The shadowing effect of initial expectation on learning asymmetry.

Area of Science:

• Cognitive Science
• Neuroscience
• Computational Psychology

Background:

• Positivity and optimism biases are observed in complex belief updates.
• A consensus on learning asymmetries in simpler reinforcement learning (RL) remains elusive.
• RL learning asymmetry involves differential sensitivity to positive and negative prediction errors (PEs).

Purpose of the Study:

• To investigate whether initial value expectations influence learning asymmetries in reinforcement learning (RL).
• To determine if default value expectations bias prediction error calculation and subsequent choices.
• To test a model incorporating both asymmetric learning rates and initial value expectations.

Main Methods:

• Two learning experiments were conducted with varying reinforcement probabilities.
• Experiments included monetary gains, losses, and mixed gain-loss environments.
• A computational model integrating asymmetric learning rates and initial value expectations was tested.

Main Results:

• Results consistently supported the proposed model.
• Initial value expectations were found to influence prediction error calculation.
• The model incorporating initial expectations explained choice preferences effectively.

Conclusions:

• Initial value expectations play a crucial role in value updating and choice behavior.
• The proposed model provides a more comprehensive understanding of learning rate asymmetry in RL.
• Understanding initial expectations is key to resolving controversies in RL learning asymmetry research.