Intelligent Data Analysis 2
Lecture and exercises, winter term 2015/2016. Prof. Tobias Scheffer, Dr. Niels Landwehr, Paul Prasse, Gerrit Gruben.
Dates
The lecture comprises 4 SWS (6 LP).
- Lecture: Monday, 10:15-11:45h, room 03.06.H01, starting 12.10.2015.
- Exercises: Tuesday, 10:15-11:45h, room 3.04.1.02, starting 13.10.2015.
In the first week, the exercise session will take place Wednesday 14.10.2015 16:00 in room 04.0.22.
Lecture content
As a follow-up to the lecture "Intelligent Data Analysis", the lecture will study machine learning algorithms in more depth and detail. Machine learning algorithms build predictive models from data, which can be used to describe an observed system and predict its future behavior. Applications of intelligent data analysis range from the prediction of credit risk to the analysis of observational data in astronomy and personalized music recommendations.
The lecture will be accompanied by a practical project. In the lecture, we will study relevant data analysis techniques and programming languages. In the practical project, students will independently solve a data analysis problem.
Lecture
1. Introduction2. Evaluation
- Slides: Risk, precision and recall, ROC curves, evaluation protocols, model selection.
- Slides: Confidence intervals, statistical tests.
- Video lecture: Confidence intervals, statistical tests.
- Slides: Feedforward networks, backpropagation, autoencoders, Boltzmann machines, convolution.
- Video lecture: Feedforward networks, backpropagation, autoencoders.
- Video lecture: Stacked autoencoders, restricted Boltzmann machines, convolution.
5. Principal component analysis
6. Graphical Models
- Slides: Syntax and semantics of graphical models, D-separation.
- Video lecture: Syntax and semantics of graphical models, D-separation.
- Slides: exact inference, message passing.
- Video lecture: exact inference, message passing.
- Slides: approximate inference, plate notation.
- Video lecture: approximate inference, plate notation.
- Slides: latent Dirichlet allocation, hidden Markov models.
- Video lecture: latent Dirichlet allocation, hidden Markov models.
8. Reinforcement Learning
Exercises
1. Lab: Introduction to Python (Gerrit) Selected solutions: Gradient descent and missclassifications.
2. Lab: Evaluation (Gerrit) Selected solution: Grid Search and CV with sklearn.
Reading recommendation: A few useful things to know about Machine Learning.
3. Theoretical: Hypothesis Testing (Nuno)
4. Lab: Neural Nets I (Nuno)
5. Lab: Neural Nets II (Nuno)
6. Lab: Recommender Systems (Gerrit) Selected solutions: Exploratory Analysis (movielens) and incremental ALS.
7. Lab: PCA and TSNE (Nuno)
8. Theoretical: Graphical Models, D-separation (Ahmed)
9. Theoretical: Graphical models: exact inference, parameter learning (Ahmed)
10. Lab: Graphical models: Gibbs sampling, logic sampling (Ahmed)
11. Lab: Latent Dirichlet allocation (Ahmed)
If you encounter an error with scikit-learn, please update the version via putting !sudo pip install -U scikit-learn in a cell and run it once.
12. Theoretical: Hidden Markov models
12. Theoretical/Practical: Reinforcement Learning