Lecture number | Day | Topics |
| Jan 9 | - NO CLASS
|
1 | Jan 14 | - Course logistics
- Linear model with n << p, sparsity
- Lasso
|
2 | Jan 16 | - l-1 regularized logistic regression
- GLMs with l-1 regularization
- l-1 SVMs, margins
- Matrix completion and trace norm
|
| Jan 21 | - NO CLASS (MLK Jr. Day)
|
3 | Jan 23 | - 1-bit matrix completion
- Non-negative matrix completion
- Covariance and concentration/precision matrix estimation
|
4 | Jan 28 | - Gaussian graphical model selection using l_1 regularized log det
- Gaussian graphical model selection using parallel l_1 regularized regressions
|
5 | Jan 30 | - High dimensional Ising model selection
- Sparse PCA: ScoTLASS, SDP relaxation of l_0 constrained PCA
|
6 | Feb 4 | - Sparse PCA: Generalized power method
|
7 | Feb 6 | - High dimensional k-means
|
8 | Feb 11 | - Sparse subspace clustering
- HW 1 out
|
9 | Feb 13 | - Lasso l_2 error bounds for the linear model case
- Restricted eigenvalue (RE) condition
- Initial project proposals due (deadline extended till Feb 15)
|
10 | Feb 18 | - Sup (i.e. l_\infty) norm error bounds and sign consistency of lasso
- Mutual incoherence condition
|
11 | Feb 20 | - Sup norm error bound continued
- When does the RE condition hold with high probability?
|
12 | Feb 25 | - Proximal methods
- Examples of prox operators
- HW 1 due
|
13 | Feb 27 | - Convergence rates for proximal methods
- Final project proposals due
|
| Mar 4 | - NO CLASS (Spring break)
|
| Mar 6 | - NO CLASS (Spring break)
|
14 | Mar 11 | - Coordinate descent methods
- HW 2 out
|
15 | Mar 13 | - Least Angle Regression (LARS)
|
16 | Mar 18 | - LARS: Lasso modification
- Estimation of high dimensional low rank matrices
|
17 | Mar 20 | - Estimation of low rank matrices: Decomposition lemma for error matrix
|
18 | Mar 25 | - Estimation of low rank matrices: Restricted Strong Convexity (RSC)
- HW 2 due
- HW 3 out
|
| Mar 27 | - NO CLASS
- Attend Prof. Andrew Gelman’s talk: “Causality and Statistical Learning” in the Ford School of Public Policy
|
19 | Apr 1 | - Bound on the maximum singular value of a matrix with iid (multivariate normal) rows
- Gordon’s Theorem
- HW 3 due
|
20 | Apr 3 | - Proof of Gordon’s theorem using Slepian’s inequality
- Gaussian concentration inequality for Lipschitz functions
- HW 4 out
|
21 | Apr 8 | - Fisher’s LDA in high dimensions
|
22 | Apr 10 | - Naive Bayes or Independence Rule in high dimensions
- HW 4 due
|
23 | Apr 15 | - Loss based classification in high dimensions
|
| Apr 17 | - Project presentations I
- Hossein
- Yuan
- Can
- Robert
- Phoenix
|
| Apr 22 | - Project presentations II
- Kam, Yiwei
- Sougata
- Naveen
- Chia Chye Yee
- Xuan
|
| Apr 26 | - Project reports due
|