Dimitrios Nikolopoulos
Bio:
Dimitrios Nikolopoulos is the John W. Hancock Professor of Engineering, Professor in the Department of Computer Science, and Professor by courtesy in the Department of Electrical and Computer Engineering at Virginia Tech. He conducts research in Computer Systems and High-Performance Computing. Some of his prior research has been adopted in the OpenMP parallel programming standard, RedHat Enterprise Linux, and several commercial system software products, including parallelizing compilers, in-memory databases, and data stream processing systems, as well as numerous experimental operating systems and programming languages for multicore and multithreaded servers. More details about his research are available in his DBLP, Google Scholar, ORCID, or Scopus profiles. Dimitrios teaches courses in Computer Organization, Systems and Networking Capstone, Multiprocessor Programming, and Warehouse-Scale Computing.
Dimitrios is a Royal Society Wolfson Research Fellow, a Distinguished Member of the ACM, and a Distinguished Contributor of the IEEE Computer Society. In addition, he has received numerous prestigious personal investigator and faculty awards (including the NSF CAREER, DOE Early Career, IBM Faculty Award, Cisco Faculty Awards, SFI-DEL Investigator Award, Marie Curie Individual Fellowship), nine best paper awards from premier conferences (including SC, PPoPP, IPDPS, CCGRID, and DATE, as well as best paper award nominations from ICS, SC, and HPDC) and multiple awards for editorial excellence (notably awards from IEEE’s Transactions on Parallel and Distributed Systems and Elsevier’s Sustainable Computing Informatics and Systems). He has chaired the Program Committee of the ACM International Conference on Supercomputing in 2022 and 2023. His research has been supported with over $35 million of external funding (personal share of a total of $102 million in competitive funding) from NSF, DOE, the European Commission, EPSRC, SFI, and the private sector.
Dimitrios served in major University leadership roles, including Head of the School of EEECS at QUB, Director of the ECIT Global Innovation Institute, and Director of the QUB HPC Working Group. During his tenure in leadership roles, he received on behalf of the School two awards for efforts to improve diversity and inclusion (Athena Swan Silver, Investors in People Silver).
Dimitrios currently leads the PEARL (Performance and Resiliency) lab at Virginia Tech. He is also Associate Director of the Stack@CS Center for Computer Systems, with a remit to advance cross-stack collaborative research against significant challenges in computing systems.
Abstracts:
Cloud-native AI Services on Edge
The notions of Edge Computing and Edge AI are somewhat innocuous: they both hinge on a straightforward path to improve AI and data analytics services by reducing latency and processing data closer to their source. Unfortunately, Edge Computing substrates lack the powerful hardware and software infrastructures of their Cloud Computing counterparts. Due to the highly heterogeneous and resource-constrained nature of edge devices and servers, established virtualization, containerization, autoscaling, and deployment methods that have worked well on the Cloud do not immediately apply to Edge Computing. This talk explores new software and hardware methods to provide Cloud-native AI services on Edge Computing devices using an application and system co-design approach. The talk specifically explores methods to develop and deploy extremely lightweight serverless computing frameworks on Edge devices and performance scalable training of and inference from AI models on these frameworks.
Transprecision Computing Paradigms
This talk explores dynamic transpression computing, a new paradigm that employs algorithmic, operating system, and hardware techniques to adaptively tune the precision of computations to improve performance and energy efficiency without compromising correctness or reliability. We will explore two research directions: One is the systematic use of variable and dynamic precision tuning in machine learning algorithms. The second is the architecting systems with heterogeneous reliability domains, where different system components can tolerate different degrees of errors and approximation with a combination of targeted hardware monitoring and operating system resilience mechanisms.
Links:
Website
LinkedIn
Facebook
