0. JD: Jon Do's idea [Jona Doetta]

1. MW: Compare implicit and explicit updates on some natural data.

2. MW: Modify the link function of logistic regression and optimize the matching loss. What is the "best" link function to use? Find the non-decreasing link function that maximizes the likelihood.[David Munday]

3. MW: Thorough experimental analysis on shrinkage versus 2-norm regularization versus early stopping (largely expanded Homework 3).[Ning Bao]

4. MW: Comparison of Naive Bayes versus SVM for spam detection.

5. MW: Use Boosting for span detection and compare against SVM and or Naive Bayes.[Carl Liu]

6. MW: Do a thorough experimental comparison of the Voted Perceptron and SVM on some natural data.

7. MW: Use Boosting as done by Viola et al for designing a face recognision algorithm. [Ben Weber]
Note: I am focusing on object detection using AdaBoost

8. MW: Use Boosting for sparse labeling for some huge corpus.

9. MW: Prove loss bounds for WM (binary labels) by analyzing the total loss along the worst-case path (See extra credit problem of HW2).

10.Active learning with boosting for spam detection. [Nikhila Arkalgud]


Last modified: Wednesday, 27-Feb-2008 20:48:00 PST