AI Prediction System for ER Chest Pain Cases, the brainchild of collaboration of the AI Center and ER, has the amazingly high accuracy of 90% reported on SCI, a venerable international medical journal.

Commonly seen on ER patients, chest pain might be the result of cardiovascular diseases. To offer a quick, effective solution for high-accuracy diagnostics, the AI Center introduced Supervised Machine Learning techniques that analyzed papers and experiences from senior doctors to elect 14 characteristics for constructing AI prediction models. The model is capable of making medical predictions on an individual basis for better treatment quality.

Today, the ER Prediction Systems not only diagnose chest pains, but also influenza of seniors, Acute Pancreatitis, Hyperglycemia, Dengue Fever and Pneumonia; and the systems are being deployed in various departments such as Anesthesiology and ICU as well.

To the IT Department, waves after waves of difficulties are a part of the daily development works; and even the most outstanding staff members are still limited by the same 24 hours in a day.

They spent most of their time maintaining the online medical systems rather than studying unfamiliar tools such as Docker or Kubernetes environments, optimizing GPU resources or debugging models in response to error messages.

Moreover, it takes days to build a model, and even more time in communications between the AI Center and IT Department itself.

In its early stages, the Academy deployed dozens of servers and a few hundred graphic cards for students to get hands-on experience on AI model constructions; however it’s an equally complex task to adequately distribute such resources among hundreds of people while maintaining stability of the entire system.

Since most students were new to AI development, the Academy called for an intuitive, friendly learning environment without the need to waste time on tedious configurations; and the system needed to be able to allow easy sharing of teaching materials, such as weekly sample files and AI model training data sets.

To meet the requirements, the Academy needed to build a multi-user, manageable computing platform in a relatively short period.

PrimeHub's intuitive interface allows teaching assistants to set-up student accounts, class materials and computing resource allocations in a snap. As AI libraries and class files evolve, assistants can also assign the environments for different library versions, as well as renew and test files before distributing them to the students.

With the self-service PrimeHub platform, students can activate their own development environments and obtain training files and data sets to start leaning on the interactive Jupyter Notebook. Thanks to the "container" model employed by PrimeHub, the completely isolated AI learning environments are able to accommodate hundreds of students. Furthermore, resources of the idle accounts can be reclaimed by the system for optimal utilization.

As the growing Academy established three sites in Hsinchu, Taichung and Southern Taiwan, InfuseAI helps each site to share the successful experience by moving the entire infrastructure to the centralized cloud for better overall resource efficiency and minimized interruption.