I am a third year Ph.D. student in Electrical and Computer Engineering at the University of Texas at Austin under the supervision of Prof. Haris Vikalo and Prof. Robert W. Heath Jr.. I am interested in the theoretical problems in the intersection of machine learning, communications, and their applications to the real world. Particularly, I am working on designing federated learning algorithms that satisfy communication and privacy constraints. I am a member of the Wireless Networking and Communications Group (WNCG).
I completed my B.S. in Mathematics from Universidad de Los Andes in 2015 in Colombia. I worked on Monte Carlo methods for hypothesis testing on contingency tables under the supervision of Prof. Mauricio Velasco. Afterwards, I had the chance to work with Google LA through the Research for industrial Problems for Students program at UCLA, and later as a data science researcher for industry and government projects at Quantil.
I was a TA for Linear Algebra and Integral Calculus at Universidad de Los Andes. At UT Austin, I was a TA for the Data Science Lab undergraduate course, and currently for convex optimization.
I investigated model pruning to reduce the number of parameters in federated settings. We achieve 85% accuracy with only 10% of the parameters. I am also studying a federated framework to factorize a rating matrix. We provide differential privacy guarantees and 30% improvement on average ranking.
The sharing of vast data resources by device networks raises privacy and communication constraints. Mobile devices and autonomous vehicles are some relevant examples. My research focuses on finding how to compress information and quantifying the uncertainty of learning models in distributed settings given this compression. I have also explored collaborative systems among sensors for scene recognition and tracking. Some code can be found here
Selecting transmission parameters in millimeter-wave (mmWave) vehicular communication is a challenging task and current solutions rely on exhaustive search, lookup tables, etc., which leads to significant overheads, delays and suboptimal solutions. We propose to learn these parameters by exploiting situational awareness and leveraging machine learning tools with the past transmissions.
I Investigated Natural Language Processing techniques for Colombia’s peace treaty and collaborated in the development of this site. I also worked on comparing different crime prediction models for the Secretary of Security of Bogota, Colombia.
I was a participant of RIPS LA, a research program for undergraduate students in industrial problems. The project, sponsored by Google LA, explored recommender systems using the Yelp Challenge Dataset and focused on recommender systems using optimization techniques and natural language processing.
Advisor: Mauricio Velasco. We considered a new algorithm to sample from a discrete set by sampling from a convex continuous set and using an effective rounding to obtain integer points from the lattice. This problem has application in sampling contingency tables for hypothesis testing.