Prerequisite: |
Vordiplom/Bachelor |
Course description: |
Neural networks are computational models for information processing problems such as pattern recognition, clustering,
novelty detection and regression analysis. This course will cover the following topics:
- Single-Layer Networks
- Multi-Layer Networks
- Error Functions
- Weights Optimization Algorithms and Regularization
- Hopfield Network
- Boltzmann Machine
- Winner Take All Neural Network (Unsupervised Learning)
- Evolving Neural Networks
- Comparison to Support Vector Machine
|
Meeting times and location: |
Friday 9:50 - 11:30, S2|02 C110, Start 04.04.2008 |
Final exam: |
Tuesday, 26.08.2008, 10:00 - 12:00, S2|02 C120 |
Topics and lecture dates |
Topics & slides |
Lecture date: |
Introduction to Neural Computation & Optimization,
Single-Layer Networks, Lecture 1 (PDF), Lecture 1 (PS.GZ) |
04.04.2008 |
Bayesian Decision, Perceptron, LMS, Lecture 2 (PDF), Lecture 2 (PS.GZ)(revised) |
11.04.2008 |
Multi-Layer Networks, Backpropagation, Lecture 3 (PDF), Lecture 3 (PS.GZ) |
18.04.2008 |
Regularization, Over/Underfitting, Momentum, Lecture 4 (PDF), Lecture 4 (PS.GZ) |
25.04.2008 |
Learning Rate, Hopfield Network, Lecture 5 (PDF), Lecture 5 (PS.GZ) |
09.05.2008 |
Boltzmann Machine, Lecture 6 (PDF), Lecture 6 (PS.GZ) |
16.05.2008 |
PCA (with Neural Networks), Feature Extraction, Lecture 7 (PDF), Lecture 7 (PS.GZ) |
30.05.2008 |
Winner-Take-All, K-Means, SOM, Lecture 8 (PDF), Lecture 8 (PS.GZ) |
06.06.2008 |
Evolving Neural Networks, Lecture 9 (PDF), Lecture 9 (PS.GZ) |
13.06.2008 |
Support Vector Machine, Lecture 10 (PDF), Lecture 10 (PS.GZ) |
20.06.2008 |
Neural Networks for Time Series Prediction, Support Vector Regression, Lecture 11 (PDF), Lecture 11 (PS.GZ) |
27.06.2008 |
|
Literature: |
|
Neural Networks for Pattern Recognition,
Christopher M. Bishop,
Oxford University Press, 1995, ISBN: 0198538642
|
|
Neural Networks: A Comprehensive Foundation (2nd Edition),
Simon Haykin,
Prentice Hall Publishers, 1998, ISBN: 0132733501
|
|
Introduction to the Theory of Neural Computation,
John Hertz, Andreas Krogh, Richard Palmer,
Addison Wesley, 1991, ISBN: 0201515601
|
|
Information Theory, Inference and Learning Algorithms,
David MacKay,
Cambridge University Press, 2003, ISBN: 0521642981
|
|
|
R Src Code: |
|
Single Layer Network (Perceptron, LMS, Pseudoinverse), Lecture 1 & 2,
download source code (single.layer.nn.R)
|
|
Network Output as Posterior Probability, Lecture 1 & 2,
download source code (logistic.discr.R)
|
|
Multi-Layer Network, Decision Regions, Lecture 3,
download source code (multi.layer.nn.R)
|
|
Momentum, Lecture 4,
download source code (rosenbrock.function.momentum.R)
|
|
Hopfield Network, Lecture 5,
download source code (hopfield.network.R)
|
|
PCA, Oja's update rule, Lecture 7,
download source code (PCA.Oja.R)
|
|
PCA example (projection + reconstruction), Lecture 7,
download source code (PCA.image.R,charly.pgm)
|
|
Winner-Take-All, Lecture 8,
download source code (WTA.R)
|
|
Support Vector Machine, Lecture 10,
download source code (SVM.R)
|
|
R Literature: |
|
Modern Applied Statistics with S (4th Edition),
W.N. Venables, B.D. Ripley,
Springer-Verlag, 2002, ISBN: 0387954570,
(Note: this book covers also R).
|
|
Programmieren mit R (Statistik und ihre Anwendungen) (2. Auflage),
Uwe Ligges,
Springer-Verlag, 2006, ISBN: 3540363327
|
|