Machine Learning


Subject to change and constantly being modified

Note: I may be off by a lecture for some of the first half of the
semester items.  Also note that if you use this to index the
videotapes in the library (e.g., for doing the final exam), the
numbering may be one off after the midterm...

LECT:	topics:

1 intro
  GP for Tetris

2 find-S

3 version spaces
  candidate elimination algorithm

4 decision trees

5 ID3

6 variations on ID3
7 numerical attributes

8 neural networks: linear units

9 sigmoid units

10 multi-layer nets and backprop
11 "how backprop works"
   NN rap: "Learn This"

12 last lecture on NNs

13 Koza GP video

14 genetic algorithms

15 othello GP HW assignment details
   midterm review   

   ** MIDTERM EXAM **  March 9


16 midterm solutions
   student NN presentations
17 student NN presentations

18 finish NN presentations beginning of class
   GP variations

19 GP variations continued

20 Structuring your final project
   Final project choices
   GP variations continued

21 Naive Bayes

22 anomoly detection -- special guest speaker Eleazar Eskin

23 Naive Bayes, continued
   Naive Bayes applied to text classification

24 financial application
   competitive fitness measures
   competitive fitness measures with empirical (fixed) training cases
   multi-objective optimization
   more meta-learning, e.g., stacked classifiers
   cross-validation for multiple purposes
   hybrid learning techniques
   evaluation/comparison of hyptheses/learners
   Grand Unified Theorem -- how all these relate
   (actually, some of above has to spill into lecture 25)

25 Final project writeup guidelines
      i.e., how to write/read research papers
   evaluation of hypothesis and of learning methods
    comparing learning techniques
    recall/precision; pareto optimality
   really coolGP variations:
      ontogeny: cellular encoding
      machine language rep and other linear reps
      Turing-complete hypotheses

26 Eleazar day again!
     Sparse Markov Transducers and Mixture techniques 
        applied to protein homology detection

27 finish up coolGP (see 25 above)
   Summary and final exam review
   universal hierarchy of learning techniques: dimensions of learning

FINAL EXAM TIME: TBA - if you find this out please email me.
   I heard it is May 11. -Eric

email: evs at cs dot columbia dot edu