NTT thiab Tsev Kawm Ntawv Qib Siab Tokyo Tsim Ntiaj Teb Thawj Qhov Kev Ntsuas AI Siv Cov Txheej Txheem Ua Haujlwm los ntawm tib neeg lub hlwb

Kev sib koom tes txhawb nqa cov kev siv tswv yim ntawm kev siv hluav taws xob tsawg, kev kub ceev AI raws li kev siv kho qhov muag

TOKYO-(BUSINESS WIRE)-#TechforGood-NTT Corporation (Thawj Tswj Hwm thiab CEO: Akira Shimada, “NTT”) thiab cov Tsev kawm ntawv ntawm Tokyo (Bunkyo-ku, Tokyo, Thawj Tswj Hwm: Teruo Fujii) tau tsim ib qho kev kawm tshiab algorithm tshwm sim los ntawm kev ua cov ntaub ntawv ntawm lub hlwb uas tsim nyog rau ntau txheej txheej neural tes hauj lwm (DNN) siv analog ua haujlwm. Qhov kev tawg no yuav ua rau txo qis ntawm kev siv hluav taws xob thiab lub sijhawm suav rau AI. Cov txiaj ntsig ntawm qhov kev txhim kho no tau luam tawm hauv phau ntawv xov xwm British scientific Xwm Communications hnub tim 26 lub XNUMX hlis^th.

Cov kws tshawb fawb tau ua tiav lub ntiaj teb thawj zaug ua qauv qhia ntawm kev ua tau zoo kho qhov muag DNN kev kawm los ntawm kev siv cov algorithm rau DNN uas siv optical analog xam, uas yuav tsum ua kom muaj kev kub ceev, tsis tshua muaj zog tshuab kev kawm. Tsis tas li ntawd, lawv tau ua tiav lub ntiaj teb qhov kev ua tau zoo tshaj plaws ntawm ntau txheej txheej neural network uas siv cov kev ua haujlwm analog.

Yav dhau los, high-load kawm suav suav tau ua los ntawm cov lej digital, tab sis qhov txiaj ntsig no ua pov thawj tias nws muaj peev xwm txhim kho qhov ua tau zoo ntawm kev kawm los ntawm kev siv cov lej analog. Nyob rau hauv Deep Neural Network (DNN) thev naus laus zis, qhov rov tshwm sim neural network hu ua kev sib sib zog nqus reservoir suav yog xam los ntawm kev xav tias qhov muag pom tau zoo li cov neuron thiab lub nplhaib tsis muaj kab hluav taws xob ua lub neural network nrog kev sib txuas rov qab. Los ntawm re-inputting lub tso zis teeb liab mus rau tib lub optical Circuit Court, lub network yog artificially tob.

DNN thev naus laus zis ua kom muaj kev txawj ntse ntse (AI) xws li tshuab txhais lus, kev tsav tsheb tsis raug cai thiab neeg hlau. Tam sim no, lub zog thiab kev suav lub sijhawm xav tau yog nce ntawm tus nqi uas tshaj qhov kev loj hlob ntawm kev ua haujlwm ntawm cov khoos phis tawj digital. DNN thev naus laus zis, uas siv cov teeb liab analog suav nrog (kev ua haujlwm analog), xav tias yuav yog ib txoj hauv kev ua kom muaj txiaj ntsig zoo thiab kev ua haujlwm siab ceev zoo ib yam li cov neural network ntawm lub hlwb. Kev sib koom tes ntawm NTT thiab University of Tokyo tau tsim cov txheej txheem tshiab tsim nyog rau kev ua haujlwm analog DNN uas tsis xav tias kev nkag siab ntawm cov kev kawm tsis suav nrog hauv DNN.

Txoj kev npaj tau kawm los ntawm kev hloov cov kev kawm tsis raws li cov txheej txheem kawg ntawm lub network thiab cov nonlinear random transformation ntawm qhov yuam kev ntawm qhov xav tau tso zis teeb liab (yuam kev teeb liab). Qhov kev suav no ua rau nws yooj yim dua los siv cov lej analog hauv cov khoom xws li optical circuits. Nws kuj tseem tuaj yeem siv tsis tau tsuas yog ua qauv rau kev siv lub cev, tab sis kuj yog cov qauv txiav-ntug siv hauv kev siv xws li tshuab txhais lus thiab ntau yam AI qauv, suav nrog DNN qauv. Qhov kev tshawb fawb no xav tias yuav pab txhawb rau kev daws teeb meem tshwm sim cuam tshuam nrog AI suav, suav nrog kev siv hluav taws xob thiab lub sijhawm suav ntxiv.

Ntxiv nrog rau kev tshuaj xyuas qhov siv tau ntawm cov txheej txheem uas tau thov hauv daim ntawv no rau cov teeb meem tshwj xeeb, NTT tseem yuav txhawb nqa qhov loj thiab me me kev sib koom ua ke ntawm kho qhov muag kho qhov muag, aiming los tsim kom muaj kev kub ceev, qis zog kho qhov muag platform rau yav tom ntej kho qhov muag. tes hauj lwm.

Kev them nyiaj yug rau kev tshawb fawb no:

JST/CREST txhawb nqa ib feem ntawm cov txiaj ntsig kev tshawb fawb no.

Phau ntawv Magazine:

magazine: Xwm Communications (Online version: Kaum Ob Hlis 26)

Lub Npe Lub Npe: Kev Kawm Txog Lub Cev Rau Lub Cev Nrog Kev Kawm Txuj Ci Biologically Inspired: Gradient-Free Approach for Physical Hardware

Sau: Mitsumasa Nakajima, Katsuma Inoue, Kenji Tanaka, Yasuo Kuniyoshi, Toshikazu Hashimoto, thiab Kohei Nakajima

Kev piav qhia ntawm Terminology:

1. Optical circuit: Ib lub voj voog uas silicon lossis quartz optical waveguides tau koom ua ke rau hauv silicon wafer siv hluav taws xob tsim hluav taws xob. Hauv kev sib txuas lus, kev sib txuas thiab kev sib koom ua ke ntawm kev sib txuas lus kho qhov muag tau ua los ntawm kev cuam tshuam ntawm qhov muag, wavelength multiplexing / demultiplexing, thiab lwm yam.
2. Backpropagation (BP) method: Feem ntau siv kev kawm algorithm hauv kev kawm tob. Gradients ntawm qhov hnyav (tsis suav) hauv lub network tau txais thaum nthuav tawm cov teeb liab yuam kev rov qab, thiab qhov hnyav tau hloov kho kom qhov yuam kev me dua. Txij li cov txheej txheem rov qab yuav tsum tau hloov pauv ntawm qhov hnyav matrix ntawm lub network qauv thiab kev sib txawv tsis sib txawv, nws yog qhov nyuaj rau kev siv ntawm analog circuits, suav nrog lub hlwb ntawm cov kab mob muaj sia.
3. Analog suav: Ib lub computer uas qhia txog qhov muaj nqis tiag siv lub cev kom muaj nuj nqis xws li kev siv zog thiab theem ntawm lub teeb thiab cov kev taw qhia thiab kev siv ntawm cov hlau nplaum sib nqus thiab ua kev suav los ntawm kev hloov cov khoom ntawm lub cev raws li cov cai ntawm physics.
4. Direct feedback alignment (DFA) method: Ib txoj hauv kev ntawm pseudo-xoo qhov teeb meem yuam kev ntawm txhua txheej los ntawm kev ua ib qho kev hloov pauv tsis zoo ntawm cov teeb liab yuam kev ntawm txheej kawg. Txij li thaum nws tsis xav tau cov ntaub ntawv sib txawv ntawm cov qauv network thiab tuaj yeem suav tsuas yog los ntawm kev sib piv random transformation, nws yog sib xws nrog analog xam.
5. Reservoir xam: Ib hom kev sib txuas ntawm neural nrog cov kev sib txuas rov ua dua hauv cov txheej txheem zais. Nws yog tus cwj pwm los ntawm kev txhim kho kev sib txuas hauv cov txheej txheem nruab nrab hu ua txheej reservoir. Nyob rau hauv kev sib sib zog nqus reservoir xam, cov ntaub ntawv ua yog ua los ntawm kev sib txuas cov reservoir txheej hauv ntau txheej.

NTT thiab NTT lub logo yog cov cim lag luam lossis cov cim lag luam ntawm NIPPON TELEGRAPH THIAB TELEPHONE CORPORATION thiab / lossis nws cov koom tes. Tag nrho lwm cov npe khoom siv yog cov cim lag luam ntawm lawv cov tswv. © 2023 NIPPON TELEGRAPH THIAB TELEPHONE CORPORATION

Muaj mob

Stephen Russell

Kev sib txuas lus Wireside®

Rau NTT

+ 1-804-362-7484

[email tiv thaiv]