IE613: Online Machine Learning, Jan-April 2017Aim of this course is to study learning algorithms and analyze their performance. Lecture HoursTuesday 3.35-5pmFriday 3.35-5pm LocationLT204Teaching AssistantsSandhya TripatiTA HoursTimings: 2-4pmVenue : Room 201, IEOR building SyllabusIntroduction to batch learning: PAC learning, bias-variance tradeoff, VC-dimension, supervised learning algorithms, brief discussion on clustering methods. Online learning: adversarial and stochastic settings, online learning with expert feedback, bandit feedback. Multi-Armed Bandits: Algorithms for simple and cumulative regret (expected and high probability), and their analysis. Online learning with side information (known and unknown feedback structure). AssingmentsAssignment 1; dataset 1, dataset 2Assignment 2 Assignment 3 A final project List ExamsMidterm, SolutionFinal, Solution Course Grades20 points: Midterm30 points: 3 Coding assingments 30 points: Final exam 20 points: Final project Reference textsClass notesLecture 1: WelcomeLecture 2: Introduction to Supervised learning Lecture 3: PAC learnability Lecture 4: A-PAC learnability Lecture 5: Bias-complexity tradeoff Lecture 6: Bias-complexity tradeoff contd. Lecture 7: VC-dimesnion and its properties Lecture 8: Fundamental theorem of satisitical learning Lecture 9: Linear predictors: Halfspaces,Linear regression, Logistic regression Lecture 10: Soft-SVM, Hard-SVM Lecture 11: Boosting Lecture 12: Introduction to Online Learning Lecture 13: Online Learning-Consitent algorithm, Little-dimension Lecture 14: Halving and Standard Optimal Algorithm (SOA) Lecture 15: Weighted Majority (WM) algorithm Lecture 16: Online Convex Optimization (OCO), Follow the Leader (FTL) Lecture 17: Follow the Regualized Leader (FTRL), Online Gradient Descent (OCD) Lecture 18: Adversarial multi-armed bandits: EXP3 Lecture 19: Adversarial multi-armed bandits: EXP3.P, EXP-IX Lecture 20: Adversarial multi-armed bandits: with side information Lecture 21: Adversarial multi-armed bandits: feedback graphs Lecture 22: Stochastic multi-armed bandits, UCB algorithm Lecture 23: Regret bound of UCB1, Lower bounds Lecture 24: Guest lecture by Prof. Shivram Kalyanakrishnan:, CSE, IITB |