Variational Bayesian Methods

Graduate students and postdocs are invited to learn from Tyler Benster amd Aaron Andalman. Neuroscience data is partially observed, only including certain regions at a particular scale, and statistically dependent, as a single neuron synapses onto thousands of other neurons. Since this violates the assumptions for commonly used statistical tools, we turn to Bayesian methods. As exact Bayesian inference is often computationally intractable, this talk will cover exciting recent results for approximating Bayesian inference as a deep learning optimization problem.