Attention modulates value normalization in human reinforcement learning by shaping reward encoding.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Behavioral Economics

Background:

• Contextual valuation in reinforcement learning (RL) is often explained by range normalization.
• Deviations from range normalization occur with equally spaced values, suggesting other factors are at play.
• Attentional processes are hypothesized to cause these distortions in value representation.

Purpose of the Study:

• To investigate the role of attention in outcome normalization within reinforcement learning.
• To test if attentional manipulation causally affects subjective valuation of options.
• To develop an RL model incorporating attention to explain observed valuation biases.

Main Methods:

• Conducted three experiments with 105 participants, using eye-tracking to monitor gaze position.
• Manipulated participant attention using top-down and bottom-up attentional cues.
• Developed a novel reinforcement learning model integrating attentional mechanisms (attentional range model).

Main Results:

• Attentional manipulations significantly increased the subjective valuation of attended options.
• Gaze duration was found to modulate the absolute value of options before range normalization.
• The proposed attentional range model provided a better fit than attention-free models.

Conclusions:

• Attention plays a causal role in shaping value representation during decision-making.
• The attentional range model successfully accounts for observed deviations in contextual valuation.
• Integrating attentional mechanisms is crucial for accurate computational models of value computation.