Smart sensors for the Internet of Things
Title of the research project
SENSEI - Sensemaking for Scalable IoT Platforms with In-Situ Data-Analytics: A Software-to-Silicon Solution for Energy-Efficient Machine-Learning on Chip
Computer Engineering, Machine Learning, Design Automation
The success of the Internet-of-Things (IoT) passes through the availability of smart sensors that autonomously convert raw data into valuable information. Our research investigated design strategies that enable this feature using low-power computing platforms with embedded artificial intelligence.
Description of the research project
The potential of the Internet-of-Things (IoT) is not the amount of data collected, it’s the ability to make sense of the information inferred from data. Known as “sensemaking”, this process currently involves off-line data analytics based on machine learning (ML) algorithms for data classification. With the exponential growth of mobile devices connected to the IoT, the challenge is how to make IoT sustainable, i.e., how to absorb the continuous streams of data preventing resource congestion. That’s the focus for the Big-Data community, active in pursuing solutions for data-centers and networks.
The SENSEI project lined up with the same urgent issue, yet following a different strategy, that is, act at the source of the problem, i.e., the sensor nodes, where data originate. The vision was that smart sensors with embedded data-analytics can generate data with high information density rather than just high volume.
Within this context, the focus turned over the design of smart objects that can run intensive ML. The challenge was to devise new design strategies for embedded computing platforms able to implement data-analytics, e.g., using Deep Neural Networks, with the tiny energy budget available in portable/mobile applications.
Impact on society
Near-sensors data-analytics is the key to sustain the IoT ecosystem. Objects that can autonomously extract and process information from the physical world might improve the quality of service in several applications, from remote health-care and domotics, to smart transportation, smart manufacturing and smart power delivery. A more efficient flow of information between humans and/or machines allows to quickly predict what is going to happen, foresee the needs, anticipate the requests and provide customized services that fit individual and/or collective specifications. With such level of smartness, IoT can really bring to a better life with affordable costs.
The activities conducted within the project focused on the implementation of optimizations that allow not just to shrink the algorithms for data analysis in terms of arithmetic complexity and use of memory but also to mimic the adaptative mechanisms used by our brain. The experiments conducted on state of art inference models show remarkable savings, both in terms of memory compression and energy efficiency. These results will pave the way toward a widespread adoption of in-situ data-analytics for a sustainable growth of future IoT technologies.
Among the main scientific results we highlight::
- 17 scientific publications and 9 project presentations to international conferences
- New collaborations with Università di Bologna (Italy), ETH Zurich (Swizerland), Brown University (United States) and ICAR- CNR (Italy)
- A patent in preparation
- Presentation to the open event “AI on the edge” in Milano (Italy) in 2018
Short CV of project coordinator
Andrea Calimera is an Assistant Professor (with tenure) of Computer Engineering at Politecnico di Torino. His research interests are mainly focused on Electronic Design Automation, with particular emphasis on CAD tools and circuit-level solutions for resilient, low-power and energy-efficient digital ICs implemented with standard Metal-Oxide-Silicon (MOS) technologies, cutting-edge Silicon-on-Insulator (SOI) devices, and emerging materials for flexible ICs, like Graphene.
Working group @Polito
Valerio Tenace, Post-Doc Fellow
Luca Mocerino, Post-Doc Fellow
Valentino Peluso, Phd Student
Roberto G. Rizzo, Phd Student
Academic partner: Università di Bologna - Italy
Non-academic partner: ST Microelectronics - France
- Budget: 149.895
- Start date: 15/09/2017
- End date: 14/09/2019