EECS 545: Machine Learning

University of Michigan, Winter 2012

Classroom: DOW 1005
Time: MW 10:30am-12:00pm
Instructor: Honglak Lee
Instructor office hours: M 2:30-4:30pm
GSI: Guanyu Zhou
GSI office hours: Tu 3:00-4:00pm and Th 4:00-5:00pm in 1637 BBB.

Contact: For all questions, please use Piazza (registration required).
NOTE: Please note that this is a tentative syllabus and subject to change.

Course Description
The goal of machine learning is to develop computer algorithms that can learn from data or past experience to predict well on the new unseen data. In the past few decades, machine learning has become a powerful tool in artificial intelligence and data mining, and it has made major impacts in many real-world applications.

This course will give a graduate-level introduction of machine learning and provide foundations of machine learning, mathematical derivation and implementation of the algorithms, and their applications. Topics include supervised learning, unsupervised learning, learning theory, graphical models, and reinforcement learning. This course will also cover recent research topics such as sparsity and feature selection, Bayesian techniques, and deep learning. In addition to mathematical foundations, this course will also put an emphasis on practical applications of machine learning to artificial intelligence and data mining, such as computer vision, data mining, speech recognition, text processing, bioinformatics, and robot perception and control. The course will require an open-ended research project.

Text books

• Chris Bishop, "Pattern Recognition and Machine Learning," Springer, 2006.
• Hastie, Tibshrani, Fiedman, "Elements of Statistical Learning," Springer, 2010. (available online)
• Sutton and Barto, "Reinforcement Learning: An Introduction," MIT Press, 1998
• (optional) Boyd and Vandenberghe, "Convex Optimization," Cambridge University Press, 2004. (available online)
• (optional) Mackay, "Information Theory, Inference, and Learning Algorithms," Cambridge University Press, 2003. (available online)

• EECS 492: Introduction to Artificial Intelligence (This is an official prerequisite, but it is not accurate in that you can still take this course if you haven't completed EECS 492.)
  
• Linear algebra (equivalent to MATH 217, MATH 417)
• Multivariate calculus
• Probability (equivalent to EECS 401)
• Programming skills (equivalent to EECS 280, EECS 281, and experience in MATLAB)

Grading
Homework: 30%
Midterm: 30%
Project: 40% (progress report 10%; final project 30%)
* Up to 2% extra credit may be awarded for active class participations.

Important dates
Project progress report due: 3/9 23:59pm
Midterm exam: 3/26 5:30pm-8:30pm, GGBL 1504
Final project report due: 4/24 23:59pm
Final project poster presentation: 4/26 9am-12pm, Tishman Hall, Beyster Building

Homework
There will be four or five (approximately bi-weekly) problem sets to strengthen the understanding of the fundamental concepts, mathematical formulations, algorithms, and the applications. The problem sets will also include programming assignments to implement algorithms covered in the class.

Project
This course offers an opportunity for getting involved in open-ended research in machine learning. Students are encouraged to develop new theory and algorithms in machine learning, or apply existing algorithms to new problems, or apply to their own research problems. Please talk to the instructor before deciding about the project topic. Students will be required to complete their project proposals, progress reports, poster presentations and the final report.

Details of topics to be covered (tentative)

• Introduction (1 lecture)
• Regression (2 lectures)
  • Linear regression
  • Gradient descent and stochastic gradient
  • Newton method
  • Probabilistic interpretation of linear regression: Maximum likelihood
• Classification (2 lectures)
  • k-nearst neighbors (kNN)
  • Naive Bayes
  • Linear discriminant analysis/ Gaussian discriminant analysis
  • Logistic regression
  • Generalized linear models, softmax regression
• Kernel methods (4 lectures)
  • Kernel density estimation, kernel regression
  • Support vector machines
  • Convex optimization
  • Gaussian processes
• Regularization (2 lectures)
  • L2 regularization
  • L1 regularization, sparsity and feature selection
  • Bias-Variance tradeoff
  • Overfitting
  • Cross validation, model selection
  • Advice for developing machine learning algorithms
• Neural networks (1 lecture)
  • Perceptron
  • MLP and back-propagation
• Learning theory (2 lectures)
  • Sample complexity
  • VC dimension
  • PAC learning
  • Error bounds
• Graphical models (4 lectures)
  • Bayesian networks
    • Representation
    • Exact inference
    • Sampling based inference
  • Learning in Bayesian networks
    • Maximum likelihood estimation
    • Expectation maximization
    • Hidden Markov Models (HMM)
  • Structure learning
  • Bayesian inference and learning
  • Markov networks
    • Inference and learning
• Unsupervised learning (4 lectures)
  • Clustering: K-means
  • Gaussian mixtures
  • Expectation Maximization (revisited)
  • PCA
  • Dimensionality reduction: ISOMAP, LLE
  • ICA
  • Sparse coding
  • Boltzmann machines and autoencoders, Deep belief networks
• Reinforcement learning (3 lectures)
  • MDP
  • Value iteration and policy iteration
  • Dynamic programming
  • Value function approximation
  • TD learning