- This event has passed.
Jay Pina (York University, Physics and Astronomy)
February 11, 2021 @ 10:40 am - 11:30 am
Visual models: A look at different approaches to understanding the mammalian visual system
The mammalian visual system is a large, complex network of neurons composed of a number of spatially separated areas, each of which is highly recurrently connected, coupled together with feedforward and feedback connections. Here, we describe two different approaches that may provide insight into dynamical and computational properties of the mammalian visual system. In one approach, we show that modeling populations of neurons as recurrent nonlocal integrodifferential equations with fixed connection strengths leads to mathematically tractable neuronal networks with dynamics that can qualitatively match experimental data. In particular, we use this neural field approach to model spatially resonant dynamics that are relevant to pattern-sensitive epilepsy, a condition in which static visual stimuli with wavenumbers within a critical range can induce seizures. Consistent with experimental results, we find that only wavenumbers within a critical range induce network activity.
In the second approach, we show that so-called “deep” feedforward neural networks with connection strengths that can vary, ubiquitous in machine learning applications such as object recognition, may provide a framework with which to model learning in mammalian brains. While much doubt has surrounded the biological plausibility of such models, recent work has addressed many of the fundamental criticisms. Here, we present experimental evidence that suggests that the mammalian visual system may instantiate such a feedforward model that learns to predict features of visual stimuli.