ARTFEED — Contemporary Art Intelligence

New Information-Theoretic Model Matches Human Categorization

ai-technology · 2026-05-07

Hey, so there's this new idea about how we categorize things, and it’s based on some pretty cool information theory. Researchers tested this theory against classic studies from Hayes-Roth and Hayes-Roth in 1977, Medin and Schaffer in 1978, and Smith and Minda in 1998. What’s interesting is that this new model does as well as or even better than existing ones, like the independent cue and context models from Medin and Schaffer, Anderson's rational model from 1991, and the hierarchical Dirichlet process by Griffiths and others from 2007. You can check out their findings on arXiv.

Key facts

  • New categorization theory based on information-theoretic rational analysis
  • Tested on experiments by Hayes-Roth and Hayes-Roth (1977)
  • Tested on experiments by Medin and Schaffer (1978)
  • Tested on experiments by Smith and Minda (1998)
  • Matches or outperforms independent cue and context models
  • Matches or outperforms Anderson's (1991) rational model
  • Matches or outperforms Griffiths et al.'s (2007) hierarchical Dirichlet process model
  • Published on arXiv

Entities

Institutions

  • arXiv

Sources