Unsupervised visual discrimination learning of complex stimuli: Accuracy, bias and generalization

Area of Science:

• Cognitive Psychology
• Neuroscience
• Perception

Background:

• Same-different judgments are crucial for environmental interaction.
• Stimulus familiarity and learning significantly impact judgment quality.
• Mechanisms of how learning shapes judgments with complex stimuli remain unclear.

Purpose of the Study:

• Investigate unsupervised visual discrimination learning with complex stimuli.
• Examine how learning influences same-different judgments without explicit labels.
• Determine the underlying processes and transferability of learned discrimination skills.

Main Methods:

• Conducted a nine-day training study with 42 participants performing same-different judgments.
• Utilized unfamiliar complex visual stimuli with equiprobable same and different pairs.
• Manipulated stimulus proportions to differentiate processing biases from proportion-induced biases.

Main Results:

• Participants improved sensitivity and criterion, reducing errors in same-different judgments.
• Developed a bias towards 'different' responses with practice.
• Demonstrated that the bias at equal proportions stemmed from stimulus processing, not just proportion effects.
• Showed full transfer of discrimination ability to novel exemplars.

Conclusions:

• Unsupervised discrimination learning involves enhanced stimulus processing and working memory precision.
• Learning leads to category-specific perceptual expertise.
• Discrimination abilities acquired through training are generalizable to new stimuli within the same category.