CMU 15-859(B), Spring 2007
MACHINE LEARNING THEORY
Lecture Notes & tentative plan
- 01/16: Introduction. PAC model and Occam's razor.
- 01/18: The Mistake-Bound model. Combining expert advice. Connections to info theory and game theory.
- 01/23: The Winnow algorithm + applications.
- 01/25: The Perceptron Algorithm, Margins, and Kernel functions.
- 01/30: Support Vector Machines [Andrew Ng's notes]
- 02/01: Maxent and maximum-likelihood exponential models. Connection to winnow.
- 02/06: Uniform convergence and VC-dimension I. Slides for 1st half + Notes for 2nd half.
- 02/08: VC-dimension II.
- 02/13: Boosting I: weak vs strong learning, basic issues.
- 02/15: Boosting II: Adaboost.
- 02/20: Offline->online optimization. Kalai-Vempala algorithm.
- 02/22: Semi-supervised learning, epsilon-cover bounds. See also the SSL book chapter.
- 02/27: Margins, Kernels and Similarity functions. See also DG paper on JL lemma.
- 03/01: Cryptographic hardness results.
- 03/06: Learning from noisy data. The Statistical Query model.
- 03/08: Characterizing SQ learnability with Fourier analysis. [BFJKMR paper]
- 03/13: [spring break - no classes]
- 03/15: [spring break - no classes]
- 03/20: Using Fourier for learning.
- 03/22: Membership queries I: basic algorithms, KM algorithm. [Mansour survey article]
- 03/27: Membership queries II: Fourier spectrum of DTs & DNFs, Bshouty's alg. [summary]
- 03/29: Membership queries III: Angluin's algorithm for learning DFAs [ppt].
- 04/03: Learning finite-state environments without a reset.
- 04/05: MDPs and reinforcement learning I.
- 04/10: MDPs and reinforcement learning II.
- 04/12: Active learning: slides and notes.
- 04/17: Margin bounds, MB=>PAC bounds.
- 04/19: [carnival - no classes]
- 04/24: Rademacher bounds, McDiarmid's inequality.
- 04/26: Leave-one-out cross-validation bounds. Course retrospective.
- 05/01: Project presentations
- 05/03: Project presentations
No comments:
Post a Comment