Studying Healthy Psychosislike Experiences to Improve Illness Prediction.

Importance: Distinguishing delusions and hallucinations from unusual beliefs and experiences has proven challenging.

Observations: The advent of neural network and generative modeling approaches to big data offers a challenge and an opportunity; healthy individuals with unusual beliefs and experiences who are not ill may raise false alarms and serve as adversarial examples to such networks.

Conclusions And Relevance: Explicitly training predictive models with adversarial examples should provide clearer focus on the features most relevant to casehood, which will empower clinical research and ultimately diagnosis and treatment.