Neuroadaptive modelling for generating images matching perceptual categories

Area of Science:

• Neuroscience
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Brain-computer interfaces (BCIs) traditionally support simple commands.
• Inferring complex intentions or adapting outputs from brain signals remains a challenge.
• Existing BCIs lack the ability to generate novel, complex information tailored to user intent.

Purpose of the Study:

• To introduce neuroadaptive generative modeling, a novel BCI paradigm.
• To demonstrate the capability of generating new information that aligns with user intentions using brain signals.
• To validate the approach in generating photorealistic human face images based on perceptual categories.

Main Methods:

• Utilized electroencephalography (EEG) signals as feedback to adapt a generative model.
• Employed a generative adversarial network (GAN) to create new images.
• Participants focused on specific perceptual categories (e.g., age) while viewing faces, with EEG data guiding model adaptation.

Main Results:

• The neuroadaptive model successfully generated images matching participants' intended perceptual categories.
• A double-blind evaluation confirmed the model's ability to produce relevant outputs.
• EEG feedback effectively updated the generative model to reflect user intentions.

Conclusions:

• Neuroadaptive generative modeling enables the creation of novel content aligned with human perceptual categories.
• This approach represents a significant advancement in brain-based creative augmentation.
• It opens new possibilities for human-computer collaboration in generating complex information.