Mind the Duality gap: Log Regret Algo. for online optimization
Online Optimization Process data sequentially for optimization Both speed and memory efficient E.g. stochastic gradient descent for solving SVM

193

Sham Kakade Shai Shalev-Shwartz

Useful for Online Learning (e.g. spam filtering, document ranking) This Work Generalize Primal-Dual framework for strictly convex functions Derive new (log-regret) algorithms