Dr. Alexandru Dobrila
Affiliation: Hivenet
Venue: TII Yas Auditorium
20th January 2025,1:30PM - 2:30PM (GST)
| Title: | Distributed AI workload : from research to real world implementation |
| Abstract: | This seminar examines distributed artificial intelligence as an alternative to centralized cloud-based computation. We present HiveNet’s architecture for cooperative distributed compute nodes enabling scalable and resilient AI workloads. The discussion is grounded in research themes from Cupseli (Inria partnership), including frugality (Memory, Data & Communication Efficiency), security & privacy and volatility & fault tolerance. Practical examples will illustrate distributed training and inference across distributed nodes. The talk concludes with open research questions relevant to large-scale, decentralized AI systems. |
| Bio: | After a PhD in distributed systems focusing on scheduling and fault tolerance I’ve spent 13 years in the industry. At Stellantis, I worked on storage systems and networking for hyperscale datacenters and from 2018 onward I led infrastructure teams responsible for network operations across two datacenters, as well as globally distributed software teams developing connected services for vehicles. In 2024 I’ve joined Hivenet as Director of Research in charge with bringing the research close to the real word challenges that we face with distributed computing and storage on a very large scale. |