Asymmetric learning and adaptability to changes in relational structure during transitive inference.

Area of Science:

• Cognitive Science
• Neuroscience
• Psychology

Background:

• Humans and animals generalize local relationships to global structures transitively.
• Asymmetric learning, updating beliefs about only winners or losers, aids inference from sparse feedback.
• Adaptability to changing relational structures and downstream effects on inference are less understood.

Purpose of the Study:

• Investigate how belief-updating biases interact with adapting to relational structure changes.
• Examine the impact of re-valuing items on inferential knowledge of unchanged items.
• Determine the optimality of asymmetric learning rules in transitive inference paradigms with changepoints.

Main Methods:

• Designed a transitive inference paradigm with two potential changepoints.
• Employed asymmetric (winner- or loser-biased) learning rules.
• Recruited 83 participants to assess sensitivity to relational structure changes.

Main Results:

• Participants showed differential sensitivity to structural changes.
• Learning an item's rank increased ('up' condition) was easier than learning a high-ranking item decreased ('down' condition).
• A reinforcement learning model with adaptable winner bias best captured behavior, showing reduced flexibility in the 'down' condition.

Conclusions:

• Asymmetric learning efficiently infers latent relational structures.
• Winner bias limits adaptability to rank decreases but can be adjusted by successful learners.
• Individual differences in adapting to structural changes relate to the flexibility of asymmetric learning strategies.