Metaplasticity as a Neural Substrate for Adaptive Learning and Choice under Uncertainty

Area of Science:

• Neuroscience
• Computational Neuroscience
• Decision Neuroscience

Background:

• Value-based decision making requires integrating reward information over time.
• This process is complex when rewards are probabilistic and change over time (non-stationary).
• Neural mechanisms for integrating uncertain rewards and estimating risk remain largely unknown.

Purpose of the Study:

• To propose and validate a neural mechanism for integrating reward under uncertainty and estimating uncertainty.
• To investigate the role of reward-dependent metaplasticity (RDMP) in adaptive decision-making.
• To model how learning rates dynamically adjust based on reward history and uncertainty.

Main Methods:

• Developed a computational model based on reward-dependent metaplasticity (RDMP).
• Simulated the model on the probabilistic reversal learning task.
• Compared model predictions with behavioral data from non-human primates.

Main Results:

• The RDMP model successfully performed the probabilistic reversal learning task.
• The model demonstrated dynamic adjustment of learning based on reward feedback.
• Model activity changes reflected unexpected uncertainty.
• Predicted time-dependent and choice-specific learning rates confirmed by primate data.

Conclusions:

• Reward-dependent metaplasticity (RDMP) offers a plausible neural mechanism for adaptive learning and decision-making under uncertainty.
• Metaplasticity enables dynamic adjustment of learning rates based on reward history.
• This mechanism supports both reward integration and uncertainty estimation in decision-making.