English

Learning to Solve the Constrained Most Probable Explanation Task in Probabilistic Graphical Models

Machine Learning 2024-04-18 v1 Artificial Intelligence

Abstract

We propose a self-supervised learning approach for solving the following constrained optimization task in log-linear models or Markov networks. Let ff and gg be two log-linear models defined over the sets X\mathbf{X} and Y\mathbf{Y} of random variables respectively. Given an assignment x\mathbf{x} to all variables in X\mathbf{X} (evidence) and a real number qq, the constrained most-probable explanation (CMPE) task seeks to find an assignment y\mathbf{y} to all variables in Y\mathbf{Y} such that f(x,y)f(\mathbf{x}, \mathbf{y}) is maximized and g(x,y)qg(\mathbf{x}, \mathbf{y})\leq q. In our proposed self-supervised approach, given assignments x\mathbf{x} to X\mathbf{X} (data), we train a deep neural network that learns to output near-optimal solutions to the CMPE problem without requiring access to any pre-computed solutions. The key idea in our approach is to use first principles and approximate inference methods for CMPE to derive novel loss functions that seek to push infeasible solutions towards feasible ones and feasible solutions towards optimal ones. We analyze the properties of our proposed method and experimentally demonstrate its efficacy on several benchmark problems.

Keywords

Cite

@article{arxiv.2404.11606,
  title  = {Learning to Solve the Constrained Most Probable Explanation Task in Probabilistic Graphical Models},
  author = {Shivvrat Arya and Tahrima Rahman and Vibhav Gogate},
  journal= {arXiv preprint arXiv:2404.11606},
  year   = {2024}
}

Comments

Will appear in AISTATS 2024

R2 v1 2026-06-28T15:57:40.090Z