dissertation

I'm excited to graduate with my PhD in computer science from UC Berkeley. If you would like to attend my exit talk, please send me an email! Details to follow, but it will likely happen during the week of August 11-15, 2025.

My work explores high-performance algorithms for tensor kernels. A tensor is a multidimensional array, and a kernel is a highly optimized piece of mathematical code. Vectors and matrices are two examples of tensors, but I'm interested in tensors with 3+ dimensions. Unfortunately, tensor kernels are underexplored compared to matrix kernels, both in theory and practice. My thesis aims to fill that knowledge gap by introducing new algorithms for one specific tensor kernel, the Matricized Tensor Times Khatri-Rao Product (MTTKRP). We use those algorithms to accelerate multiple critical computational applications, as well as a launching pad to study other related problems.

A PhD by the Numbers

Graduate work is "bursty": in computer science, it involves long periods of exploration followed by high-intensity code sprints and paper writing. Here's an annotated graph of my average daily Github contributions over time, along with the deadlines for some of the conferences we submitted our work to:

The ICML'23 submission became our NeurIPS'23 paper, and the work for SPAA'24 was completed long before submission. For ACDA'25, most contributions were pull requests that squashed several commits (hence the lower peaks).

Here are some statistics from the past five years. Some numbers (like thesis page count, research diary length, or Github contributions) are pointless and clearly influenced by noise. Others, I'm very proud of.