Sylvia Madhow

Reveal Contact Info

PhD Student

Neuroscience

Bouchard Lab

Current Research

As the mammalian brain receives an auditory signal, it effectively processes, filters, and interprets the stimulus to extract complex, ethologically relevant information. I am interested in harnessing insight into signal processing and stimulus representation in the mammalian brain to develop algorithms that are resilient to noise and variability for machine sensing tasks.

The brain’s robust representation of complex auditory objects and its ability to separate many sources from a low-dimensional signal, compared to modern sensing algorithms, suggest that neural feature selectivity is a highly effective method of coding important information.

One of my primary goals is to characterize this selectivity in the auditory cortex, and how neural representations map sound stimuli to ethologically relevant cues. More specifically, I study distributed representations of natural sound stimuli in the rat auditory cortex. To model feature representation and correlated variability in A1, I use task-optimized, nonlinear deep networks as phenomenological models for simultaneously recorded ECoG and intracortical polytrode data. This permits investigation of higher-order statistics in the neural data from a large set of the auditory cortex and study of population dynamics of sensory encoding.

Background

I grew up with a little too much Asimov on my shelf, and that has influenced every project I have chosen as an adult, from jaunts in statistical mechanics to the tense AI/neuroscience border where I now reside.

Before starting my doctoral studies in neuroscience at Berkeley, I studied physics at the beach (i.e., UC Santa Barbara). There, I worked as an undergraduate researcher on analysis of astronomical data and signal processing for detector arrays in the Mazin lab. I will always have a soft spot for all matters relating to astronomy, complex analysis, general relativity, binary systems, and SETI.

As a senior and post-grad, I also worked in infrared video signal processing algorithms and data analytics at FLIR Systems. Among other things, I learned many ways to NOT use a screwdriver, and various methods for breaking camera lenses. A few of my long-term projects involved developing calibration models for cameras with temperature-dependent dark current, applying machine learning to manufacturing data, and developing and testing signal processing algorithms for infrared cameras.

Outside of the lab, I spend my time dancing Argentine tango, reading books to stay well-supplied with fresh bad ideas, getting more paint on my face than my sketchbook, having long discussions about terrible movies, and making fresh pasta.