Hierarchical Neural Representation of Dreamed Objects Revealed by Brain Decoding with Deep Neural Network Features

Area of Science:

• Neuroscience
• Cognitive Science
• Computer Vision

Background:

• Dreaming is often linked to spontaneous brain activity during sleep.
• Previous studies predicted dreamed objects from sleep brain activity.
• The neural representation of visual features in dreams remains unclear.

Purpose of the Study:

• To investigate how visual image features of dreamed objects are represented in the brain.
• To utilize deep neural network (DNN) models to analyze brain activity during dreaming.

Main Methods:

• Trained DNN decoders with stimulus-induced brain activity and image features.
• Applied decoders to fMRI data collected during dreaming.
• Compared decoded dream features with image database features for object categories.

Main Results:

• Decoded DNN features from dream fMRI data positively correlated with mid- to high-level visual features of object categories.
• Dreamed object categories were identified above chance using decoded features.
• Hierarchical visual feature representations are involved in dreaming.

Conclusions:

• Dreaming appears to recruit hierarchical visual feature representations.
• These representations may contribute to the subjective experience of dreams.
• Findings support a link between brain activity during sleep and waking visual processing.