ETP4HPC Webinar - In-memory computing for AI applications

11am-noon CET

To feed our thoughts on the evolution of HPC/AI systems, we invited Robert Haas from IBM to share his insights on in-memory computing for AI applications.

The "memory wall" challenge in traditional digital systems has been aggravated by the explosive growth in data-hungry applications. A radically new approach is analog in-memory computing, where certain computational tasks are performed in an analog manner in the memory itself. Given the precision of analog matrix-vector operations that can be performed on arrays of analog devices, analog in-memory computing is particularly well suited when approximate solutions are acceptable, such as many AI tasks. Furthermore, new computing paradigms such as hyperdimensional computing are lending themselves extraordinarily well to the inherent variability of analog in-memory computing. This talk will illustrate these exciting new research directions with efforts undertaken at IBM Research Europe in Zurich.

Dr. Robert Haas leads one of the technical research departments at IBM Research Europe based in Zurich. His department delivers critical system hardware and software innovation that powers the cloud, AI, server and storage platforms of IBM, as well as leads the race towards in-memory computing. He has an M.Sc. in Communications Systems from EPFL and the Eurécom Institute (Sophia-Antipolis, France), a Ph.D. from ETHZ, and an MBA from Warwick Business School (UK). He has contributed to over 60 patents and several open standards in networking and security.

Agenda:

- Introduction, Michael Malms, ETP4HPC (5 min)

- In-memory Computing for AI Applications, Robert Haas, IBM (30 min)

- ETP4HPC’s webinar plans for 2021, Maria Toth, ETP4HPC (5 min)

- Q&As (20min)