Syllabus
This list is very much in flux - and in particular, is overly ambitious.
We probably will not get to the last couple of topics, so please check back often.
- Linear regression
- LMS algorithm
- Normal equations
- Matrix derivatives
- Least squares
- Probablistic motivation
- Locally weighted linear regression
- Nearest neighbors
- Overfitting
- Classification and logistic regression
- Sigmoid loss function
- Perceptron
- Iteratively weighted least squares
- Generalized linear models
- Exponential family
- Bernoulli
- Gaussian
- Recap with GLM models
- Linear regression
- Logistic regression
- Softmax regression
- Generative models
- Guassian/Quadratic discriminant analysis
- Multi-variate gaussian
- Probablistic model
- Comparison with logistic regression
- Naive Bayes
- Laplace smoothing
- Decision Trees
- Entropy gain
- Subspace methods
- Principle Component Analysis (PCA)
- Singular Value Decomposition (SVD)
- Linear Discriminant Analysis (LDA)
- Canonical Correlation Analysis (CCA)
- Independant Componant Analysis (ICA)
- Neural nets
- Seperating hyperplanes
- Hidden layer models
- Backpropogation
- Support vector machines
- Functional and geometric margins
- Quadratic Program (QP) Primal forumaiton
- Lagrange duality
- Support vectors
- Kernals
- Non-separability
- Sequential Minimal Optimization (SMO)
- Coordinate ascent
- SMO
- Kernalized subspace methods
- Boosting
- Exponential loss function
- Adaboost
- Viola Jones face detection
- Learning Theory
- Bias/Variance
- Consistency
- Bounds
- Union bound
- Chernoff bound
- Provably Approximately Correct (PAC) models
- Loss functions
- Empircal versus structural risk
- Sample complexity
- VC dimension
- Regularization and model selection
- Cross validation
- Feature selection
- Bayes statistics for regularization
- Maximum likelihood (ML) versus Maximum a-Posteriori (MAP)
- Structured prediction
- Multi-class generalization
- Vitterbi optimization of markov models
- Margin-based training
- Conditional Random Feilds
- Expectation maximization
- K-means clustering
- Guassian mixture models
- Expected complete log-likelihood