deep ganguli

image

about me

I am a Ph.D. student in the Center for Neural Science at New York University (NYU). I work with Eero Simoncelli in the Laboratory for Computational Vision. Before coming to NYU, I received my B.S. in Electrical Engineering and Computer Science (EECS) from the University of California Berkeley. Over the years my coursework has been focused on statistical signal processing, information theory, machine learning, and computational neuroscience. Outside of the lab, I enjoy longboarding around Brooklyn and volunteering Sunday mornings as a barista at the Housing Works Bookstore Cafe.

contact: dganguli (at) cns dot nyu dot edu

cv: pdf

research interests

Aspects of human perception, sensorimotor control, and cognition, appear to be consistent with Bayesian inference. However, very little is known about how the probability distributions required by the Bayesian machinery might be represented and computed with in the brain. In collaboration with Eero Simoncelli, I have developed information theoretic predictions for how prior probability distributions over sensory variables can be implicitly encoded with populations of noisy neurons. I have supported a subset of these predictions with natural image/sound statistics, neurophysiological measurements, and human perceptual data. We are currently developing neural decoding algorithms that can combine encoded prior information with likelihood information to perform approximate Bayesian inference. Additionally, Jeremy Freeman and I are developing psycophysical methods to assess what kinds of probability distributions the brain can encode, and to look for signatures of their use in perceptual tasks.

Prior to this work, I spent my first year at NYU as a rotation student in Bijan Pesaran's lab learning electrophysiology and working on decoding algorithms for brain machine interfaces. As an undergraduate I worked in Frederic Theunissen's lab fitting probabilistic models of neural coding to physiological data, and examining the influence of higher order statistics of natural sounds on speech perception. Throughout my work, I have remained interested in how insights from biological systems can lead to superior engineered systems.

recent publications

Neural implementation of Bayesian inference using efficient population codes.   abstract   poster
D Ganguli and E P Simoncelli Computational and Systems Neuroscience (CoSyNe), Feb 2012.

Do humans use Occam's Razor when learning probability distributions?   abstract   poster
J Freeman, D Ganguli and E P Simoncelli Computational and Systems Neuroscience (CoSyNe), Feb 2011.

Implicit encoding of prior probabilities in optimal neural populations.   abstract   pdf
D Ganguli and E P Simoncelli Adv. Neural Information Processing Systems 23 (NIPS*10) , vol.23 pp. 658--666, 2010. Presented at NIPS, Dec 2010.

Orientation statistics at fixation.   abstract   poster
D Ganguli, J Freeman, U Rajashekar and E P Simoncelli 10th Annual Meeting, Vision Sciences Society, May 2010.

Representation of environmental statistics by neural populations.   abstract
D Ganguli and E P Simoncelli Computational and Systems Neuroscience (CoSyNe), Feb 2010.