Pedro R. C. Dall'Antonia

prof_pic.jpeg
School of Applied Mathematics.

Rio de Janeiro, Brazil

I am a final-year Master’s student in Applied Mathematics at the School of Applied Mathematics, Getulio Vargas Foundation (FGV EMAp), in Rio de Janeiro, where I also obtained my B.Sc. degree. Currently, I am working under the supervision of Diego Mesquita.

Previously, I worked as a Quantitative Researcher at Kadima Asset Management, following my service as a Midshipman in the Brazilian Navy.

My research lies broadly in Probabilistic Machine Learning. Most recently, I have been focusing on Generative Flow Networks (GFlowNets), investigating methods to help them explore sparse combinatorial spaces more effectively while mitigating mode collapse.

Recently, I have expanded my research to Uncertainty Quantification and rigorous statistical inference with black-box models, working with Conformal Prediction and Prediction-Powered Inference (PPI). Additionally, I am currently working on methodological developments in Computer Vision in collaboration with researchers at the Vision and Computer Graphics Laboratory (Visgraf) of IMPA.

news

Nov 12, 2025 New paper on arXiv: Boosted GFlowNets. We introduce a boosting method to tackle mode collapse in sparse environments.
Oct 17, 2025 New preprint on arXiv: Extending Prediction-Powered Inference through Conformal Prediction. Our approach enables PPI with privacy, robustness, and validity under continuous distribution shifts.

selected publications

  1. Boosted GFlowNets: Improving Exploration via Sequential Learning
    Pedro Dall’Antonia, Tiago Silva, Daniel Augusto Souza, and 2 more authors
    2025
  2. Extending Prediction-Powered Inference through Conformal Prediction
    Daniel Csillag, Pedro Dall’Antonia, Claudio José Struchiner, and 1 more author
    2025