The emergence of task-relevant representations in a nonlinear decision-making task

Area of Science:

• Cognitive Neuroscience
• Decision Science
• Machine Learning

Background:

• Understanding the neural basis of decision-making is crucial.
• Task performance is linked to the ability to form internal representations.
• Latent features often govern complex stimulus-outcome associations.

Purpose of the Study:

• To investigate the relationship between decision-making performance and the emergence of task-relevant neural representations.
• To identify the temporal dynamics of these representations using electroencephalography.
• To determine if better representations correlate with superior task performance.

Main Methods:

• Participants performed two learning tasks with a discoverable latent feature.
• Behavioral accuracy was analyzed based on feature values.
• Representation similarity analysis (RSA) was applied to electroencephalography (EEG) data.
• Participants were categorized as good or bad performers based on classification accuracy.

Main Results:

• Better task performance correlated with a more accurate representation of the latent feature space.
• Task-relevant neural representations were decoded from EEG data.
• These representations emerged approximately 700 ms after stimulus presentation.
• Successful decoding was exclusive to participants with high task performance.

Conclusions:

• Effective decision-making necessitates the creation and extraction of low-dimensional task representations.
• The emergence of these representations is a critical factor for successful performance.
• Neural representations supporting decision-making unfold over a specific time window post-stimulus.