Problem Decomposition and the Learning of Skills

Abstract

One dimension of “divide and conquer” in problem solving concerns the domain and its subdomains. Humans learn the general structure of a domain while solving particular learning problems in it. Another dimension concerns the solver's goals and subgoals. Finding good decompositions is a major AI tactic both for defusing the combinatorial explosion and for ensuring a transparent end-product. In machine learning, pre-occupation with free-standing performance has led to comparative neglect of this resource, illustrated under the following headings. 1. Automatic manufacture of new attributes from primitives (“constructive induction”). 2. Machine learning within goal-subgoal hierarchies (“structured induction”). 3. Reconstruction of skills from human performance data (“behavioural cloning”).

Cite

Text

Michie. "Problem Decomposition and the Learning of Skills." European Conference on Machine Learning, 1995. doi:10.1007/3-540-59286-5_46

Markdown

[Michie. "Problem Decomposition and the Learning of Skills." European Conference on Machine Learning, 1995.](https://mlanthology.org/ecmlpkdd/1995/michie1995ecml-problem/) doi:10.1007/3-540-59286-5_46

BibTeX

@inproceedings{michie1995ecml-problem,
  title     = {{Problem Decomposition and the Learning of Skills}},
  author    = {Michie, Donald},
  booktitle = {European Conference on Machine Learning},
  year      = {1995},
  pages     = {17-31},
  doi       = {10.1007/3-540-59286-5_46},
  url       = {https://mlanthology.org/ecmlpkdd/1995/michie1995ecml-problem/}
}