Today, distributed systems are increasingly needed to support distributed applications that involve machine learning (ML) operations for both service providers and end users. These operations include training ML models across distributed infrastructures and deploying them over the network for inference. While the training of these models, whether centralized or decentralized, has been extensively studied, the orchestration of inference models across the network remains largely unexamined. To ensure high-quality applications run effectively over these infrastructures, orchestration presents several challenges, including i) the interaction of models with other software components distributed across the network; ii) deploying such components on heterogeneous and computationally limited resources; and iii) adapting orchestration strategies to continuous changes in the data fed into these models. This seminar aims at exploring and discussing these methodological and technological challenges in the design of algorithmic solution for orchestrating AI-based applications within distributed computing systems.
Short Bio. Francescomaria Faticanti is a postdoctoral researcher at ENS de Lyon, France. From January 2022 to August 2023 he was a postdcotoral researcher at Inria Sophia Antipolis, France. He received the PhD in Information and Communication Technology from University of Trento and Fondazione Bruno Kessler (FBK), Italy, in November 2021. He serves as reviewer, and member of the TPC, for several international conferences and journals in the field of Networking and Distributed Systems. His research interests include algorithms for resource allocation in distributed systems, distributed ML and combinatorial optimization.