Feature selection for inductive generalization

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Machine Learning

Background:

• Similarity judgments are fundamental cognitive processes enabling prediction and generalization.
• Current understanding assumes selective attention to salient features, but the mechanisms of feature selection and weighting remain unclear.

Purpose of the Study:

• To develop and present a computational method for assessing feature weights in similarity judgments.
• To investigate how individuals select and weigh features when judging similarities and categorizing stimuli.

Main Methods:

• A novel procedure integrating image-processing techniques with a machine-learning algorithm was employed.
• The method was used to assess feature weights that could explain behavioral data from similarity and categorization tasks.

Main Results:

• The computational analysis identified a small set of local features crucial for explaining behavioral data.
• Feature weights derived from the model accounted for both similarity and categorization judgments.

Conclusions:

• The developed computational method offers insights into the feature selection and weighting processes underlying human similarity judgments.
• A limited number of local features appear to play a disproportionately important role in cognitive similarity assessments.