Guang R. Gao ACM Fellow and IEEE Fellow Endowed Distinguished Professor University of Delaware
The field of parallel computing (systems, and beyond) and the field of AI (including machine learning) have one aspect in common: both have experienced dramatic changes from a period of flourish, followed by a long dormant period and decline, then coming back again to another period of flourish with fantastic recovery and growth again.
In this talk, we will review the situation from the angle of parallel computation – models and systems. We articulate why we should start from the angle of a parallel Turing machine model. The program execution model of the proposed PTM is presented that addresses the weakness of some existing work on parallel computation models due to the well known “problems of threads” --- program understandability, performance predictability, optimizability and composability.
We will comment on the challenges (to computer systems) from the recent flourish in the field of AI, machine learning and brain-like (or brain-inspired) computing and advocate the need of our community to consider longer term solutions in a wider scope. To this end, we conclude the talk with a few open problems and conjectures.
Guang R. Gao is a computer scientist and a Professor of Electrical and Computer Engineering at University Delaware. Gao is a founder and Chief Scientist of ETI (ET International Inc., Newark, Delaware).
Gao has devoted a majority of his research and academic careers in carrying on the legacy of the MIT dataflow model research that he has participated and contributed during his Ph.D project under Prof. Jack. B. Dennis and Arvind. The legacy of Gao’s own research is to show that the fundamental value of the dataflow model of computation can be effectively explored and efficiently realized – and the superiority dataflow can be demonstrated even in parallel computer systems that are made of classical microprocessors with von-Neumann architectures and other components. To this end, Gao has led a series of parallel architecture and system projects where various aspects of dataflow models are improved and integrated in the design and implementation – ranging from innovations in programming paradigms, architecture features, system software technology, including novel program optimization and runtime system techniques. Gao’s contribution is recognized by receiving the ACM Fellow and IEEE Fellow in 2007. Through 35+ years persistent R&D and entrepreneur effort – Gao and his students have propelled the impact of dataflow model inspired technology of computation beyond their laboratory in the US to other parts of the world including EU and Asia.
The legacy of Gao’s work has also been associated with his entrepreneur effort applying his dataflow R&D results for real world applications through ETI (ET International Inc.). A unique achievement of Gao’s team at ETI is its critical role in the now legendary supercomputing system project – funded by DoD and IBM - known as IBM Cyclops-64 Supercomputer (https://en.wikipedia.org/wiki/Cyclops64). The success on Cyclops64 supercomputer is recognized by the selection ETI as winner of a Supercomputing disruptive technology award in 2007 (http://www.etinternational.com/index.php/projects/) as one of the largest supercomputer build on many-core chip technology at that time and in production use in real world.