From the edge to the cloud: Exploring AI inference (yes, including generative AI) across the computing continuum

This talk explores recent research on AI inference across the computing continuum, from the edge to the cloud, with a focus on the challenges and opportunities brought by hardware and software heterogeneity, as well as automation requirements. Roberto will present insights into collaborative, edge-centric generative AI inference, including efforts to benchmark language models on constrained devices and to route queries efficiently across distributed nodes. He will also discuss recent work on automating the lifecycle of extreme edge devices using LLMs, demonstrating how these models can support code generation, adaptation, and deployment under

tight resource constraints.