Ricerca ovunque con Google ...

Machine Learning - A.Y. 2011/2012

Instructor	Telephone	Office hours	Studio
Paola Velardi	06-49918356	send e-mail	Via Salaria 113 - 3° floor n. 3412

Course schedule

II semester:

When		Where
tuesday	14.00-15.30	aula seminari
thursday	14.00-15.30	aula seminari

Important Notices

The course is taught in English.

Summary of Course Topics

The course introduces motivations, paradigms and applications of machine learning. Supervised learning: decision trees, instance-based learning, naīve Bayes, Bayesian networks, support vector machine. Unsupervised learning: clustering. Semi-supervised learning. Reinforcement learning. Genetic algorithms and genetic programming. Issues in machine learning: underfitting, overfitting, model selection, error analysis. Students will experiment the main algorithms using the Weka ML software Weka.

Textbooks

• Tom Mitchell, Machine Learning, McGraw-Hill, 1997
• Ethem Alpaydin, Introduction to Machine Learning, MIT Press, 2004
• Pang-Ning Tan, Michael Steinbach, Vipin Kumar, Introduction to Data Mining, Pearson Addison Wesley, 2005

Additional useful texts:

• Stuart Russell, Peter Norvig, Artificial Intelligence: A Modern Approach, Prentice Hall, 2009
• Philipp Koehn, Statistical Machine Translation, Cambridge University Press, 2010

Exam

• Written exam on course material (50%)
• Project (teams of 2-3, joint project with NLP or Web Information Retrieval possible), using the Weka ML toolkit (from 25% to 50%)
• Read and present (in English) one paper on your choice in a (large) list of hot topics (25%, required only if project is "simpler")
• Written exam can be done in two steps , mid-term (april, part I) and end of course (early june, part II), or in july (on the full program)

Mid-Term (if returned by mid-may, is +3 on final grade)

Get the breast cancer data from txt . Use the Neural Network and Decision Tree classication algorithms to classify benign and malignant cases (labeled by +1 and -1). a)How does the algorithm perform on this data set (report the training and test error)? b) What do you think about the results? Is the NN or DT a good choice for this task? Explain why or why not.

Test Examples

Slides and course materials

Timetable	Topic	PPT	Details
2012	Introduction	ppt	Introduction, architecture of ML systems, chessgame example
2012	Inductive Classification	ppt	Learning for categorization, Find-S, Version Space
2012	Decision Trees	ppt	Decision trees, feature selection
2012	Performance Evaluation	ppt	Empirical error analysis
2012	Learning Theory	ppt
2012	Ensemble methods	ppt
2012	Rule-based Learning	ppt
2012	Neural Networks	ppt
2012	Support Vector Machines	ppt
2012	Naive Bayes
2012	Clustering
2012	Reinforcement Learning and Q-Learning
2012	Genetic and Evolutionary Methods

Syllabus

Part A:

• Concept Learning and the General-to-Specific Ordering
• Decision Tree Learning
• Ensamble methods
• Evaluation methods: experimental and theoretical methods
• Rule learning
• Artificial Neural Networks
• Support Vector Machines
• Bayesian Learning
• Reinforcement learning
• Q-learning and Genetic Algorithms
• Clustering

Part B:advanced topics (tentative)

• Populating the Web on a vast scale
• Machine learning to analyse human brain activity

Papers on selected topics

• Learning Chess Games download paper
• Discovery of a set of biologically-driven semantic dimensions underlying the neural representation of concrete nouns download paper
• Active learning Survey download paper