State Space Models for Natural and Artificial Intelligence

Join us for a talk by Scott Linderman, Assistant Professor of Statistics and an Institute Scholar in the Wu Tsai Neurosciences Institute at Stanford University. This talk is part of the Kempner Seminar Series, a research-level seminar series on recent advances in the field.

New recording technologies are revolutionizing neuroscience, allowing us to measure the spiking activity of hundreds to thousands of neurons in behaving animals. These technologies offer exciting opportunities to link brain activity to behavioral output, but they also pose statistical challenges. Neural and behavioral data are noisy, high-dimensional time series with nonlinear dynamics and substantial variability across subjects. I will present our work on state space models (SSMs) for such data. The key idea is that these high-dimensional measurements reflect the evolution of latent states, which shed light on the neural computations underlying natural intelligence. For example, we have used SSMs to study how attractor dynamics in the hypothalamus encode persistent internal states during social interaction, and to connect stereotyped movements to moment-to-moment fluctuations in brain activity. Perhaps surprisingly, stacks of simple SSMs like these now form the basis of several state-of-the-art models for sequential data, including natural language. Next, I will present a parallel line of work in my lab that focuses on deep SSMs and algorithms for evaluating them in sublinear time using modern hardware. Together, these projects highlight the central role of state space models in our studies of both natural and artificial intelligence.

Coming from Longwood? Sign up to take the shuttle here.