Analogy as Nonparametric Bayesian Inference over Relational Systems
A new paper on arXiv (2006.04156) proposes a nonparametric Bayesian framework for analogy, treating it as inference over relational systems. The study, conducted by researchers, involved an online behavioral experiment where participants played virtual games with underlying relational structures. Results showed that exposure to a particular structure biased participants' expectations in a test game, with the effect scaling with frequency of exposure. The paper suggests that analogy can be modeled as a nonparametric Bayesian process, offering a computational account of how prior experiences shape generalizations. The work contributes to cognitive science by formalizing analogy as a statistical inference problem, potentially bridging cognitive psychology and machine learning. The paper is available at arxiv.org/abs/2006.04156.
Key facts
- Paper arXiv:2006.04156 proposes analogy as nonparametric Bayesian inference over relational systems.
- Study used online behavioral experiments with virtual games.
- Exposure to relational structures biased expectations in test games.
- Effect scaled with number of exposures.
- Research addresses fundamental question in cognitive science about generalization.
- Paper available at arxiv.org/abs/2006.04156.
Entities
Institutions
- arXiv