Challenges and Opportunities for Composable AI-Integrated Applications at the Digital Continuum: Keynote
İlkay Altintaş · 2020
Summary form only given. The complete presentation was not made available for publication as part of the conference proceedings. Cyberinfrastructure is everywhere in diverse forms. From IoT to extreme scale computing, data and computing has never been as distributed with potential for real-time integration via fast networking and container management. The growth of new processors over the last decade including GPUs, FPGAs, and edge accelerators opened the way to a diverse set of applications using machine learning on top of distributed nontraditional hardware. The common theme to these applications, mostly composed of artificial intelligence (AI) workloads, is their need to run in specialized environments for reasons such as on demand or 24×7 nature of the tasks they are performing, and difficulties regarding their portability, latency, privacy and performance optimization. In many data-driven scientific applications there is a need for integration of these AI-workloads with traditional high-throughput computing (HTC) or high-performance computing (HPC) tasks for AI-integrated science. This talk will discuss example AI-integrated applications, describe some of the new systems that enabled these applications, and overview our recent research to enable composable applications including an application development methodology, intelligent middleware and workflow composition.