PAC analogues of Perceptron and Winnow via boosting the margin

Machine Learning

Volumn 47, Issue 2-3, 2002, Pages 133-151

  Servedio, Rocco ^a  

^a HARVARD UNIVERSITY   (United States)

Author keywords

Boosting; Linear threshold functions; Probably approximately correct learning

Indexed keywords

COMPUTATION THEORY; ERROR ANALYSIS; LEARNING ALGORITHMS; MATHEMATICAL MODELS; NEURAL NETWORKS; ONLINE SYSTEMS; THEOREM PROVING;

PROBABLY APPROXIMATELY CORRECT (PAC) LEARNING;

LEARNING SYSTEMS;

EID: 0036568031     PISSN: 08856125     EISSN: None     Source Type: Journal    
DOI: 10.1023/A:1013633619373     Document Type: Article

References (39)

1. Queries and concept learning
2. Tracking the best disjunction
3. Generalization performance of support vector machines and other pattern classifiers
4. The perceptron algorithm is fast for nonmalicious distributions
5. A polynomial lime algorithm for learning noisy linear threshold functions
6. Learnability and the Vapnik-Chervonenkis dimension
7. Worst-case analysis of the perceptron and exponentiated update algorithms
8. Learning noisy perceptrons by a perceptron in polynomial time
9. An improved boosting algorithm and its implications on learning complexity
10. Boosting a weak learning algorithm by majority
11. Game theory, on-line prediction and boosting
12. A decision-theoretic generalization of on-line learning and an application to boosting
13. Large margin classification using the perceptron algorithm
14. The robustness of the p-norm algorithms
15. General convergence results for linear discriminant updates
16. Majority gates vs. general weighted threshold gates
17. Probability inequalities for sums of bounded random variables
18. Learning sparse perceptrons
19. Combinations of weak classifiers
20. Recent results on boolean concept learning
21. On the boosting ability of top-down decision tree learning algorithms
22. Cryptographic limitations on learning boolean formulae and finite automata
23. The perceptron algorithm versus winnow: Linear vs. logarithmic mistake bounds when few input variables are relevant
24. Learning quickly when irrelevant attributes abound: A new linear-threshold algorithm
25. From online to batch learning
26. Redundant Noisy attributes, attribute errors, and linear-threshold learning using winnow
27. Halfspace learning, linear programming, and nonmalicious distributions
28. How fast can a threshold gate learn?
29. The strength of weak learnability
30. Drifting games
31. On PAC learning using Winnow, Perceptron, and a Perceptron-like algorithm
32. Boosting the margin: A new explanation for the effectiveness of voting methods
33. Improved boosting algorithms using confidence-rated predictions
34. Identification criteria and lower bounds for Perceptron-like learning rules