Why do valence asymmetries emerge in value learning? A reinforcement learning account

Area of Science:

• Cognitive Psychology
• Neuroscience
• Computational Modeling

Background:

• The Value Learning Task (VLT) assesses how learned value influences stimulus perception and processing.
• Participants learn associations between images and probabilistic outcomes (win, loss, no-change) under uncertainty.
• Human behavior shows an asymmetry: win associations are learned more effectively than loss associations.

Purpose of the Study:

• To investigate the nature of the win-loss learning asymmetry in the Value Learning Task (VLT).
• To determine if a standard reinforcement learning model can replicate this behavioral asymmetry.
• To explain individual differences in learning asymmetry and its effect on value categorization.

Main Methods:

• Utilized a standard error-driven reinforcement learning model with a softmax choice rule.
• Applied the model to data from the Value Learning Task (VLT).
• Analyzed model parameters for empirical fit and task payoff maximization.

Main Results:

• The reinforcement learning model successfully replicated the human learning asymmetry between wins and losses.
• The asymmetry arises from the interplay of neutral initial value estimates and an explore-exploit choice policy.
• Individual learning rates correlate with the magnitude of win-loss asymmetry, explaining behavioral variability.

Conclusions:

• A standard reinforcement learning model accounts for the observed win-loss learning asymmetry without explicit valence coding.
• The model's value estimates provide a parsimonious explanation for post-learning explicit value categorization.
• Understanding this learning bias is crucial for interpreting downstream effects of acquired value in cognitive tasks.