Probabilistic Reasoning and Learning for Trustworthy AI
summary
Frames fairness, robustness and explainability as probabilistic inference problems and builds a backend that actually answers those queries under uncertainty, using a learned probabilistic model to handle biased or noisy labels and missing data. Develops theory for probabilistic circuits that proves which queries can be computed exactly and efficiently without losing expressiveness, and uses those circuits to produce and enforce quantitative guarantees for downstream models, with ongoing work mapping these ideas into language-model latent spaces.
AI-written summary · about
chapters (5)
- 0:00Trustworthy AI concerns
- 1:18Integrated framework approach
- 2:29Probabilistic modeling needs
- 4:46Probabilistic circuits
- 5:51Q&A on language models
related talks
24mToward Trustworthy AI: Explainability and Robustness
46mCan you trust your AI?
35mWhy Do You Need an Explainable Ai?
01h 05mLogic and Probabilistic Circuits 1
16mTrust by Design: Building Reliable AI for Sensitive Environments
41mFostering Trust via Explainable ML Inferences
41mTheory of Modern AI: Learning Theoretic, Game Theoretic, and Algorithmic Perspectives
46mBlueprints for a Universal Reasoning Machine