Research

Current research

To survive in ever-changing environments, animals must learn the dynamics of their environment across different timescales and adapt their decisions accordingly. This adaptivity has emerged through evolution, development, and learning, giving rise to diverse computational and biological solutions suited to different environments. Our goal is to uncover the shared- and species-specific mechanisms of such adaptivity along the evolutionary spectrum to identify computational, biological, and ecological principles governing decision-making.

We specifically focus on foraging behavior as a natural decision process shared among species. During foraging, animals must constantly decide whether to exploit a currently depleting resource or leave to explore for alternatives. This exploration-exploitation tradeoff lies at the core of many ethologically relevant decisions that animals face daily, such as searching for food, water, mates, or shelter. By integrating computational models and analysis of data from theory-driven experiments across different species, we aim to uncover (1) what decision strategies different animals use during foraging, (2) how these multi-timescale decisions arise from underlying neural circuits, and (3) relate to the spatiotemporal structure of their unique ecological niche, as well as their sensory-motor constraints.

Our approach combines various methods from reinforcement learning, dynamical systems, and AI to identify computational, algorithmic, and neural mechanisms underlying adaptive decision-making in naturalistic settings. Working in close collaboration with multiple experimental labs, we design theory-driven, comparable foraging tasks across different species, such as C. elegans, zebrafish, and mice, complemented by targeted human experiments conducted in our group. This cross-species, interdisciplinary approach provides a comprehensive understanding of mechanisms shaping adaptive decisions, while highlighting algorithmic and structural inductive biases that can enhance the adaptivity of AI systems.

Recorded talks

BrainNet 2025: Adaptive behavior across timescales.

 

Cosyne 2024 Main Meeting: Long timescales needed for memory tasks arise from distinct mechanisms shaped by learning curricula.

 

Cosyne 2024 Workshops: A census of neural timescales across the mouse brain.