Alex Gomez-Leos
e-mail: alexgomezleos (at) gmail (dot) com
Welcome to my professional / academic homepage. I recently earned a M.Sc. from the ECE Department at UT Austin, where my studies were generously funded by the Virginia & Ernest Cockrell, Jr. Fellowship in Engineering. At UT, I was lucky to work with Gustavo de Veciana and Sanjay Shakkottai in the Wireless Networking and Communications Group. Before that, I earned a B.Sc. in Electrical and Computer Engineering from Texas A&M University, working with Anoosheh Heidarzadeh. At both these schools I got to work with countless brilliant peers and students. It was a joy to teach and learn from each of them.
At UT, my work focused on problems in machine learning and algorithmic statistics: bandits & MDPs, robust regression, distribution testing, and low-rank tensor estimation. At Texas A&M, I fell in love with computing as a scientific and intellectual interest. I don't get to work on much research-level math these days, but I still like working on combinatorial problems in my free time. Here is a research summary!
These days, I work as an R&D engineer with a focus on embedded systems. Most of this is under NDA, so this page primarily catalogues my academic projects.
Publications
-
New! Simple and Nearly-Optimal Sampling for Rank-1 Tensor
Completion via Gauss-Jordan
with Oscar Lopez
Transactions on Machine Learning Research (TMLR), 2025. (final conference version)
Summary: We reduce rank-1 tensor completion to solving a special pair of random linear systems. This view allows us to bound the number of required samples for the original problem. The sample complexity of the latter is given by a matrix sketching problem over GF2, which can't be analyzed using leverage score sampling. To circumvent this, we study a random walk mixing time on a subspace graph. -
Single-Server Private Information Retrieval With Side Information Under Arbitrary Popularity Profiles
with Anoosheh Heidarzadeh
Information Theory Workshop (ITW), 2022. (final conference version) (slides)
Summary: Suppose you want to obtain some item from a database without revealing its identity. A common two-party protocol based on linear codes can ensure perfect privacy guarantees. In the prior work, these guarantees break down if the items are not equally likely to be queried. We developed a randomized protocol to solve the problem. A key technical ingredient was proving an extremal property of our codes.
Other Projects
- New! Progress on Envy-Free Cake-Cutting (pdf, slides) (TL;DR: smoothed EF cake-cutting likely admits a polyquery algorithm)
- Reversible Markov Decision Processes: Potential Applications? (slides) (TL;DR: policy iteration can be substantially faster)
- On Variance-Aware Bandits (pdf) (TL;DR: potential techniques for a new family with controllable but unknown variance)
- Learning Social Influence Functions (pdf) (TL;DR: "natural" graph epidemic processes are PAC-learnable)
- A Note on LinUCB's Smallest Eigenvalue (pdf) (TL;DR: open question - how well does LinUCB uniformly explore?)
Teaching
“Teachers should prepare the student for the student’s future, not for the teacher’s past.” - Richard Hamming- Fall 2024: ECE 360C (UT): Algorithms & Data Structures (TA)
- Spring 2024: ECE 351K (UT): Probability & Random Processes (TA)
- Spring 2022: ECEN 325 (Texas A&M): Analog Circuits Analysis (Peer Mentor)
Fun
My family is from Ciudad Juárez. I love cooking, traveling, and playing the guitar.
Last updated Oct. 2025.