Speaker: Dr. Manuele Rusci (KU Leuven)
Abstract:
Bringing intelligence inside battery-powered tiny sensors that populate the extreme-edge of the network is extremely challenging because of the severe computation, memory and power constraints of the system and its processing units. In the first part of the talk reviews the main quantization and hardware-software innovations that we adopted to port audio and visual deep inference tasks on low-power RISC-V microcontrollers. Then, It will challenge the widespread train-once-deploy-everywhere paradigm by describing our effort to bring customization capabilities on-device.
By taking an audio classification problem as a case-study, the talk will show our recent results to learn the deep learning sensor functions on our microcontroller-powered systems using data collected in the field to eventually gain a personalized smart audio sensor.
Short Bio
Dr. Manuele Rusci received his Ph.D. in electronic engineering from the University of Bologna in 2018. He is currently a Marie Curie postdoc fellow at KU Leuven. Before that, he was with GreenWaves Technologies and Università di Bologna. His research work focuses on edge machine learning on ultra-low-power processors.
Organazier: Dr. Alessandro Capotondi