math189r

Info

Mathematics of Big Data I

Professor Weiqing Gu
Harvey Mudd College
Fall 2016

Readings should be done before class. Reading summaries (specification in info section) are due for all non-Murphy readings at the start of class.

Monday Thursday
Aug. 29 Linear Algebra and Matrix Calculus Review

Convex Optimization
Overview I


(notes from review)
Supervised Learning
Sept. 5
Introduction to Big Data. Linear Regression. Normal Equations and Optimization Techniques. Solving the Normal Equations efficiently (Cholesky Decomposition). Various forms of Linear regression.

Read: Murphy 1.{all} (DEFINITELY READ BEFORE FIRST CLASS)
Read: Murphy, 7.{1,...,5}
Probability Review

(notes from review)

(also Murphy 2.{all} is a great resource)
Sept. 12 Classification. K-Nearest Neighbors. Logistic Regression. Exponential Family and Generalized Linear Models. Logistic Regression as a GLM.

Read: Murphy, 8.{1,2,3,5} \ 8.{3.4,3.5}, 9.{1,2.2,2.4,3} Due: Homework 1 (first third) (solutions)
Convex Optimization
Overview II


Tutorial on Scientific Python
Sept. 19 Generalized Linear Models continued. Poisson Regression. Softmax Regression. Covariance matrix; Multivariate Gaussian Distribution. Marginalized Gaussian and the Schur Complement. Regularization continued. Big Data as a Regularizer.

Read: Murphy 9.7,
4.{1,2,3,4,5,6} (important background)
Due: Homework 1 (second third)
Sept. 26 Dimensionality Reduction; Spectral Decomposition, Singular Value Decomposition, and Principal Component Analysis.

Generative Learning Algorithms. Gaussian Discriminant Analysis.

Due: Final Project Proposal Due: Homework 1 (last third)
Scientific Computing Review / Help Session
Oct. 3 Naive Bayes. L1 Regularization and Sparsity. Lasso. Support Vector Machines. Kernels.

Read: Murphy 14.{1,2,3,4} \ 14.{4.4}
Read: MapReduce: Simplified Data Processing on Large Clusters
Due: Homework 2 (first half) (solutions)
Midterm Review
Unsupervised Learning
Oct. 10
Introduction to Unsupervised Learning. Clustering. K-Means. Mixture of Gaussians. Expectation-Maximization (EM) Algorithm.

Read: Murphy 11.{1,2,3,4} \ 11.{4.6,4.9}
Read: Pegasos: Primal Estimated sub-GrAdient SOlver for SVM
Read: Random Features for Large-Scale Kernel Machines (deep insight)
Due: Homework 2 (second half)
Constrained Optimization
Oct. 17 Fall Break MapReduce and
Distributed Computation
and Learning
Oct. 24 Principal Component Analysis (PCA) Review. Kernel PCA. One Class Support Vector Machines.

Read: Murphy 12.2.{0,1,2,3} 14.4.4
Read: Support Vector Method for Novelty Detection
Due: Midterm. (solutions) (results)
Due: Final Project Progress Report.
Evaluating Models; Deployment
Learning Theory
Oct. 31
Learning Theory. VC Dimension. Bias/Variance Trade-off. Union and Chernoff/Hoeffding Bounds.

Read: Large-Scale Sparse Principal Component Analysis with Application to Text Data
Read: On the Convergence Properties of the EM Algorithm
Due: Homework 3 (first half) (solutions)
Applications of EM: Hidden Markov Models
Recommender Systems
Nov. 7
Introduction to Recommender Systems. Collaborative Filtering. Non-Negative Matrix Factorization. Using Non-Negative Matrix Factorization for Topic Modelling.

Read: Murphy 27.6.2
Read: Netflix Update: Try This at Home
Due: Homework 3 (second half)
Graph Methods
Nov. 14
Graphs. Graph representations as data. The Laplacian and usage of Spectral (Eigenvalue-Eigenvector) information.

Directed Graphical Models (Bayesian Networks). Conditional Independence. Naive Bayes as a Graphical Model. Plate Notation.

Read: Murphy 10.{1,2,3,4,5,6}
Due: Homework 4 (first half) (solutions)
XSEDE Demo
Bayesian Learning
Nov. 21
Recap of Bayesian Reasoning. Bayesian Linear Regression (which we've already seen). Bayesian Logistic Regression. Intractable Integrals and Motivation for Approximate Methods.

Read: Murphy 5.{1,2,3.0,3.2} 7.6, 8.4
Read: Bayesian Online Changepoint Detection
Due: Homework 4 (second half)
Thanksgiving
Nov. 28 Monte-Carlo Methods. Rejection Sampling. Importance Sampling. Intro to Markov-Chain Monte-Carlo. Gibbs Sampling. The Metropolis-Hastings Algorithm.

Read: Murphy 23.{1,2,3,4} \ 23.4.3, 24.{1,2.(1,2,3,4), 3,4} \ 24.{3.7}
Optional Read: Murphy 24.{5,6} Due: Homework 5 (all) (solutions)
Application: Identifying Rapidly Deteriorating Water Quality Locations With Gaussian Processes

Some (~10) Final Project Presentations will be given one day this week.
Dec. 5 Latent Dirichlet Allocation. Nonparametric Models. K-Nearest-Neighbors as a Nonparametric Model. Gaussian Processes. Dirichlet Processes and the infinite mixture of Gaussians.

Read: Murphy 15.{1,2,3,4}, 25.2, 27.{1,2,3}
Read: Gaussian Process Kernels for Pattern Discovery and Extrapolation
Due: Final Project!
Final Project Presentations. We will distribute these across 2 days, where on the final day we will watch presentations of class and instructor chosen extraordinary projects.
Dec. 12 Finals Finals