Professor Abu-Mostafa has focused on three medical applications of AI plus a fundamental AI project. A new, major medical AI project is starting this year.

1. Ultrasound technology to measure the absolute arterial blood pressure non-invasively without the need for any calibration. This research has already led to a device that can produce the full waveform of blood pressure accurately, not just the systolic and diastolic values. AI was employed to make the measurement accurate, robust, and adaptive.
   
2. Using AI to detect Congestive Heart Failure and other conditions based on videos of the so-called sniff test in Medicine. This is part of our ongoing collaboration with UCSF.
   
3. Detecting mini clots in the bloodstream non-invasively based on a Caltech patent. The results would lead to prediction of impending stroke before it actually takes place, thus allowing preventive and mitigative intervention.
   
4. A new project to create a foundation model for cardiology is starting this year based on a unique data set that is obtained from a major US hospital, one of the biggest medical data sets in an AI lab (100+ Terabytes). The data will be hosted in our lab at Caltech, and all the legal formalities, IRB protocols, and data anonymization processes are already in place. The data pipeline has already started.

In all of these projects, AI/ML is an essential technology to deal with the realities of the human body that do not fit idealistic assumptions or lend themselves to conclusive mathematical analysis.

In addition, we have revived interest in Learning from Hints in Neural Networks. This encompasses a set of techniques to incorporate prior knowledge in the learning process as virtual data, in addition to the regular training data. Most recent results are using generative models to create more effective virtual data, which resulted in uniform improvement in performance.