A study on Structural Learning Algorithms for Probabilistic Circuits

• Grant number: 22/01798-5
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: April 01, 2022
• End date: July 31, 2023
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Denis Deratani Mauá
• Grantee: Thiago Peres Casagrande
• Host Institution: Instituto de Matemática e Estatística (IME). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

This text describes an undergraduate research project for a study on probabilistic circuits, which are generative probabilistic models in which it is possible to realize different types of inference in a exact and computationally efficient way.A probabilistic circuit is specified through its structure, a directed acyclic graph that models the conditional relations of (in)dependence between the random variables in the model, and its parameters, which are both the weights associated to the edges of the graph and the parameters of the univariate distributions associated to the leafs.In this project the student will do a revision of the literature of the main structural learning algorithms of probabilistic circuits based on a dataset, comparing their complexity in regard to the dimensionality of the problem and the amount of data available.Based on this study, the student will implement a selection of techniques using the open source library of the research group in the language Julia, comparing them on their inference efficiency.