Ryan Zarcone

Reveal Contact Info

PhD Student

Biophysics

Olshausen Lab
DeWeese Lab

Current Research

Measured in FLOPS/watt, the human brain is (roughly) 10^5 times more efficient than a comparable supercomputer. But this efficiency comes at a cost: just as Berkeley undergrads are often allowed two of either sleep, fun, or good grades (but not all three), nature seems to have enforced a three-way tradeoff between power, precision, and speed. Accordingly, the brain has developed algorithms that can quickly and efficiently use imprecise components. Inspired by the brain’s efficiency, I’m interested in information storage/computation with stochastic components. My current work is a first step in this direction: developing algorithms for the storage of analog data on a set of low-power, stochastic storage devices. Specifically, I’m developing an autoencoder framework to learn a mapping from natural images to a set of Phase Change Memory (PCM) devices. From an information theoretic perspective, this can be thought of as performing joint source-channel coding. In the future, I hope to take what I learn from this project and apply it to the more general problem of computing with stochastic elements.

Background

I studied physics at UChicago for undergrad. A summer REU studying granular materials at UChile my sophomore year got me interested in soft condensed matter physics. The rest of my undergraduate research focused on studying phase transitions and dynamics in colloidal systems. By the time I came to grad school I was more interested in studying complex systems than anything I had encountered in the traditional physics canon. And, because the brain is the most interesting complex system (to me), I decided to join the Redwood Center.