Credal Models for Uncertainty and Logic Treatment

on YouTube ↗52m

" summary

Argues that replacing single probabilities with credal sets (convex sets of distributions) lets models represent ignorance, partial specifications and logical constraints; implements this by converting learned sum–product networks into probabilistic circuits whose sum weights become intervals (epsilon-contamination), then measures per-instance epsilon robustness — the largest perturbation that keeps the same decision. Epsilon is computable for a fixed value in linear time per epsilon (via max/product propagation and tiny knapsack LPs), and experiments show robustness correlates with higher accuracy and useful outlier/uncertainty detection.

AI-written summary · about

" chapters (10)

1. 0:00Introduction to AI challenges
2. 4:37Imprecision in decision making
3. 9:18Probabilistic circuits overview
4. 13:58Random forests and P(x)
5. 17:12Credal sets and robustness
6. 22:19Operations on credal sets
7. 28:04Epsilon robustness in AI
8. 36:04Neuro-symbolic AI integration
9. 40:20Q&A and research directions
10. 45:17Open problem in graph models

" related talks

• 01h 05mLogic and Probabilistic Circuits 1Guy Van den Broeck · Logic and Algorithms in Database Theory and AI Boot Camp
• 01h 03mLogic and Probabilistic Circuits 2YooJung Choi · Logic and Algorithms in Database Theory and AI Boot Camp
• 01h 01mLogic and Probabilistic Circuits 4Guy Van den Broeck · Logic and Algorithms in Database Theory and AI Boot Camp
• 29mA Pre-Expectation Calculus for Probabilistic SensitivityJustin Hsu, Alejandro Aguirre, Gilles Barthe, Joost-Pieter Katoen, Benjamin Lucien Kaminski, Christoph Matheja · POPL 2021
• 34mDistributional Robustness, Learning, and Empirical LikelihoodJohn Duchi · Optimization, Statistics and Uncertainty
• 07mProbabilistic Reasoning and Learning for Trustworthy AIYooJung Choi · 9th Annual Simons Institute Industry Day
• 37mEfficient Algorithms for Reliable Machine LearningAdam R. Klivans · The Role of TCS in Modern Machine Learning
• 09mTowards Credible and Effective Data-Driven Decision-MakingAngela Zhou · Meet the Fellows Welcome Event