Competitive Algorithms for Online Control
summary
Presents a reduction from LQR-style problems (input-disturbed squared regulator) to a new online convex optimization with structured memory, then designs an algorithm called optimistic regularized online balance descent that provably achieves a horizon-independent constant competitive ratio versus the offline optimal. Uses analysis from smoothed online convex optimization and convex body chasing, proves dimension-free constant bounds under stabilizability and bounded prediction sets, reports NeurIPS submission and experiments where the method outperforms best static linear controllers in practical tracking/control cases.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:15Online Optimization
- 2:03Competitive Ratio
- 4:23Sustainable Data Centers
- 5:49Robotics and Control
- 6:48Convex Body Chasing
- 8:41Online Control
- 22:38Quantifying Uncertainty
- 23:49Lower Bounds
- 26:48Be the Leader
- 27:33Linear Controllers
related talks
42mA Regret Minimization Approach to Mutli-Agent Control and RL
35mOnline Optimization and Learning Under Long-Term Convex Constraints and Objectives
43mContextual Inverse Optimization: Offline and Online Learning
01h 01mRobustness, Stochastics, Uncertainty 2
43mOnline Learning via Offline Greedy Algorithms: Applications in Market Design and Optimization
01h 10mAnalysis and Design of Optimization Algorithms via Integral Quadratic Constraints
31mLearning to Control Unknown Systems
01h 00mContinuous Methods 4