Inductive Inference of an Approximate Concept from Positive Data
Abstract
In ordinary learning paradigm, a target concept, whose examples are fed to an inference machine, is assumed to belong to a hypothesis space which is given in advance. However this assumption is not appropriate, if we want an inference machine to infer or to discover an unknown rule which explains examples or data obtained from scientific experiments. In their previous paper, Mukouchi and Arikawa discussed both refutability and inferability of a hypothesis space from examples. In this paper, we take a minimal concept as an approximate concept within a hypothesis space, and discuss inferability of a minimal concept of the target concept which may not belong to the hypothesis space. That is, we force an inference machine to converge to a minimal concept of the target concept, if there is a minimal concept of the target concept within the hypothesis space. We also show that there are some rich hypothesis spaces that are minimally inferable from positive data.
Cite
Text
Mukouchi. "Inductive Inference of an Approximate Concept from Positive Data." International Conference on Algorithmic Learning Theory, 1994. doi:10.1007/3-540-58520-6_85Markdown
[Mukouchi. "Inductive Inference of an Approximate Concept from Positive Data." International Conference on Algorithmic Learning Theory, 1994.](https://mlanthology.org/alt/1994/mukouchi1994alt-inductive/) doi:10.1007/3-540-58520-6_85BibTeX
@inproceedings{mukouchi1994alt-inductive,
title = {{Inductive Inference of an Approximate Concept from Positive Data}},
author = {Mukouchi, Yasuhito},
booktitle = {International Conference on Algorithmic Learning Theory},
year = {1994},
pages = {484-499},
doi = {10.1007/3-540-58520-6_85},
url = {https://mlanthology.org/alt/1994/mukouchi1994alt-inductive/}
}