MW 3:30-4:45 PM, Fall 2013, Math Science 5147 www.stat.ucla.edu/~sczhu/Courses/UCLA/Stat_231/Stat_231.html
Course DescriptionThis course introduces fundamental concepts, theories, and algorithms for pattern recognition and machine learning,
which are used in computer vision, speech recognition, data mining, statistics, information retrieval, and bioinformatics.
Topics include: Bayesian decision theory, parametric and non-parametric learning, data clustering, component analysis,
boosting techniques, kernel methods and support vector machine, and fast nearest neighbor indexing and hashing.
Prerequisites
- Math 33A Linear Algebra and Its Applications, Matrix Analysis
- Stat 100B Intro to Mathematical Statistics,
- CS 180 Intro to Algorithms and Complexity.
- R. Duda, P. Hart, D. Stork, "Pattern Classification", second edition, 2000. [Good for CS students]
- T. Hastie, R. Tibshurani, and J.H. Friedman, "The Elements of Statistical Learning: Data Mining, Inference, and Prediction", Spinger Series in Statistics, 2001. [Good for Statistics students]
- Prof. Song-Chun Zhu, sczhu@stat.ucla.edu, 310-206-8693, office BH 9404.
Office Hours: Monday 1:00-3:00pm - Reader: Joyce Meng, Statistics Graduate Student Lounge, joycemeng@ucla.edu
Office hours: Friday 9-11:00am
Two Homework assignments [td=37]20% |
Three projects:
|
Middle-Term Exam: No. [td]0% |
Final Exam: Dec 10, Tuesday 11:30AM-2:30PM (we only need 2 hours 12:15-14:15, close book exam) [td]35% |
- Homework policy:
Homework must be finished independently. Do not discuss with classmates. - Project policy:
You are encouraged to work and discuss in a group, but each person must finish his/her own project. Hand in
(i) a brief description of the experiment in hard copy, (ii) results and plots in hard copy, (iii) your code in e-copy to the reader. - Late policy:
You have a total of three late days for the class, but after using the three late days, no credit will be given for late homework/project.
Lecture | Date | Topics | Reading Materials | Handouts |
1 | 09-30 | Introduction to Pattern Recognition[Problems, applications, examples, and project introduction] | Ch 1 | |
2 | 10-02 | Bayesian Decision Theory I[Bayes rule, discriminant functions] | Ch 2.1-2.6 | |
3 | 10-07 | Bayesian Decision Theory II [loss functions and Bayesian error analysis] | Ch 2.1-2.6 | |
4 | 10-09 | Component Analysis and Dimension Reduction I:[principal component analysis (PCA)], face modeling][Explanation of Project 1: code and data format] | Ch 3.8.1, Ch 10.13.1Project 1 | HW1Lect4-5.pdf |
5 | 10-14 | Component Analysis and Dimension Reduction II:[Fisher Linear Discriminant ][Multi-dimensional scaling (MDS)] | Ch 3.8.2, Ch10.14 | |
6 | 10-16 | Component Analysis and Dimension Reduction III:[Local Linear Embedding (LLE), Intrinsic dimension] | paper | |
7 | 10-21 | Boosting Techniques I: [perceptron, backpropagation and Adaboost] | Ch 9.5 | |
8 | 10-23 | Boosting Techniques II:[RealBoost and Example on face detection][ Explanation of project II ] | ||
9 | 10-28 | Boosting Techniques III:[Probabilistic analysis, Logit boost] | ||
10 | 10-30 | Non-metric method I: [tree structured Classification: principle and example] | Ch 8.1-8.3 | |
11 | 11-04 | Non-metric method II:Syntactic pattern recognition and example on human parsing | Ch 8.5-8.8 | |
12 | 11-06 | Support vector machine I: Kernel-induced feature space | ||
11-11 | Veterans day holiday |
| ||
13 | 11-13 | Support vector machine II: [Support vector classifier][Explanation of project III] | Ch 5.11 | |
14 | 11-18 | Support vector machine III:[Loss functions, Latent SVM, Neual networks and DeepNet] | ||
15 | 11-20 | Parametric Learning [ Maximum Likelihood Estimation (MLE) ] [ Sufficient Statistics and Maximum entropy ] | Ch 3.1-3.6 | |
16 | 11-25 | Non-parametric Learning I | Ch 4.1-4.5 | |
17 | 11-27 | Non-parametric Learning II:[K-nn classifer and Error analysis] | Ch 4.6handout | |
18 | 12-02 | Non-parametric Learning III: [K-nn fast approximate computing:KD-tree and Hashing ] | ||
19 | 12-04 | Data Clustering and Bi-clustering: [K-mean clustering, EM clustering by MLE, Provable 2-step EM,mean-shift and landscape ] | Ch 10.1-10.4Handout |