FA17: ENGR-E 599 Machine Learning for Signal Processing: 35109

FA17: ENGR-E 599 Machine Learning for Signal Processing: 35109

MLSP_Flyer.jpg

  1. Course Title and Instructor

Title: ENGR E599 – Machine Learning for Signal Processing

  • Class hours: 4:00 PM - 5:15 PM Mondays and Wednesdays

 

  1. Course Description

The course discusses advanced signal processing topics as an application of machine learning. Various hands-on signal processing tasks are introduced in class and tackled by using a problem-solving manner, so that the students can grasp some important machine learning concepts along the way. Eventually, the course can help them learn how to build an intelligent signal processing system in a systematical way.

 

  1. Course pre-requisites

The students are assumed that they are accustomed to topics as follows:

Calculus, Linear Algebra, Probability Theory, Machine Learning (CS B555) and one of the following scientific programming languages: MATLAB, Python3, and R.

 

  1. Course Content/ Topics covered 

Weeks

Topic

Details and Activities

Assignments

1

Overview of the MLSP systems; linear algebra basics

Be motivated about the course and accustomed to the real-world data processing; linear algebra basics

 

2

Optimization and Probability Theories

Review basic Optimization and Probability Theories

Problem set

3

Static feature transforms

Traditional ways to extract features from raw signals. Fourier transform and short-time Fourier transform

 

4

Clustering and unsupervised feature learning

Clustering, mixture models, EM. From domain-specific features to data-driven methods using signal examples

Problem set

5

Dimension reduction for high-dimensional signals

General template learning for feature extraction. Principal component analysis, matrix factorization, and probabilistic topic modeling.

 

6

Pattern classification 1

Decision theory, hyperplanes, perceptron

Problem set

7

Pattern classification 2

Template matching and detection problems. Classification of non-sequential signals.

 

8

Temporal data modeling

Time domain signal recognition and tracking: hidden Markov models and Kalman filtering

Problem set

9

Non-linearity in signals

Trajectory of high-dimensional signals. Manifold learning and kernels.

 

10

Regression and Deep Learning for signal processing

Linear regression, logistic regression, and deep learning and their applications in signal processing

Problem set

11

Statistical signal processing: Bayesian methods and graphical models

Markov Random Fields, Bayesian Methods, and inference techniques

 

12

Audio Signal Processing

Speech/music signal processing, recognition, music informatics, and source localization

 

13

Sequential text data processing

Topic modeling, sentiment analysis, and machine translation

 

14

Multimodal signals and beyond

Audio-visual signal processing, EEG denoising/recognition, etc.

 

15

Project Presentations

In class presentation

 

16

Project Presentations

In class presentation

 

  

 5. Teaching and learning methods

 

6. Representative bibliography

     Students are encouraged to read the following books though not required:

 

 7. Student learning outcomes

When students complete this course, they should be able to:

  • Understand basic machine learning and signal processing topics covered in the course
  • Understand advanced machine learning topics covered in the course
  • Fluently utilize the machine learning topics for solving basic signal processing problems
  • Fluently utilize at least one of the advanced machine learning topics to develop an intelligent system for their real-world signal processing problems
  • Redesign existing methods and come up with a better solution in a principled way

 

 8. Grading

Grade Item

Percentage

Projects

40%

Assignments

40%

Quiz

10%

Participation at the piazza discussion boards:

(Piazza sign-up: http://piazza.com/iu/fall2017/fa17blengre59935109 )

 10%

Course Summary:

Date Details