Transferring and generalizing deep-learning-based neural encoding models across subjects.

Area of Science:

• Neuroscience
• Computer Science
• Machine Learning

Background:

• Deep learning models excel at mapping brain information processing for natural vision.
• Current encoding models require extensive subject-specific data, limiting generalization.
• Challenges exist in creating robust models across diverse subjects and populations.

Purpose of the Study:

• To develop methods for transferring and generalizing deep learning-based brain encoding models across subjects.
• To enable efficient creation of subject-specific and population-wide models.
• To improve the scalability and applicability of neural encoding models.

Main Methods:

• Utilized Bayesian inference to refine pre-trained models with limited target subject data.
• Developed progressive training strategies for population-level encoding models.
• Applied methods to functional magnetic resonance imaging (fMRI) data from subjects viewing naturalistic videos.

Main Results:

• Demonstrated efficient transfer and generalization of encoding models across subjects.
• Successfully established subject-specific predictive models of visual cortical representations.
• Created population-wide models capturing hierarchical visual features.

Conclusions:

• The developed methods offer an effective approach for building generalized neural encoding models.
• These advancements facilitate a deeper understanding of brain information processing.
• The study paves the way for more robust and scalable brain-computer interfaces and analyses.