On universal learning algorithms

https://doi.org/10.1016/S0020-0190(97)00117-8Get rights and content

Abstract

We observe that there exists a universal learning algorithm that PAC-learns every concept class within complexity that is linearly related to the complexity of the best learning algorithm for this class. This observation is derived by an adaptation, to the learning context, of Levin's proof of the existence of optimal algorithms for NP.

References (12)

There are more references available in the full text version of this article.

Cited by (2)

Presented in the Impromptu Session of COLT96.

On sabbatical leave at LCS, MIT.

2

Research was supported in part by an NSF Postdoctoral Fellowship.

View full text