Parameter learning but not structure learning: a Bayesian network model of constraints on early perceptual learning

Area of Science:

• Cognitive psychology
• Computational neuroscience
• Visual perception

Background:

• Perceptual learning allows adaptation to new environments.
• Existing models do not fully explain the constraints on perceptual learning.

Purpose of the Study:

• To investigate constraints on early perceptual learning.
• To differentiate between parameter learning and structure learning in cue acquisition.
• To test if learning is limited to potentially dependent variables.

Main Methods:

• Formalized learning constraints using Bayesian networks.
• Designed five experiments to test cue acquisition in novel environments.
• Manipulated the dependency between scene and perceptual variables.

Main Results:

• Subjects successfully acquired new cues when learning required parameter learning (dependent variables).
• Subjects failed to acquire new cues when learning required structure learning (independent variables).
• Results indicate a bias in early perceptual learning mechanisms.

Conclusions:

• Early perceptual learning is constrained; individuals can modify existing statistical relationships (parameter learning) but not create entirely new ones (structure learning).
• Learning is limited to variables already considered potentially dependent.
• This bias facilitates adaptation in naturalistic environments but limits learning in novel, unrelated contexts.