Learning Under Requirements

on YouTube ↗09m

" summary

Proposes replacing penalty-weighted adversarial training with constraint optimization that enforces robustness as a constraint, reports matching adversarial robustness while improving clean accuracy on state-of-the-art adversarial benchmarks; uses optimization duality by relaxing nonconvex finite problems to infinite-dimensional duality where strong-duality-like properties hold, proves empirical dual solutions generalize, and implements dual-ascent style algorithms for practical training that recover feasible models and empirical improvements.

AI-written summary · about

" chapters (9)

1. 0:00Intro
2. 0:41Robust image recognition
3. 0:57Adversarial training
4. 2:13Constrained learning for robustness
5. 3:30Beyond robustness...
6. 4:14Constrained learning theory
7. 5:21Non-convex variational duality
8. 5:39Sparse logistic regression
9. 6:45What's next?

" related talks

• 34mOffline Reinforcement Learning and Model-Based OptimizationSergey Levine · Reinforcement Learning from Batch Data and Simulation
• 01h 24mAdversarial ExamplesSebastien Bubeck · Deep Learning Boot Camp 2019
• 01h 17mRobust Optimization and GeneralizationJohn Duchi · Modern Paradigms in Generalization Boot Camp 2024
• 47mReinforcement Learning via an Optimization LensLihong Li · Emerging Challenges in Deep Learning (Simons Institute workshop, 2019)
• 53mOff-policy Policy OptimizationDale Schuurmans · Emerging Challenges in Deep Learning (Simons Institute workshop, 2019)
• 42mExploring Pillars of Trustworthy AI: Robustness and FairnessNiharika Shrivastava · Open Source Summit Japan + AI_dev 2024
• 38mBest of Both World Algorithms from I.I.D. to Adversarial DataJulian Zimmert · Quantifying Uncertainty: Stochastic, Adversarial, and Beyond
• 01h 01mRobustness, Stochastics, Uncertainty 2Sasha Rakhlin, Ben Recht, Laurent El Ghaoui · Bridging Continuous and Discrete Optimization Boot Camp 2017