Dynamics Learning Rate Bias in Pigeons: Insights from Reinforcement Learning and Neural Correlates

Area of Science:

• Neuroscience
• Cognitive Science
• Animal Behavior

Background:

• Reinforcement learning research shows animals process reward prediction errors differently.
• Learning rate bias is observed in humans and animals, but its dynamic changes during learning remain unclear.

Purpose of the Study:

• To investigate dynamic changes in learning rate bias during a probabilistic learning task in pigeons.
• To explore the relationship between behavioral learning strategies and neural activity in the avian striatum.

Main Methods:

• Recorded behavioral choices and local field potentials (LFPs) in pigeon striatum during a probabilistic learning task.
• Applied reinforcement learning models with and without learning rate bias to fit behavioral data and estimate option values.
• Analyzed correlations between striatal LFP power and model-estimated option values.

Main Results:

• Pigeon learning rate bias shifted from negative to positive throughout the learning process.
• Striatal Gamma power (31-80 Hz) correlated with option values influenced by dynamic learning rate bias.
• Behavioral and neural data support a dynamic learning strategy in pigeons.

Conclusions:

• Pigeons utilize a dynamic learning strategy, adjusting their learning rate bias over time.
• Striatal neural activity, specifically Gamma power, reflects this dynamic learning process.
• Findings provide valuable insights into non-human animal reinforcement learning mechanisms.