Augmented Hebbian reweighting: interactions between feedback and training accuracy in perceptual learning.

Area of Science:

• Cognitive Psychology
• Neuroscience
• Machine Learning

Background:

• Perceptual learning is influenced by feedback, but its exact role remains debated.
• Existing models struggle to explain the complex interaction between feedback and learning.
• The augmented Hebbian reweighting model (AHRM) proposes feedback acts as an input, not a direct teaching signal.

Purpose of the Study:

• To test the predictions of the AHRM in a perceptual learning task.
• To investigate the interaction between feedback and training accuracy.
• To determine if feedback is essential for learning under different accuracy conditions.

Main Methods:

• Gabor orientation identification task over six training days.
• Accelerated stochastic approximation to track contrast thresholds.
• Four subject groups: high (85%) and low (65%) training accuracy, with and without feedback.

Main Results:

• Contrast thresholds improved in high accuracy groups regardless of feedback.
• Thresholds improved in the low accuracy group only when feedback was provided.
• Results align with AHRM predictions and contradict pure supervised or Hebbian models.

Conclusions:

• Feedback is critical for perceptual learning when training accuracy is low.
• The findings support models where feedback modulates learning rate rather than acting as direct error correction.
• The augmented Hebbian reweighting model provides a viable framework for understanding feedback in perceptual learning.