My research
Interpretable time series analysis with Gumbel dynamics, NeurIPS 2026 [paper] [code]
We developed the Gumbel Dynamical Model (GDM), which learns interpretable dynamical components and captures overlapping behavioral patterns with gradual, stochastic transitions.
Identifying connectivity distributions from neural dynamics using flows, ICML 2026 [paper] [code]
We characterized how different connectivity patterns can produce the same neural activity and developed flow-based inference over distributions of compatible connections.
Sketching stochastic valuation functions, INFORMS Journal on Optimization, 2026 [paper] [code]
We proposed compact summaries that approximate a wide range of stochastic valuation functions with controllable representation size and a constant approximation guarantee.
Evidence from spatial transcriptomics for the mosaic hypothesis and pure cell types in the cortex, Cell Reports, 2025 [paper] [code] [poster]
We provided evidence for the long-standing mosaic hypothesis using spatial point processes, showing that neurons of the same molecularly defined cell type tend to avoid one another spatially.
Combinatorial bandits for maximum value reward function under value-index feedback, ICLR 2024 [paper] [code] [poster]
We developed a combinatorial bandit algorithm that observes only the best-performing item's identity and value. When item values come from finite sets, it achieves regret comparable to methods that observe every selected item's outcome.
Two-agent Gumbel dynamics (2MA-GDM), Work in Progress
We are extending Gumbel dynamics to model interacting agents and study how their behavior changes through interaction.
Cell-type-specific changes in psilocybin-induced psychedelic states, Work in Progress
We study psilocybin-induced changes in neuronal spiking dynamics and functional interactions between L5p cell types.
Uncovering cell type neighborhood changes in Alzheimer's disease [poster]
We study changes in cell-type distributions as Alzheimer's disease progresses.
Relational learning for set value functions, PhD thesis, LSE, 2022 [thesis]
My thesis studies inference and optimization for group outcomes, including maximum-likelihood estimation for beta models of random hypergraphs.
Capturing the Purpose of Websites [report]