Probabilistic circuits: a unified framework for tractable probabilistic modeling

• Grant number: 21/09990-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): October 01, 2021
• Effective date (End): July 31, 2022
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Denis Deratani Mauá
• Grantee: Jonas Rodrigues Lima Gonçalves
• Host Institution: Instituto de Matemática e Estatística (IME). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

This is a proposal for a Scientific Initiation project that aims to study Probabilistic Circuits, which are parametric probabilistic models based on neural networks that allow several inferences to be computed accurately and efficiently. Throughout the project, the student will become familiar with the main algorithms and techniques for parametric learning of these models from data, as well as with the main applications of probabilistic circuits. It is proposed that the student will implement some recently proposed techniques for regularization in parametric gradient learning in the open-source library being developed by the associated research group, thus contributing to current and cutting-edge scientific research. (AU)