COSYNE 2026 Workshop
on
Agent-based Models in Neuroscience:
Theory, Autonomy, Embodiment and Environment

Abstract

The brain is not a computer in a vat; it evolved to control a body interacting with a complex, dynamic environment. To understand complex naturalistic behavior and its neural underpinnings, we must model this fundamental closed loop. This workshop brings together a diverse group of experimentalists and theorists to explore a powerful, growing approach in neuroscience: using agent-based models to link neural processes to the complex, multi-step behaviors they produce. As these approaches proliferate, new questions arise: What levels of detail are necessary to capture animal behavior? How should we balance abstraction versus realism in embodiment? To what extent do neural and behavioral data constrain or validate these models? How can such models complement ongoing experiments and inspire new ones?

This year, we aim to highlight three complementary strands of work, across multiple scales of modeling: The first uses cognitively-oriented agents, typically artificial neural networks trained with deep reinforcement learning, to model complex naturalistic tasks like foraging, social interaction, and long-term planning. These models serve as powerful platforms to test theories of memory, decision-making, and representation, offering compelling explanations for both animal behavior and neural dynamics.

The second paradigm centers on embodied agents, i.e. veritable "digital twins" of animals like flies, fish, and rodents. By embodying neural circuits within physically realistic musculoskeletal models, this approach provides unprecedented insight into the interplay of brain, body, and environment. It enforces morphological constraints on computation, helping to uncover the principles of sensorimotor control, agility, and dexterity.

The third focuses on agentic neuro-theory, fostering dialogue between researchers developing abstract principles and those modeling detailed empirical phenomena. We will explore emerging frontiers that expand the scope of agent-based modeling, such as learning dynamics, incorporating evolutionary pressures through reward functions, internally generated goals, and the role of internal bodily signals (interoception) in shaping behavior.

Through invited talks and a multidisciplinary panel discussion, this workshop will bring together experimental and computational researchers from systems neuroscience, theoretical neuroscience, NeuroAI, reinforcement learning, and biomechanics. We also hope to contrast this approach with the data-driven "foundation model" approach, which is concurrently gaining traction. Our goal is to catalyze conversations across scales and approaches, and to chart how agent-based modeling can advance our understanding of brain function, behavior, and cognition.

Schedule (tentative)

Speakers (confirmed)

Omri Barak

Technion

Cristina Savin

New York University

Lilian Weber

Osnabrück University

Justin Wood

Indiana University

Hannah Choi

Georgia Tech

Xaq Pitkow

CMU

Caroline Haimerl

Champalimaud Foundation

Srini Turaga

HHMI Janelia

Eva A. Naumann

Duke University

Alexander Mathis

EPFL

Jonathan Kao

UCLA (panelist)

Organizers

Satpreet Singh • X

Harvard University

Aran Nayebi • X

Carnegie Mellon University

Elliott Abe • X

University of Washington

Past Editions