I work in theoretical and data-driven cosmology. My current interests are as follows:

• Implicit-likelihood (simulation-based) inference using deep generative models. In particular, I have been interested in multi-fidelity training since it appears to be essential for trustworthy SBI at modern data scale and maps well to the properties of cosmological simulations.
• Higher-order statistics for the non-linear late-time universe. (weak gravitational lensing, galaxy redshift surveys, tSZ, ...).
• Cosmic voids and filaments. The cosmic web is a worthy object of study in its own right, for example to understand galaxy formation. I am also interested in utilizing the physical understanding of the cosmic web as guardrails for learned feature extraction.
• Foundation models for astrophysics (more to come).
• Agentic AI for theoretical physics. I believe that lack of pretraining data and difficulty to construct verifiable RL problems limits the eventual capabilities of agentic AI in theoretical physics. I have therefore initiated a research program aimed at empowering individual physicists to construct their own agentic AI with enhanced capabilities.
• Theoretical understanding of Deep Learning. I hope that a better first-principles theoretical understanding of deep neural networks will enable more trustworthy applications to data analyses in cosmology and beyond. To this end, I am collaborating with high-energy theorists to identify the correspondence between neural networks and physical systems with many degrees of freedom.

Since March 2024, I work as a project assistant professor in the Center for Data-Driven Discovery at Kavli-IPMU/U-Tokyo. I received a PhD in Physics from Princeton University where David Spergel advised me. Previously, I received a BA from Oxford University and obtained an MSc through the PSI program at Perimeter Institute.

Contact: leander (dot) thiele (at) ipmu (dot) jp