ePerceptive Gives Energy-Harvesting Sensors High-Performance, Best-Effort Deep Learning Capabilities

Researchers from Nokia Bell Labs, in collaboration with the Universities of Washington and Cambridge, TU Delft, Newcastle University, have released details of a system dubbed "ePerceptive" and designed to offer "reactive embedded intelligence" for battery-free sensor systems. "For long, we have studied tiny energy harvesters to liberate sensors from batteries. With remarkable progress in embedded deep learning, we are now reimagining these sensors as intelligent compute nodes," the researchers explain in their paper's draft. "Naturally, we are approaching a crossroad where sensor intelligence is meeting energy autonomy enabling maintenance-free swarm intelligence and unleashing a plethora of applications ranging from precision agriculture to ubiquitous asset tracking to infrastructure monitoring. One of the critical challenges, however, is to adapt intelligence fidelity in response to available energy to maximize the overall system availability." "To this end, we present the design and implementation of ePerceptive: a novel framework for best-effort embedded intelligence, i.e., inference fidelity varies in proportion to the instantaneous energy supplied. ePerceptive operates on two core principles. First, it enables training a single deep neural network (DNN) to operate on multiple input resolutions without compromising accuracy or incurring memory overhead. Second, it modifies a DNN architecture by injecting multiple exits to guarantee valid, albeit lower-fidelity inferences in the event of energy interruption. The combination of these techniques offers a smooth adaptation between inference latency and recognition accuracy while matching the computational load to the available power budget." To prove the concept, the team built physical implementations of battery-free camera and microphone sensors developed around Texas Instruments' MSP430 microcontroller and using off-the-shelf radio frequency and solar energy harvesting systems. By automatically adjusting to the available power, rather than designing for a worst-case baseline, the team found that they could boost the inference throughput of the system by up to 80 percent for a maximum drop in accuracy of under six percent. The team's work was presented at the 18th ACM Conference on Embedded Networked Sensor Systems (SenSys 2020), and is available under open-access terms from co-author Fahim Kawsar's website . The ePerceptive system is able to tailor its operation to interruptible, unreliable harvested power. (📷: Montanari et al)

Read more here: https://www.hackster.io/news/eperceptive-gives-energy-harvesting-sensors-high-performance-best-effort-deep-learning-capabilities-a0bbf85bf7b4

Content Attribution

This content was originally published by Gareth Halfacree at Hackster News - Hackster.io, and is syndicated here via their RSS feed. You can read the original post over there.

%d bloggers like this: