Robustness, Stochastics, Uncertainty 3
summary
Gives a practical tour: shows that tiny implementation noise (example cited: 0.001% error breaks feasibility) ruins some linear-program designs, converts nominal LPs into tractable second-order cone programs or SDPs under ellipsoidal uncertainty, and formulates affine feedback (recourse) policies for time‑staged demand (power plant example) so decisions adapt as observations arrive. Covers distributionally robust and chance‑constraint bounds (Chebyshev, KL), connects SVM/L1 regularizers to uncertainty models, and notes applications in finance and sparse text models but limited ML deployments so far.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:56Robust Optimization
- 6:10Distributionally Robust Optimization
- 14:07Recourse
- 22:25Formula
- 33:02Dynamic Programming
- 38:52Supervised Learning
- 48:12Questions
- 51:54Theory
- 53:19Examples
- 57:46Message
related talks
42mExploring Pillars of Trustworthy AI: Robustness and Fairness
48mProvable Robustness Beyond Bound Propagation
01h 24mAdversarial Examples
01h 29mLLM Safety, Alignment, and Generalization
47m1Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers
53mSecurity of Machine Learning Systems
28mSecurity of Machine Learning
01h 49mAI engineering with open access LLMs that lie, curse, and steal