Machine learning within a sensor: how and when?
Machine learning models such as Minerva, PaLM, GPT-4 are dramatically increasing the memory wall and their complexity in a “more is better” fashion. However these models let serious doubts to be raised: how much energy is it required to train them? how can they scale across four billion android users? Is there any limit to model hyper-parametrization? how to avoid data contamination? what is the proper training data vs parameter ration? Is this approach sustainable for the future of the planet?
For experts on tiny machine learning, the way to go is the opposite: the less is more by looking to deployable machine learning solutions on resource restricted devices. Indeed, since 2019 TinyML Foundation and MLCommons created a vibrant community focused on developing low power devices with open benchmarks mainly concentrating on micro-controllers and neural processing units. Unfortunately, sensors were poorly considered as execution targets because of their extreme constraints.
Therefore, this talk will focus on machine learning computing aimed for sensors with built in capability to infer AI workloads to push forward the tiny concept to the extreme low boundary both in term of power consumption, die area and accuracy. Two examples will be elaborated: inertial and pressure sensor with two different computing paradigms.
About the Speaker
Danilo Pau graduated in 1992 at Politecnico di Milano, Italy. On 1991, he joined SGS-THOMSON (now ST Microelectronics) as interns on Advanced Multimedia Architectures, and he worked on memory reduced HDMAC HW design. Then MPEG2 video memory reduction. Next, on video coding, transcoding, embedded (Khronos) 2/3D graphics, and (ISO CDVS and CDVA) computer vision.
Currently, his work focuses on developing solutions for tiny machine learning including tools (STM32Cube.AI, Stellar.AI, SPC-Studio.AI). Since 2019 Danilo is an IEEE Fellow; AAIA Fellow on 2023; he served as Industry Ambassador coordinator for IEEE Region 8 South Europe, was vice-chairman of the “Intelligent Cyber-Physical Systems” Task Force within IEEE CIS, was IEEE R8 AfI member in charge of internship initiative. Today he is a Member of the Machine Learning, Deep Learning and AI in the CE (MDA) Technical Stream Committee CESoc. He was AE of IEEE TNNLS. He wrote the IEEE Milestone on Multiple Silicon Technologies on a chip, 1985 which was ratified by IEEE BoD in 2021 and IEEE Milestone on MPEG Multimedia Integrated Circuits, 1984-1993 which was ratified in 2022. He served as TPC member to TinyML Symposium, Summit, EMEA forum and is the chair of the TinyML on Device Learning working group. He serves as 2023 IEEE Computer Society Fellow Evaluating Committee Members.
With 78 and 68 respectively European and US application patents, 167 publications, 113 ISO/IEC/MPEG authored documents and 80 invited talks/seminars at various Universities and Conferences, Danilo's favorite activity remains supervising undergraduate students, MSc engineers and PhDs.