Biased Neural Representation of Feature-Based Attention in the Human Frontoparietal Network

Area of Science:

• Cognitive Neuroscience
• Neuroimaging
• Computational Neuroscience

Background:

• Selective attention is crucial for information processing, modulating neural responses across brain regions.
• The computational principles of attentional modulation and neural representation remain incompletely understood.
• Prevailing theories suggest high-dimensional, distributed neural representations for cognitive variables.

Purpose of the Study:

• To investigate the neural representation of feature-based attention.
• To determine if neural representations are high-dimensional and distributed or simpler and biased.
• To examine how attentional modulation varies across the cortical hierarchy.

Main Methods:

• Analysis of a large dataset from five human functional Magnetic Resonance Imaging (fMRI) studies.
• Examination of voxel-wise response biases to attended versus unattended features.
• Assessment of the strength and hierarchical progression of neural biases.

Main Results:

• A consistent, univariate bias in neural responses favoring the attended feature was observed across multiple cortical areas.
• This bias increased in magnitude along the cortical hierarchy, being strongest in frontoparietal areas.
• Early visual areas showed a lack of this univariate bias, contrasting with higher-level areas.

Conclusions:

• Feature-based attention is represented by a surprisingly simple, low-dimensional biased code, particularly in higher cortical areas.
• This biased representation transitions from analog in early visual areas to more abstract in frontoparietal regions.
• Such a low-dimensional code may facilitate robust cognitive representations and efficient readout for functions like working memory and decision-making.