Sergey NikolenkoMain page Books Research papers Talks and posters Students Popular science Other stuff Research CS and crypto Bioinformatics Machine learning Algebraic geometry Algebra Bayesian networks Earth sciences Teaching 2014 ML, KFU Game Theory, HSE Mech. Design, HSE ML, CSClub Kazan Game theory, HSE Math. logic, AU Machine learning, STC Machine learning, AU 2013 Discrete math, HSE Machine learning, STC Math. logic, AU Cryptography, AU 2012 Machine learning, STC Math. logic, AU Machine learning II, AU Machine learning, AU Machine learning, EMC 2011 Cryptography, AU Math. logic, AU Machine learning, AU 2010 Math. logic, AU Machine learning, AU Cryptography, AU 2009 Crypto in CS Club Statistics Machine learning, AU Cryptography 2008 Speech recognition MD for CS Club ML for CS Club Mechanism design 2007 Machine Learning Probabilistic learning External links Google Scholar profile DBLP profile LiveJournal account nikolenko (in Russian) | |
Teaching activities |
Machine Learning at the Kazan Federal University, 2014
This is a semester-long machine learning course presented at the
Kazan Federal University
with financial aid from the Dynasty Foundation; see also the
course page at the CSClub website.
The course itself (all slides and lecture notes are in Russian):
- 1. Introduction. History of AI. Probability theory basics. Bayes' theorem and maximal a posteriori hypotheses.
- Slides (.pdf, 304kb)
- 2. Probability distributions. Bernoulli trials. Maximum likelihood, ML estimates for Bernoulli trials and multinomial distribution. Prior distributions, conjugate priors. Beta distribution as a conjugate prior for Bernoulli trials. Predictive distribution: Laplace's rule. Dirichlet distribution as a conjugate prior for multinomial distributions.
- 3. Gaussian distribution. Maximum likelihood estimates for the Gaussian; why the ML estimate for variance is biased. Multidimensional Gaussian. Conditional and marginal Gaussians.
- Slides for lectures 2-3 (.pdf, 741kb)
- 4. Least squares regression. Least squares as an ML estimate for Gaussian noise.
- Slides (.pdf, 329kb)
- 5. Overfitting. Regularization. Ridge regression and lasso regression. Predictive distribution for linear regression. Classification: 1-of-K representation, linear decision functions. Fischer's linear discriminant.
- Slides (.pdf, 1902kb)
- 6. Bayes theorem for classification. LDA and QDA. Logistic regression.
- Slides (.pdf, 1290kb)
- 7. Statistical decision theory. Regression function, optimal Bayesian classifier. Nearest neighbors. Curse of dimensionality. Bias-variance-noise decomposition.
- Slides (.pdf, 545kb)
- 8. Reinforcement learning: multiarmed bandits. Greedy policies, exploration vs. exploitation. Confidence intervals. Minimizing regret: UCB1.
- Slides (.pdf, 265kb)
- 9. Reinforcement learning: Markov decision processes. On-policy and off-policy learning. TD-learning. Machine learning in games (backgammon, chess, go).
- Slides (.pdf, 686kb)
- 10. Clustering. Hierarchical clustering, graph-based clustering. The EM algorithm. EM in general, minorization-maximization, why EM improves the likelihood. EM for clustering.
- Slides (.pdf, 805kb)
- 11. Hidden Markov models. Baum-Welch algorithm. Applications of hidden Markov models to speech recognition.
- Slides (.pdf, 292kb)
- 12. Probabilistic graphical models: basic idea, factorizations, d-separation. Directed and undirected models. Factor graphs.
- Slides (.pdf, 930kb)
- 13. Inference on factor graphs. Belief propagation with the message passing algorithm.
- Slides (.pdf, 820kb)
- 14. Case study: Bayesian rating systems. Bradley–Terry models. Expectation Propagation, TrueSkill, and its extensions.
- Slides (.pdf, 2398kb)
- 15. Approximate inference in PGMs. Loopy belief propagation. Variational approximations (idea).
- 16. Sampling and approximate inference with sampling. Markov chain Monte Carlo methods.
- Slides (.pdf, 658kb)
- 17. Case study: text mining. Naive Bayes. Latent Dirichlet allocation and its extensions.
- Slides (.pdf, 826kb)
- 18. Support vector machines. Kernel trick for SVMs.
- Slides (.pdf, 553kb)
- 19. Case study: recommender systems. Nearest neighbors: user-based and item-based. Locality sensitive hashing.
- 20. Case study: recommender systems. SVD extensions. Additional information in recommender systems. Course review.
- Slides for lectures 19-20 (.pdf, 1100kb)
Selected references.
- Christopher M. Bishop. Pattern Recognition and Machine Learning. Springer, Information Science and Statistics series, 2006.
- Kevin Murphy. Machine Learning: A Probabilistic Perspective, MIT Press, 2012.
- David J. C. MacKay. Information Theory, Inference, and Learning Algorithms. Cambridge University Press, 2003.
|