Elsevier

Theoretical Computer Science

Volume 268, Issue 2, 17 October 2001, Pages 351-366
Theoretical Computer Science

A comparison of identification criteria for inductive inference of recursive real-valued functions

https://doi.org/10.1016/S0304-3975(00)00275-9Get rights and content
Under an Elsevier user license
open archive

Abstract

In this paper we investigate the inductive inference of recursive real-valued functions from data. A recursive real-valued function is regarded as a computable interval mapping. The learning model we consider in this paper is an extension of Gold's inductive inference. We first introduce some criteria for successful inductive inference of recursive real-valued functions. Then we show a recursively enumerable class of recursive real-valued functions which is not inferable in the limit. This should be an interesting contrast to the result by Wiehagen (1976, Elektronische Informationsverarbeitung und Kybernetik, Vol. 12, pp. 93–99) that every recursively enumerable subset of recursive functions from N to N is consistently inferable in the limit. We also show that every recursively enumerable class of recursive real-valued functions on a fixed rational interval is consistently inferable in the limit. Furthermore, we show that our consistent inductive inference coincides with the ordinary inductive inference, when we deal with recursive real-valued functions on a fixed closed rational interval.

Cited by (0)