DocumentCode :
2356842
Title :
On learning discretized geometric concepts
Author :
Bshouty, Nader H. ; Chen, Zhixiang ; Homer, Steve
Author_Institution :
Dept. of Comput. Sci., Calgary Univ., Alta., Canada
fYear :
1994
fDate :
20-22 Nov 1994
Firstpage :
54
Lastpage :
63
Abstract :
We present a polynomial time online learning algorithm that learns any discretized geometric concept generated from any number of halfspaces with any number of known (to the learner) slopes in a constant dimensional space. In particular, our algorithm learns (from equivalence queries only) unions of discretized axis-parallel rectangles in a constant dimensional space in polynomial time. The algorithm also runs in polynomial time in l if the teacher lies on l counterexamples. We then show a PAC-learning algorithm for the above discretized geometric concept when the example oracle lies on the labels of the examples with a fixed probability p⩽½-1/r that runs in polynomial time also with r. We use these methods, as well as a bounded version of the finite injury priority method, to construct algorithms for learning several classes of rectangles. In particular we design efficient algorithms for learning several classes of unions of discretized axis-parallel rectangles in either arbitrary dimensional spaces or constant dimensional spaces
Keywords :
computational complexity; computational geometry; learning (artificial intelligence); PAC-learning; classes of rectangles; discretized geometric concepts; equivalence queries; finite injury priority method; geometric concept; halfspaces; online learning algorithm; polynomial time; Algorithm design and analysis; Approximation algorithms; Computer science; Injuries; Polynomials;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Foundations of Computer Science, 1994 Proceedings., 35th Annual Symposium on
Conference_Location :
Santa Fe, NM
Print_ISBN :
0-8186-6580-7
Type :
conf
DOI :
10.1109/SFCS.1994.365705
Filename :
365705
Link To Document :
بازگشت