophy is to maintain a good balance of fundamental understanding and practical grounding, and to always keep the research open-ended.
Miscellany
- Personal interests and other information not provided
