Lectures
Lecture Notes for CS491/591-001, Spring 2006
Lecture notes and other class materials will be posted here as they become available.
(TDRL)
- Lecture 02: Jan 19: Introduction to Supervised Learning; definitions and notation; preliminary problem statement.
-
- Lecture Notes
- |Keynote slides| |PowerPoint slides| |PDFified slides|
- Lecture 03: Jan 24: Homework 1 assigned; hypothesis spaces; intro to decision trees; risk functions; geometry of DTs.
-
- Lecture Notes
- |PDFified slides| |PowerPoint slides| |Keynote slides|
- Lecture 04: Jan 26: Cost matrices; learning bias; the entropy impurity function; the final splitting criterion; practical considerations; measuring performance.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 05: Jan 31: HW1 FAQ; function extrema and the vector derivative; performance assessment; cross-validation.
-
- Lecture Notes
- |PDFified slides| |PowerPoint slides| |Keynote slides|
- Lecture 06: Feb 02: Learning curves; variance of a classifier; metric functions; the nearest neighbor rule; geometry of 1-NN; k-nearest neighbors.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 07: Feb 07: HW2 assigned; more on nearest neighbor; the Bayes optimal classification and Bayes optimal thresholds; k-NN as an approximation to the Bayes optimal classifier; nearest-neighbor classification in daily life.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 08: Feb 09: Introduction to linear methods; the squared error loss function; formulation of the learning problem; matrix derivatives; the Gram matrix and the pseudoinverse.
-
- Lecture Notes
- |PDFified slides| |PowerPoint slides| |Keynote slides|
- Lecture 09: Feb 14: Staw poll; group reading #1; solution of the LSE regression problem; multi-class data with binary splits; intro to support vector machines; nonlinear data projections.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 10: Feb 16: Hyperplane geometry; the maximum margin objective function; the quadratic programming problem; linearly nonseparable data and slack variables; kernel functions; interpretations of kernels; the support vector solution.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 11: Feb 21: Monotonicity and optimization; SVMs: "putting it all together"; discussion of Reading 1.
-
- Lecture Notes
- |PDFified slides| |PowerPoint slides| |Keynote slides|
- Lecture 12: Feb 23: Beginning of Bayesianness; meditations on assumptions; the standard Gaussian example; the Bayes-optimal decision rule;analytic Gaussian forms; exercise.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 15: Mar 02: IID samples cont'd; maximum likelihood in action; the Gaussian distribution; joint, marginal, and conditional probabilities; the a posteriori estimator; Bayesian coin flipping.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 16: Mar 07: Administrivia; follow-up on Bayesian estimation; swallows reprised.
-
- Lecture Notes
- |PDFified slides| |PowerPoint slides| |Keynote slides|
- Lecture 17: Mar 09: Introduction to reinforcement learning; definitions: reward function, state space, action space, aggregate reward.
-
- Lecture Notes
- |PDFified slides| |PowerPoint slides| |Keynote slides|
- Lecture 18: Mar 28: RL: outcome uncertainty; transition functions; Markov processes/chains; probability of trajectories/histories; the Markov decision process (MDP) formalism.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 19: Apr 06: Value of a policy in a stochastic world (MDP); principle of maximum expected utility; the formal reinforcement learning problem; the learning problem vs. the planning problem; the policy evaluation problem; the Bellman equation.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 20: Apr 11: R3 assigned; pictorial examples of an MDP, policies, and value functions; the policy iteration algorithm; visualization of policy iteration.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 21: Apr 13: The Q function; policy iteration re-written; the Q-learning algorithm; Q-learning in action; examination of the algorithm.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 22: Apr 18: Notes on final project; action selection; exploration vs. exploitation; on- vs. off-policy learning; explorative policies; use of experience; SARSA(lambda).
-
- Lecture Notes
- |PDFified slides| |PowerPoint slides| |Keynote slides|
- Lecture 23: Apr 20: Presentation do's and don't's; more on eligibility traces; the "forward view" of eligibility; incrementing vs. replacing traces.
-
- Lecture Notes
- |PowerPoint slides| |PDFified slides| |Keynote slides|
- Lecture 24: Apr 25
-
- Lecture Notes
- |Keynote slides| |PDFified slides| |PowerPoint slides|
Last updated: Mon 08 May 2006 04:45:39 PM MDT
