Research and Publications

This page describes my research contributions, both current and past. Please refer to my Google Scholar to view my publication list.

Vector Symbolic Architecture, Machine Learning and Quantum Computing

At my current PhD position with the Imani group at UC Irvine, I work on the intersection of vector symbolic architecture, machine learning and quantum algorithms. I develop new quantum algorithms that can be used to efficiently decompose semantic information from large dimensional representations generated from neural network models using methods from vector symbolic architecture.

Some of my past work with the Imani group involves DNA pattern recognition, signal compression, and graph representation. My research so far has been published in the leading systems conferences like Design Automation Conference (DAC), International Symposium on Computer Architecture (ISCA), Frontiers in Neuroscience, and International Conference on Computer-Aided Design (ICCAD)

I am also interested in generative models, having taken relevant courses at UCI. I am studying methods to analyze the training dynamics of probabilistic generative models, particularly using methods inspired from the Neural Tangent Kernel (NTK) that has been successfull in explaining the training dynamics of neural network and understanding the information flow through individual weights.

Theoretical Physics

Before working in CS, I was a researcher in theoretical condensed matter physics. I was in the Physics PhD program at University of Maryland, College Park for a year, and a research assistant at the University of Luxembourg before that. My research was primarily in the field of topological superconductivity, and twisted 2-dimensional heterostructures. One of my works involved defining models that hosted perfectly localized topological Majorana modes with higher order topology, and another involves studying a novel superconducting phase hosted in two-dimensional twisted graphene that results from the interplay of spin and superconducting fluctuations. Though it looks like a simple BCS superconductor, it has very strong non-BCS behaviour through the spectral function. These works have been published in Physical Review B and Nature Communications.

Even before that, I had a small stint in biophysics, where I studied the dynamics of lipids. I showed that non-Gaussian features in the nucleotide diffusion over the lipid bilayers can be described as a combination of Gaussian distribution with different scale-dependent diffusion constants. (In hindsight, with my current knowledge of generative models, this seems obvious.) Particularly, I showed a very clean model-invariant relationship between the diffusion constants and a well-defined measure of non-affinity in the local motion of neighbouring lipids.