Keynote

Decentralized Intelligence: towards collaborative and sustainable learning

Speaker: Paolo Dini

Abstract

The deployment of sensors, mobile services and industrial applications using artificial intelligence (AI) is growing fast and becoming pervasive. The architectures to provide such computing services are challenging the centralized processing system typically used by AI. In fact, modern information sources are distributed and generate data far from the cloud computing centers. Streaming data to remote centers results to be highly inefficient and energy consuming.
Recently, edge intelligence have emerged, proposing a distributed and pervasive computing system where data can be processed directly or closer to the sources. In such a way, communication overhead, latency, memory requirements, energy consumption are reduced and privacy concerns are mitigated.
However, highly accurate AI models (e.g., deep learning) are complex, need huge amount of data and high computing capabilities that could not be (always) present locally in a single resource-constrained device. Therefore, decentralized intelligence at the edge requires collaborative learning paradigms to assist training and inference, maintain high accuracy and good generalization properties, and lower energy consumption of AI models.
Current research efforts aim to design flexible / scalable architectures and algorithms enabling fast convergence, robustness against failures and attacks, and high energy efficiency. Knowledge transfer, federated and continual learning represent key enablers of such decentralized intelligence.
In this keynote I present the current state of collaborative and sustainable AI ecosystem, the interplay among the different enablers, how we can use them in different application scenarios, and what challenges we need to face in the future.

Bio

Paolo Dini is a Senior Researcher within the Centre Tecnologic de Telecomunicacions de Catalunya (CTTC) in Spain, where he coordinates the activities of the Sustainable Artificial Intelligence research unit. His research interests include sustainable computing and networking, distributed optimization and machine learning, multi-agent systems and decision making processes, data mining for cyber-physical systems. His research activity is documented in more than 90 peer-reviewed scientific journal and international conference papers. He received two awards from the Cisco Silicon Valley Foundation for his research on mobility management in heterogeneous mobile networks (2008 and 2011). He has been continuously participating in several research projects and collaborates with multiple research centers and universities around the world