Luiz Facury

Machine learning researcher · UFMG

Luiz Facury

I’m an MSc student in Computer Science at UFMG, supervised by Wagner Meira Jr. I work on self-supervised learning, joint-embedding predictive architectures (JEPAs), and world models.

I also work with Randall Balestriero on the stable family of open-source repositories, JEPA architectures, and world models more broadly.

Publications View CV PDF
Luiz Facury
01

Research

My current work focuses on online JEPA-based learning for Atari using SIGReg regularization, and on planning in learned latent spaces.

I also work on self-supervised and multimodal ECG representations, including zero-shot diagnosis and action-conditioned models of cardiac dynamics.

  • Self-supervised learning
  • World models
  • Latent-space planning
  • ECG representations
02

Selected publications

Google Scholar
2026

NeurIPS 2026 Poster

stable-worldmodel: A Platform for Reproducible World Modeling Research and Evaluation

Lucas Maes*, Quentin Le Lidec*, Luiz Facury, Nassim Massaudi, Ayush Chaurasia, Francesco Capuano, Richard Gao, Taj Gillin, Dan Haramati, Damien Scieur, Yann LeCun, Randall Balestriero

* Equal contribution.

Read paper : stable-worldmodel: A Platform for Reproducible World Modeling Research and Evaluation Code : stable-worldmodel: A Platform for Reproducible World Modeling Research and Evaluation
2025

Computing in Cardiology 2025 Workshop version: LXAI @ ICML 2025 · Oral

Clinically Interpretable Zero-Shot ECG Classification via Multimodal Learning and Expert-Aligned Descriptors

Luiz Facury de Souza, Jose Geraldo Fernandes, Pedro Robles Dutenhefner, Turi Andrade Rezende, Gisele L. Pappa, Gabriela Miana Paixão, Antonio Luiz Pinho Ribeiro, Wagner Meira Jr.

The workshop version appeared as ‘Bridging the Gap in Clinical AI: Zero-Shot Multimodal ECG Analysis with Textual Explanations’.

Read paper : Clinically Interpretable Zero-Shot ECG Classification via Multimodal Learning and Expert-Aligned Descriptors
More publications 6
03

Experience & education

Research & experience

  1. 2024 — Present

    Researcher · SPEED Lab, UFMG

    Self-supervised and multimodal ECG diagnosis, cardiac dynamics, and world models.

  2. Jan — Dec 2025

    Researcher I · Petrobras

    Developed methods to locate and segment regions of interest in geological data for oil exploration.

  3. Feb — Aug 2024

    Research Intern · Philips, Paris

    Developed a self-supervised foundation model for ECGs and submitted two patent applications for automated clinical diagnosis.

  4. Jun — Jul 2023

    Summer Research Intern · Sorbonne Université

    Computer vision for tracking and maintaining focus on a single-celled organism.

Full experience in my CV

Education

  1. 2025 — Present

    MSc in Computer Science

    Universidade Federal de Minas Gerais

    Self-supervised world models and latent-space planning.

  2. 2022 — 2024

    Diplôme d’Ingénieur

    Télécom Paris · Institut Polytechnique de Paris

    Double degree in data science, image processing, and AI. Final project on dense self-supervised segmentation, graded 19/20.

  3. 2018 — 2024

    BS in Electrical Engineering

    Universidade Federal de Minas Gerais

    Specialization in machine learning and computer engineering.

Academic service

Reviewer for the World Models Workshop at ICLR 2026 and Physical Understanding for Decision-Making at NeurIPS 2026.

04

Beyond research

Outside of work, I'm really into cinema — French Nouvelle Vague and Japanese film in particular — and literature, especially art criticism and essays (Sontag, Barthes, etc.). If you want to talk movies, find me on Letterboxd.

Beyond that, I like cooking, photography, and traveling. I've also been reading One Piece weekly since 2013, so it's been a long journey.

Contact

luizfysouza@gmail.com