deep neural net using fiber optics & filters for weights. should be possible to, e.g., project an image and get instantaneous classification
-
Show this thread
-
Kyle McDonald Retweeted Kyle McDonald
my first drawing of this ideahttps://twitter.com/kcimc/status/776494200305115136?s=20 …
Kyle McDonald added,
2 replies 2 retweets 10 likesShow this thread -
now there's some exciting research in this direction that is only visible light, using some kind of caustic refraction? i need to study it more https://www.newscientist.com/article/2208975-ai-made-from-a-sheet-of-glass-can-recognise-numbers-just-by-looking/ …pic.twitter.com/0G5MPQwfjG
1 reply 1 retweet 24 likesShow this thread -
looking at their paper, it seems they had the exact same idea ("filters act like weights + ReLU for light"). but my biggest issue was the lack of negative numbers ("negative light/negative filters").. not clear if they got around this? https://www.osapublishing.org/prj/fulltext.cfm?uri=prj-7-8-823&id=415059 …pic.twitter.com/W6idO2NiKf
2 replies 0 retweets 10 likesShow this thread -
if this all seems vaguely familiar you might be thinking of this 2018 work using terahertz light, not exactly designed to function outside a research labhttps://www.the-scientist.com/news-opinion/ai-object-recognition-system-operates-at-speed-of-light--64569 …
2 replies 2 retweets 14 likesShow this thread -
-
Replying to @fchollet
great reference. here's a link for anyone else curious to learn more about their systemhttps://www.osapublishing.org/ao/abstract.cfm?uri=ao-21-2-274 …
1 reply 1 retweet 1 like
I was thinking of this onehttps://www.osapublishing.org/josa/abstract.cfm?uri=josa-72-5-556 …
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.