Rafael Elberg

I study how communication shapes learning systems: when messages become compositional, how compression can support generalization, and how agents can coordinate through learned representations.

I completed my Master's degree in Computer Science at Pontificia Universidad Catolica de Chile, where I worked with Denis Parra and Mircea Petrache. I will start my PhD in the Department of Computer Science at the University of Toronto.

research focus

Communication in learning

How bottlenecks, messages, and interaction protocols change what models learn and how well they generalize.

Compositional representations

How structure emerges in learned representations, especially when systems must communicate efficiently.

Multi-agent intelligence

How reinforcement learning agents can develop useful coordination strategies through complex interaction.

current

Teaching Discrete Mathematics (IIC1253) at PUC Chile.
Preparing to join the University of Toronto CS PhD program.
Broadly interested in machine learning theory and neurosymbolic AI.

teaching

Alongside my research, I have worked as a teaching assistant for Deep Learning, Advanced Topics in AI, Artificial Intelligence, Data Structures and Algorithms, Introduction to Programming, and Advanced Programming at PUC Chile.

elsewhere

In my free time, I enjoy reading, especially science fiction. I keep a small bookshelf with books I have been reading and would like to revisit.

snapshots

news

Jun 02, 2026	I just came back from my first Dagstuhl seminar, plus a week of traveling around Germany. Was a lot of fun!
Mar 26, 2026	I was awarded the best Master’s thesis award by the PUC computer science department for my work “On Compositionality in Emergent Communication”!
Feb 02, 2026	Will officially be starting my PhD next fall in the CS department at University of Toronto :maple-leaf:
Jan 15, 2026	I will be teaching Discrete Mathematics at PUC this semester!
Dec 22, 2025	I got awarded with Honorable Mention for Best Research of 2025 in the annual CENIA event, for my paper in Neurips 2025.

selected publications

NeurIPS

A Compressive-Expressive Communication Framework for Compositional Representations

Rafael Elberg, Felipe del Rio, Mircea Petrache, and Denis Parra

In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

Paper Bib

@inproceedings{elberg2025a,
  title = {A Compressive-Expressive Communication Framework for Compositional Representations},
  author = {Elberg, Rafael and del Rio, Felipe and Petrache, Mircea and Parra, Denis},
  booktitle = {The Thirty-ninth Annual Conference on Neural Information Processing Systems},
  year = {2025},
  url = {https://openreview.net/forum?id=z6mwI6VcHA},
}