Neha Wadia

Reveal Contact Info

PhD Student


DeWeese Lab

Current Research

I am interested in understanding the effects of structure in data on the geometry of the training landscape of a neural network. In collaboration with Charles Frye, I am currently studying how the distribution of solutions on the training landscape (as a function of height on the landscape) of a deep network learning to classify images changes as the statistics of the training images are tuned in well-defined ways.

I also work on problems in the field of out-of-equilibrium stochastic processes. In collaboration with Ryan Zarcone and Dibyendu Mandal, I am developing solutions to Fokker-Planck equations for driven systems under various assumptions, and using these solutions and other techniques to develop methods to predict optimal protocols for small stochastic systems.


I was an undergraduate at Amherst College, where I majored in physics because I enjoy the way physicists think about problems. My interests at that time were in the evolution of language and culture, and also in genomics. So in addition to my physics classes, I took courses in psychology, linguistics, biology, and archaeology, but I didn’t double major in any of these subjects because there wasn’t enough mathematics in the requirements! For my senior thesis at Amherst I worked for a year and half on a table-top atomic physics experiment, advised by Larry Hunter. After I graduated from Amherst I spent a year at Perimeter Institute in Waterloo, Canada doing a Masters in theoretical physics. I studied edge effects in two-dimensional spin systems for applications in quantum computing, advised by David Cory. At Perimeter Institute I had the great privilege of meeting information theory and algorithms for the first time (both through studying quantum information), which are not subjects normally to be found in a physics curriculum. After I graduated from Perimeter Institute, I spent a year at the National Center for Biological Sciences (NCBS) in Bangalore, India in the theory group of Mukund Thattai, who studies the evolution of the eukaryotic cell plan. The exposure to the quotidian ins and outs of experimental biology that I gained there greatly influence the way I have thought about science and the role of theory in science ever since. It was also at NCBS that I first met the beautiful theories of stochastic processes and probability, now two of the foundations on which my work stands.

By the time I arrived at Berkeley Biophysics in 2015 I was quite certain that the biological system I was really interested in was the nervous system. I spent my first year rotating in experimental labs, working with flies and bats, until I finally realized once and for all that I preferred doing theory to experiment. Thus I happily arrived at the Redwood Center, where my research now involves a funny combination of all the things I learned and was interested in before coming here, and some wonderful new things such as machine learning. My advisor at the Redwood Center is Mike DeWeese. My specific interests are still developing but I am generally interested in studying toy models (artificial neural networks) to inform and develop theoretical tools to ask and answer questions about brain networks.

When I’m not working I enjoy cooking, reading (especially recipe books), playing piano, and taking long walks.