Jascha Sohl-Dickstein

I am a graduate student in the Redwood Center for Theoretical Neuroscience, at University of California, Berkeley. I am a member of Bruno Olshausen's lab.

I am interested in how we learn to perceive the world. That is, how we start receiving visual input, and auditory input, and tactile input ... and after a while figure out the way the world works. We learn that the light intensity on adjacent patches of retina is correlated, that the world is made up of edges and surfaces, how occlusion works, that objects exist, that if you drop them they fall and make a noise... We basically infer all of physics from examples of sensory input.

How this unsupervised learning problem is solved - how we learn the structure inherent in the world just by experiencing examples of it - is not understood. This is the problem I am interested in tackling.

There are two large (known) parts to this problem. The first is the design of models which are flexible enough to describe *anything* without being unwieldy (greater flexibility frequently comes at the cost of an explosion in the number of parameters, and more computationally costly implementation). The second is practically training these models once you've designed them - estimating their parameters for a particular dataset (something called the partition function problem - a result of the constraint that probabilities of all states must sum to 1 - means that almost any model you can write down is intractable to exactly evaluate). I am interested in both halves, but the majority of my work has been on the parameter estimation side.