Speaker: Siavash Golkar

Center for Computational Neuroscience
Flatiron Institute
New York, NY


Title: Explorations in NeuroAI, paths less trod


Abstract: The machine learning and neuroscience community have historically had a symbiotic relationship with advances in one often leading to new insights in the other. In this talk I will take the neuro-to-ML perspective and ask in what ways research in neuroscience can inspire new directions in machine learning. I will give a broad overview of the different learning scales in neuroscience and draw analogs in machine learning literature. Then I will focus on areas that are less often considered: First I will talk about learning at the single neuron level and how neuroscience can motivate potential advances in single neuron design. Then I will talk about structural learning at the population level and show how using a bio-inspired curriculum based framework can drastically reduce the sample complexity in a certain class of learning problems.