Roberto Garcia

About

I’m a PhD student in Computational and Mathematical Engineering at Stanford, advised by Prof. Chris Ré. I previously did my masters in ICME at Stanford and worked on mid-frequency trading at Hudson River Trading. I have also interned at Jane Street, Meta and Google.

I am interested in developing machine learning methods from mathematical first principles, with a focus on understanding neural architectures and training dynamics. My recent work explores how structural and geometric insights can be used to both improve efficiency and reveal fundamental limits of neural architectures.

Selected Publications

Constructing Efficient Fact-Storing MLPs for Transformers

Owen Dugan*, Roberto Garcia*, Ronny Junkins*, Jerry Liu*, Dylan Zinsley, Sabri Eyuboglu, Atri Rudra, and Chris Ré

paper

Adaptive Rank Allocation: Speeding Up Modern Transformers with RaNA Adapters

Roberto Garcia, Jerry Liu, Daniel Sorvisto, and Sabri Eyuboglu

paper · ICLR 2025

Combining Constructive and Perturbative Deep Learning Algorithms for the Capacitated Vehicle Routing Problem

Roberto García-Torres, Alitzel Adriana Macias-Infante, Santiago Enrique Conant-Pablos, José Carlos Ortiz-Bayliss, and Hugo Terashima-Marín

paper