« zurück

Kolloquium | Edge AI hardware reliability assessment and enhancement techniques

10. Juni 2025 | 11 Uhr s.t.

Dr. Mahdi Taheri von der Brandenburgische Technische Universität Cottbus-Senftenberg hält am

10. Juni 2025 um
11 Uhr s.t.
im AGRA-Raum (MZH 4380)

ein Informatik-Kolloquium zum Thema "Edge AI hardware reliability assessment and enhancement techniques".

Wir freuen uns, Dich bei uns willkommen zu heißen.

With advancements in technology, the use cases of Artificial Intelligence (AI) on edge devices are increasing. This shift brings new reliability demands to hardware platforms, especially in safety- and mission-critical systems. A major challenge in ensuring the reliability of edge-AI chips stems from the constrained computational and energy resources of edge devices. This talk will address methods for assessing and enhancing soft-error reliability in Deep Learning accelerators. It will also explore the potential of approximate computing and delve into the architectural considerations of systolic arrays, dataflow-based designs, and industrial-grade AI accelerators implemented on ASICs and FPGAs.

Biografie

Mahdi received his B.Sc. degree in electrical engineering in 2017 and his M.SC. degree in electrical engineering in 2021. He received his Ph.D. at Tallinn University of Technology in Jan. 2025, and since then, he has been a postdoctoral researcher at BTU Cottbus. His research interests include reliability and fault tolerance design, approximate computing, neural networks (NNs), hardware accelerators, and applications of field-programmable gate arrays (FPGAs).

Mehr Informationen gibt es bei Rolf Drechsler.