Reinforcement learning in multidimensional environments relies on attention mechanisms.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Reinforcement learning (RL) theories, inspired by computational neuroscience, explain dopamine's role in basal ganglia learning.
• Traditional RL algorithms struggle with high-dimensional environments, limiting their real-world applicability.

Purpose of the Study:

• To investigate how the human brain performs "representation learning" to reduce environmental dimensionality for effective reinforcement learning.
• To identify the neural mechanisms and algorithms underlying this dimensionality reduction process.

Main Methods:

• An experiment was designed to assess human representation learning strategies.
• Neural activity was likely monitored (details not provided in abstract) during a reward-based learning task.

Main Results:

• A bilateral attentional control network, including the intraparietal sulcus, precuneus, and dorsolateral prefrontal cortex, was identified.
• This network appears crucial for selecting relevant task dimensions and updating representations via trial and error.

Conclusions:

• Cortical attention mechanisms interact with basal ganglia learning to solve the "curse of dimensionality" in reinforcement learning.
• Representation learning, mediated by the attentional network, is key to adapting RL to complex, real-world scenarios.