A big trend in AI is the transition from cloud to edge computing. Instead of AI devices doing their computation remotely via an internet connection, they\u2019re increasingly handling things locally, with algorithms working directly on-device. Benefits of this approach can include faster results, greater security, and more flexibility. But how far can you push this model?<\/p>\n\n\n\n

Seattle-based startup Xnor is certainly right at the bleeding-edge. This week the company unveiled a\u00a0prototype AI camera\u00a0that runs entirely off solar power \u2014 no battery or external power source required. The camera has a piddling 320 x 320 resolution, an FGPA chip to handle processing, and is loaded with a state of the art object recognition algorithm.<\/p>\n\n\n\n

You could, theoretically, stick a device like this anywhere outdoors and have it transmit data back to you indefinitely. It\u2019s compatible with a\u00a0few\u00a0different\u00a0low-energy wireless communication protocols (because Wi-Fi drains too much battery) which let it send information over tens of kilometers. And, says Xnor, if you fit it with a battery, it can store enough power during the day to keep it running during low-sunlight hours and at night.<\/p>\n\n\n\n

\u201cWe\u2019re investigating several use-cases for these devices right now,\u201d Mohammad Rastegari, Xnor\u2019s CTO, tells\u00a0The Verge<\/em>. \u201cFrom large-scale civilian projects, to monitoring inside the cabins of autonomous cars, to attaching cameras to drones.\u201d<\/p>\n\n\n\n

Xnor has a strong background in this sort of AI miniaturization. It was spun off from the Allen Institute for Artificial Intelligence\u00a0back in 2017\u00a0on the back of a proprietary method for creating super-efficient machine learning systems. (Key to this technology is a type of logic circuit known as an\u00a0XNOR gate, hence the name.) It\u2019s also proved the utility of its software by running them on low-power, low-compute devices\u00a0like the Raspberry Pi Zero.<\/p>\n\n\n\n

This solar-powered AI camera is not yet ready to sell though. Although it\u2019s entirely self-contained, there are some constraints on its operation. For example, the amount of frames it can process each second depends on how much sun it\u2019s getting. Xnor says that on a sunny day it runs at 32fps, but this can be compensated for with bigger solar cells.<\/p>\n\n\n\n

What\u2019s clear is that devices like this are only going to become more common in the future. They\u2019re relatively cheap (Xnor\u2019s model costs $10), and more convenient for operators. And, as the photos and videos they take never leave the device, they\u2019re potentially more private.<\/p>\n\n\n\n

But a question remains: are we comfortable with a world full of AI eyes that are always watching? It\u2019s a problem we\u2019re already grappling with in the context of\u00a0smart CCTV, and as Xnor\u2019s work shows, the technology is only going to get smaller and more unobtrusive.<\/p>\n\n\n\n

From: https:\/\/www.theverge.com\/circuitbreaker\/2019\/2\/15\/18225972\/tiny-solar-powered-ai-camera-xnor<\/p>\n","protected":false},"excerpt":{"rendered":"

A big trend in AI is the transition from cloud to edge computing. Instead of AI devices doing their computation remotely via an internet connection, they\u2019re increasingly handling things locally, with algorithms working directly on-device. Benefits of this approach can include faster results, greater security, and more flexibility. But how far can you push this…
Read more<\/a><\/p>\n","protected":false},"author":5,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"nf_dc_page":"","footnotes":""},"categories":[3],"tags":[],"class_list":["post-79","post","type-post","status-publish","format-standard","hentry","category-solarnews"],"yoast_head":"\n