MACHINE LEARNING for SIGNAL PROCESSING

**Coordinator:**
Murat Saraçlar (http://busim.ee.boun.edu.tr/~murat)**Instructors**:
Murat Saraçlar, Burak Acar

**Catalog Description: **This course is
designed for graduate students. The students will be
exposed to contemporary machine learning approaches as applied to
signal processing. The emphasis of the course is not only the theory
but also practical applications.

**Textbook:**

Duda, Hart, Stork, Pattern Classification, 2nd Ed., Wiley-Interscience, 2004

**Reference Texts:**

Hastie, Tibshirani, Friedman, The Elements of Statistical Learning, Springer, 2001.

Ethem Alpaydin, Introduction to Machine Learning, 2004.

**Prerequisites
by topic:**
Probability

**Topics:**

- Overview of Learning (DHS ch1, EA ch1, HTF ch2)
- Statistical Decision Theory (DHS ch2, EA ch3)
- Parameter Estimation: ML, Bayesian (DHS ch3)
- Linear Regression: Least Squares, Subset Selection and Shrinkage (HTF ch3)
- Linear Methods for Classification: LDA, Perceptron, SVM (DHS ch5, EA ch10)
- Dimensionality Reduction: PCA, Reduced Rank LDA (DHS ch3.8, EA ch6)
- Kernel Methods: Support Vector Machines and Support Vector Regression
- Nonparametric Methods: Parzen Windows, Nearest Neighbors (DHS ch4)
- Model Assessment and Selection: BIC, MDL, VC-Dimension, Cross Validation (DHS ch9, HTF ch7, EA ch14)
- Tree-Based Methods (DHS ch8.2-8.4, HTF ch9.2, EA ch9)
- Ensemble Methods: Voting, Bagging, Bayesian Model Combination, Stacking, Boosting (HTF ch8.7-8.8, ch10, EA ch15)
- Unsupervised Learning: Clustering, Mixture Models (DHS ch10, EA ch7)

**Course Structure:** Three lectures per week, assignments, survey, one
final exam.

**Grading:**

Assignments:
problems and computer implementation, every other week
(40%)

Survey/Project (40%)

Final Exam (20%)

**Prepared
by:** Murat Saraçlar

**Last
revised:** 24/09/07