Gabriel Patrón

Gabriel Patrón

Ph.D. candidate (Statistics, University of Michigan). Research in personalized generative models. Award-winning VFX Supervisor for Reflejos (first Latin American feature with diffusion-based VFX). Personalized generative models, personalized generation, and vision–language systems for generative recommendation.

Gabriel Patron

About Me

I am a Ph.D. candidate in Statistics at the University of Michigan (advised by Jeffrey Regier). My research focuses on generative models, personalized generation, and vision–language systems for generative recommendation.

Before academia, I led AI systems in creative workflows. I was the AI-VFX Supervisor for Reflejos, the first Latin American feature film to use diffusion-based visual effects, winner of Best Visual Effects at the Bogotá Horror Film Festival (2025) and the Tabloid Witch Awards (2025). Prior to that, I helped develop one of the first AI-generated virtual influencers.

I believe content consumption should drive content creation, and my work sits exactly in that feedback loop: personalization that generates, not just ranks.

Research Interests

Generative Models: Diffusion models, flow matching, and score-based generative methods, and their applications to science and media.

Variational Inference & Probabilistic Learning: Applying Variational Bayesian methods to improve inference speed with theoretical guarantees at the scale of future astronomical surveys.

Personalized Content Generation: How can we use pre-trained generative foundational media models to move beyond generic recommendation into adaptive, user-driven creation.

Recommender Systems: In the later stage of my PhD, I have developed a keen interest towards recommender systems, and I am excited to incorporate the insights of this field into my own research on personalized generation.

Selected Publications

📄 Recommendations Beyond Catalogs: Diffusion Models for Personalized Recommendation
Gabriel Patrón. Extended Abstract, ACM RecSys 2025 (Doctoral Symposium), Prague.
DOI

📄 An Interpretable Automated ML Credit Risk Model
Gabriel Patrón, Daniel León, Edwin López, Guillermo Hernández. Applied Computer Sciences in Engineering (Springer), 2020.
Springer

Preprints / In Preparation

🧪 Diffusion Models for Inferring Structured Backgrounds in Astronomical Images
Gabriel Patrón, Jeffrey Regier. In preparation for submission to Annals of Applied Statistics (2025).
Summary: diffusion-based EM approach for astronomical background modeling.
(Preprint link will be added here when available.)

🧪 REBECA: Diffusion Priors for Personalized Image Generation
Gabriel Patrón, Zhiwei Xu, Ishan Kapnadak, Felipe Maia Polo. In preparation for submission to CVPR 2026.
Summary: user-conditioned diffusion-prior framework; benchmarks vs. LoRA/VLM personalization on 200K user–image pairs.
(Preprint link will be added here when available.)

Work Experience

Director, AI-VFX & Associate Producer 📍 Bogotá, Colombia

🎬 Reflejos (First AI-VFX-powered Feature Film in Latin America)

  • Led the AI-driven VFX pipeline for the film, integrating generative models for synthetic imagery and video.
  • Developed and showcased AI-generated scene demos to secure investor funding.
  • Built and managed a Machine Learning team, hiring engineers with both technical and artistic expertise.

AI Specialist for Media 📍 Bogotá, Colombia

Dirty Kitchen

  • Developed Diffusion model-based and DeepFake techniques for AI-powered virtual influencers.
  • Built ML-powered tools to automate media production workflows, reducing manual work in post-production.
  • Created @melodijoflora, an AI virtual influencer with 100K+ social media impressions.

Machine Learning Engineer 📍 Bogotá, Colombia

ADAO

  • Developed risk capital compliance software (SIMM by ISDA, FRTB by Basel Committee).
  • Researched NLP-based text mining on 100GB+ financial datasets.
  • Designed benchmarking and testing frameworks for code efficiency.

Awards & Fellowships

  • Reflejos: Award-winning VFX Supervisor (first Latin American feature with diffusion-based VFX).
  • Best Undergraduate Thesis in Statistics (2020) (UNAL)
  • Honorary Degree in Statistics (2020) (UNAL)
  • First Prize – Banistmo Regional (LATAM) Hackathon (USD 5,000)
  • Best Poster – International Statistics Symposium (Stochastic Symphonies)
  • Full Scholarship – Honorary Academic Merit (UNAL) (Top GPA in Statistics Dept.)

Talks & Seminars

🎤 Generative Adversarial Networks for Climate Simulation
📍 Columbia University, IRI Institute (2022)
🔗 [Watch Talk]

🎤 Training GANs with Wasserstein Distance
📍 Quantil AI, Bogotá (2022)
🔗 [Watch Talk]

🎤 AI for Music: Stochastic Symphonies (Best Poster – Int’l Stats Symposium XXVIII)
📍 Universidad del Atlántico, Barranquilla (2018)
🔗 [GitHub]

Teaching

📌 Graduate Student Instructor, University of Michigan (2022 - Present)

  • Winter 2025DATASCI 306 (Majors): Intro to Statistics & Data Analysis
  • Fall 2024DATASCI 101 (Intro): Intro to Data Science
  • Winter 2024STATS 503 (Grad): Statistical Learning II (Multivariate Analysis)
  • Fall 2023STATS 415 (Majors): Data Mining & Statistical Learning
  • Winter 2023STATS 413 (Majors): Applied Regression Analysis
  • Fall 2022STATS 250 (Intro): Intro to Statistics & Data Analysis

Service & Mentoring

🎓 Graduate Mentor – Undergraduate Research Program in Statistics (2024)

  • Supervised Galaxy Cluster Detection research for promising undergraduates.

Selected Software Projects

Before my Ph.D., I worked on various projects that combined my passion for AI with creative applications. Here are some highlights:

🌀 Markovian Music Composer
Trained a Markov Chain-based model to compose original music.

📜 Ancient Greek Text Summarizer
Developed a probabilistic summarization system for Ancient Greek texts (pre-LLM era).

Contact

📧 Email: gapatron@umich.edu
🔗 Google Scholar: Scholar Profile