CSC2515 Fall 2003 - Lectures

Tentative Lecture Schedule

• Sept 10 -- FIRST CLASS ONLY ON A WEDNESDAY
  Lecture 1: Introduction to Machine Learning, Generalization and Capacity (notes [ps] [pdf])

• Sept 16 --Classification 1:
  KNN, linear discriminants, decision trees (notes [ps] [pdf])

• Sept 23 -- Classification 2:
  naive Bayes, logistic regression, neural nets for classification (notes [ps] [pdf])

• Sept 30: Assignment 1 (Classification) posted

• Sept 30 -- Regression 1:
  constant model, linear models, generalized additive models (e.g. RBFs), locally weighted regression, multilayer perceptrons/neural networks (notes [ps] [pdf])

• Oct 7 -- Objective Functions and Optimization:
  error surfaces, weight space, gradient descent, stochastic gradient, conjugate gradients, second order methods, convexity, enforcing constraints (notes [ps] [pdf])

• Oct14: Assignment 1 due at the start of class

• Oct 14 -- Regression 2 and Supervised Mixtures:
  credit assignment problem, neural networks, radial basis networks, kolmogorov's theorem, backprop algorithm for efficiently computing gradients, mixtures of experts, piecewise linear models (notes [ps] [pdf])

• Oct21: Assignment 2 (Regression) posted

• Oct 21 -- Unsupervised Learning 1: Trees & Clustering
  K-means, heirarchical clustering, soft competitive learning (Mixtures of Gaussians), maximum likelihood trees, optimal tree structure (notes [ps] [pdf])

• Oct28 -- Unsupervised Learning 2: Mixture models and the EM Algorithm:
  missing data, hidden variables, Jensen's inequality, lower bound on marginal likelihood, free energy interpretation, inference, (notes [ps] [pdf])

• Nov 4: Assignment 2 due

• November 4 -- Unsupervised Learning 3:
  Continuous latent variable models, Factor Analysis, (Probabilistic) PCA, Mixtures of Factor Analyzers, Independent Components Analysis (notes [ps] [pdf])

• Nov 11: Assignment 3 posted

• Nov 11 -- Time Series Models
  autoregressive/Markov models, hidden Markov models (notes [ps] [pdf])

• Nov 18 -- Capacity Control:
  generalization and overfitting, No free lunch theorems, high dimensional issues.
  capacity control methods: weight decay, early stopping, cross validation, model averaging, intro to Bayesianism (notes [ps] [pdf])

• Nov 25: Assignment 3 due at the start of class

• Nov 25 -- Meta-Learning Methods:
  stacking, bagging, boosting (notes [ps] [pdf])

• Dec 2 -- Kernel methods:
  the kernel trick, support vector machines, kernel perceptrons, sparsity, capacity control, dual problems (notes [ps] [pdf])

• December 19 -- projects due

• Extra topics we may or may not have time for
  • other kernel machines: gaussian processes
  • linear dynamical systems, Kalman filtering
  • Approximate inference and learning:
    sampling, variational approximations, loopy belief propagation
  • Spectral Methods:
    Isomap,LLE, spectral clustering,
  • MaxEnt models:
    maximum entropy/energy-based models, iterative scaling, products of experts, dependency nets,
  • Matrix Factorizations:
    aspect models/LDA/plaids, non-negative matrix factorization
  • Automatic Structure Learning:
    sparsity priors, empirical Bayes, automatic relevance determination (MLII), structural EM

CSC2515 - Machine Learning || www.cs.toronto.edu/~roweis/csc2515/