Inportant web pages

The web page for the Computation Learning Theory conference COLT
(In particular check out the link ``COLT Resources'' in the left column)

Web page on Kernal Machines
 

Links were you can find data

 UCI
 DELVE
 Gunnar's benchmarks
Open problems
----------------------------------------------------------

Summary of lectures
 

 Lecture 1:
General introduction to Machine Learning, halfing algorithm,
on-line mistake bounds, VC dimension

Accompaning paper that introduces the on-line model and the
Winnow algorithm ( paper )

Lecture 2:
Learning disjunctions, VC dimension, various algorithms, a reduction

Lecture 3:
Standard Optimal algorithm, mistake bounds for the Winnow and
Weighted Majority algorithm ( paper )

Homework 1

Lecture 4:
How to prove relative loss bounds in the expert's framework ( paper )
More papers on my web page

Lecture 5:
Bayes rule, independence, Bayes algorithm and how it fits into
the expert framework

Lecture 6:
Solution to Homework 1 ( paper containing solution to last problem).
Kraft's inequality. Upper and lower bound
on expected code length. Huffman codes. Entropy and relative entropy.
Motivation of Bayes rule and the expert weight update using a relative
entropy to the prior.

Homework 2

Lecture 7:
Derivation of updates
Gradient Descent versus the Exponentiated Gradient Algorithm ( paper )

Lecture 8:
Expansion into feature space
Kernal functions
Mercer's theorem

Lecture 9:
Opimization Theory
  Langrangians
  Duality

Homework 3

Lecture 10:
 Computing the dual optimization problem
 Geometry of SVMs

Lecture 11:
 SVMs
 Linear Hinge Loss
 Bounds on expected error via leave-one-out

Lecture 12:
 PAC bounds
     Occam's Razor bound for the finite case
     Othogonal rectangles bound
     Compression Schemes

Lecture 13:
  Example of Compression Schemes
  Compression Schemes from mistake bounded learning algorithms
  Open Problem re Compression Schemes
  Sauer's Lemma

Lecture 14:
 Background on boosting
 Ada-boost
  Intro. paper    Tutorial  Confidence-Rated Predictions

Lecture 15:
 Adaboost as entopy projection
 Corrective and totally corrective algorithm
  Trading off the min. edge against entropy

Lecture 16:
 Predicting the stock market
 Cover's algorithm
 EG algorithm
 Shifting expert's algorithm

Lecture 17:
 Density estimation with the exponential family
 Deriving and analyzing algorithms with the Bregman divergence
 In vitro selection and on-line algorithms

Manfred's home page
Back to the CE / CS Class Home Pages
Back to the CE / CS Home Page
Last modified Wednesday, 05-Apr-2000 23:13:17 PDT