Accelerating discovery of data-driven models and theories of neural dynamics
How does cognition emerge from the complex interaction of biological elements in the brain? Neural dynamics offers a potential bridge between structure and function, where population activity is shaped by a variety of circuit properties and harnessed for computations at multiple timescales. In dissecting these relationships, we are now amassing brain measurements at an unprecedented scale. But while artificial intelligence excels at uncovering associations in large datasets, how AI can complement and advance mechanistic theories of neural dynamics remains an open question.
In this talk, I outline how recent developments—in particular, probabilistic generative models and sequence models—can help link neural dynamics with brain structure and function. First, I discuss how simulation-based inference enables estimation of circuit properties from experimental recordings using a mechanistic model with interpretable parameters. Second, I detail our recent works using state space models and diffusion models to synthesize realistic field potential and spiking data. To close, I share how these approaches can be integrated for building tractable multi-scale models of brain dynamics to bridge neurobiology and neural computation.