Visa version

Version skapad av Saikat Chatterjee 2013-02-01 15:02

Visa < föregående | nästa >
Jämför < föregående | nästa >

Course plan

Course literature and content

We will follow the book -- PETTERN RECOGNITION AND MACHINE LEARNING by CHRISTOPHER M. BISHOP. Following the book, we will cover the following topics.

Introduction to the course (Chapter 1)
Necessary probability theory (Chapter 2)
Linear models for regression (Chapter 3)
Linear models for classification (Chapter 4)
Kernel methods (Chapter 6)
Sparse Kernel Machines (Chapter 7)
Graphical models (Chapter 8)
Mixture models and EM (expectation-maximization) (Chapter 9)
Approximate Inference (Chapter 10)

General information

Course credit: 8 points
Instructor: Saikat Chatterjee
Instructor's office: Osquldas väg 10, floor 3.
Instructor's email: sach@kth.se
Meeting times: For discussion, mail to Instructor and fix a time
Class Room: The list will be given in the 'Schedule' table
Work load: 3 hours per lecture and assignment.
Prerequisites: Good understanding of probability theory.
Teaching and learning methodology: The lectures will be based on blackboard and slides. The style will be more discussion oriented. Students have to present papers and execute projects.

Outcome: Good understanding of the topic and eventually some good research problem formulation, solution and publication.

Schedule and lecture notes

Lecture	Time	Place	Content	Slides
1	2011-10-24 10:00 - 12:00	Osquldasväg 6B plan 1 (Q13)	Introduction to the Course	Slide 1
2	2011-10-31 10:00 - 12:00	Drottning Kristinasväg 30BV (L31)	Uncertainty and Uniqueness	Slide 2
3	2011-11-07 10:00 - 12:00	Drottning Kristinasväg 30 (L42)	Pursuit Algorithms - Design	Slide 3
4	2011-11-14 10:00 - 12:00	Osquldasväg 6B plan 1 (Q11)	Pursuit Algorithms - Guarantees	Slide 4
5	2011-11-21 10:00 - 12:00	Drottning Kristinasväg 30 (L42)	From Exact to Approximate Solutions	Slide 5
6	2011-11-28 10:00 - 12:00	Osquldasväg 6B plan 2 (Q21)	From Exact to Approximate Solutionss	Slide 6
7	2011-12-05 10:00 - 12:00	Osquldasväg 6B plan 1 (Q15)	Iterative-Shrinkage Algorithm	Slide 7
8	2011-12-12 10:00 - 12:00	Osquldasväg 6B plan 1 (Q15)	Towards Average Performance Analysis	Slide 8

List of papers for presentation

For paper presentation, each group consists of 3-4 persons. Each group has to read two good papers and make a technical note followed by presentation. Please find some papers which you may choose.

No	Title
1	A Simple Proof of the Restricted Isometry Property for Random Matrices
2	Asymptotic Analysis of MAP Estimation via the Replica Method and Compressed Sensing
3	Decoding by Linear Programming
4	Distributed Basis Pursuit
5	Information Theoretic Bounds for Compressed Sensing
6	Message Passing Algorithms for Compressed Sensing: Part I and II
7	MIMO Radar Using Compressive Sampling
8	Optimally Sparse Representation in General (non-Orthogonal) Dictionaries via L_1 Minimization
9	Performance Analysis for Sparse Support Recovery
10	Regime Change: Bit-Depth versus Measurement-Rate in Compressive Sensing
11	Signal Processing With Compressive Measurements
12	Support Recovery of Sparse Signals
13	Uncertainty Principles and Ideal Atomic Decomposition
14	Beyond Nyquist: Efficient Sampling of Sparse Bandlimited Signals

According to interest, you may choose other quality papers. Please inform the paper title to the instructor.

Objective: For writing good technical notes, the group members should do intense discussion and exchange of ideas, revolving around the chosen papers. By this way, students are expected to learn from each other.

Tasks for technical note writing: (1) Read the paper with most attention. (2) Understand the central issues - what is the research and why and how? (3) If theorems are there, try to relate with reality. (4) You may not need to rework all the hard proofs - but, for your own learning, try to go as far as possible. (5) Then write a good summary of the research work (within 2-3 pages). (6) Present the work so that audience understands the summary. (7) Provide a thought process such that the papers can be useful for your own research.

Presentation dates: 15th and 16th December, 2011. If we do not finish within these two days, then we may finish by 17th December (thats a Saturday). I prefer that the presentations should be finished by these dates. However, if somebody wants more time, that can be arranged in the first week of March, 2012. You may present on 5th March and 6th March, 2012.

Presentation time limit: Approximately one hour. Ultimately, it depends on the audience interactions.

List of projects

For projects, we are interested in hands on learning. The first project will be given by the instructor to the group. The second project is individual where a student has to choose a good paper (mostly dealing with a practical problem) and understand the paper followed by reproducing the results in the paper. The individual paper work has to be presented which will be counted as the individual paper presentation as well as individual project presentation. The overall format is as follows.

Type	Title
Group	Instrutor will provide the problem. The problem is to implement several algorithms and compare them. This project will be decided after third lecture.
Individual	Choose a practical paper. Read it carefully and try to reproduce the experimental / numerical results. Present the work.

Project submission and presentation date: 20th December, 2011. I prefer that the presentations should be finished by this date. However, if somebody wants more time, that can be arranged in the first week of March, 2012. You may present on 5th March and 6th March, 2012.

Presentation time limit: Approximately 40 minutes. Ultimately, it depends on the audience interactions.

The details of the group project assigned by the istructor is here. Download from ProjectAssignedByInstructor

Ultimate Objective

Being a new research area with lots of activities, the purpose of the course will be served if the students are able come up with new research results.