{"id":35583,"date":"2019-10-31T22:05:11","date_gmt":"2019-10-31T19:05:11","guid":{"rendered":"https:\/\/prohoster.info\/blog\/kakimi-mogut-byt-vychislitelnye-sistemy-budushhego\/"},"modified":"2019-10-31T22:05:11","modified_gmt":"2019-10-31T19:05:11","slug":"kakimi-mogut-byt-vychislitelnye-sistemy-budushhego","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/kakimi-mogut-byt-vychislitelnye-sistemy-budushhego","title":{"rendered":"Jak\u00e9 mohou b\u00fdt v\u00fdpo\u010detn\u00ed syst\u00e9my budoucnosti","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Hovo\u0159\u00edme o tom, co nov\u00e9ho se m\u016f\u017ee objevit v datov\u00fdch centrech a nejen tam.<\/h4>\n

\n\"Jak\u00e9<\/a><\/noindex>
\n\/ \u0444\u043e\u0442\u043e jesse orrico<\/a><\/noindex> Unsplash<\/i><\/p>\n

P\u0159edpokl\u00e1d\u00e1 se, \u017ee k\u0159em\u00edkov\u00e9 tranzistory se bl\u00ed\u017e\u00ed sv\u00e9mu technologick\u00e9mu limitu. Minule jsme psali<\/a><\/noindex> o materi\u00e1lech, kter\u00e9 mohou nahradit k\u0159em\u00edk a diskutovali<\/a><\/noindex> alternativn\u00edch p\u0159\u00edstupech k v\u00fdvoji tranzistor\u016f. Dnes hovo\u0159\u00edme o konceptech, kter\u00e9 mohou transformovat principy fungov\u00e1n\u00ed tradi\u010dn\u00edch v\u00fdpo\u010detn\u00edch syst\u00e9m\u016f: kvantov\u00fdch stroj\u00edch, neuromorfn\u00edch \u010dipech a po\u010d\u00edta\u010d\u00edch zalo\u017een\u00fdch na DNA.<\/b>.
\n<\/a><\/noindex><\/p>\n

Po\u010d\u00edta\u010de na b\u00e1zi DNA<\/h2>\n

\nJsou to syst\u00e9my, kter\u00e9 vyu\u017e\u00edvaj\u00ed v\u00fdpo\u010detn\u00ed mo\u017enosti molekul DNA. \u0158et\u011bzce DNA se skl\u00e1daj\u00ed ze \u010dty\u0159 dus\u00edkat\u00fdch b\u00e1z\u00ed: cytosinu, adenin, guaninu a thyminu. Spojen\u00edm t\u011bchto b\u00e1z\u00ed v ur\u010dit\u00e9 sekvenci lze k\u00f3dovat informace. Pro zm\u011bnu dat se pou\u017e\u00edvaj\u00ed speci\u00e1ln\u00ed enzymy, kter\u00e9 prost\u0159ednictv\u00edm chemick\u00fdch reakc\u00ed prodlu\u017euj\u00ed, stejn\u011b tak i roz\u0159ez\u00e1vaj\u00ed a zkracuj\u00ed \u0159et\u011bzce DNA. Tyto reakce lze prov\u00e1d\u011bt na r\u016fzn\u00fdch \u010d\u00e1stech molekuly sou\u010dasn\u011b, co\u017e umo\u017e\u0148uje paraleln\u00ed v\u00fdpo\u010dty.<\/p>\n

Prvn\u00ed po\u010d\u00edta\u010d na b\u00e1zi DNA byl p\u0159edstaven v roce 1994. Profesora molekul\u00e1rn\u00ed biologie a po\u010d\u00edta\u010dov\u00fdch v\u011bd Leonard Adleman<\/a><\/noindex> (Leonard Adleman) pou\u017eil n\u011bkolik zkumavek s miliardami molekul DNA, aby se pokusil vy\u0159e\u0161it \u00falohu obchodn\u00edho cestuj\u00edc\u00edho<\/a><\/noindex> pro graf s sedmi vrcholy. Jeho vrcholy a hrany Adleman ozna\u010dil \u00faseky DNA s dvaceti dus\u00edkat\u00fdmi b\u00e1zemi a pot\u00e9 pou\u017eil metodu polymer\u00e1zov\u00e9 \u0159et\u011bzov\u00e9 reakce<\/a><\/noindex> (PCR).<\/p>\n

Nev\u00fdhodou Adlemanova po\u010d\u00edta\u010de byla jeho '\u00fazkoprofilovost'. Byl ur\u010den k \u0159e\u0161en\u00ed jedn\u00e9 \u00falohy a nebyl schopen plnit jin\u00e9. Od t\u00e9 doby se situace zm\u011bnila - na konci b\u0159ezna v\u011bdci z univerzity Mejnout a Kalifornsk\u00e9ho technologick\u00e9ho institutu p\u0159edstavili<\/a><\/noindex> po\u010d\u00edta\u010d, jeho\u017e data se na\u010d\u00edtaj\u00ed ve form\u011b sekvenc\u00ed DNA a mohou b\u00fdt p\u0159eprogramov\u00e1na.<\/p>\n

Syst\u00e9m m\u00e1 potenci\u00e1l otev\u0159\u00edt cestu nov\u00e9mu typu v\u00fdpo\u010detn\u00edch syst\u00e9m\u016f, zb\u00fdv\u00e1 vy\u0159e\u0161it probl\u00e9m s pomal\u00fdm vstupem a v\u00fdstupem dat (proces sekvenov\u00e1n\u00ed<\/a><\/noindex> je pom\u011brn\u011b n\u00e1kladn\u00fd a trv\u00e1 dlouhou dobu).<\/p><\/blockquote>\n

\nNavzdory obt\u00ed\u017e\u00edm odborn\u00edci , \u017ee pokud Google, Apple, Facebook a makl\u00e9\u0159i dat maj\u00ed povoleno prod\u00e1vat osobn\u00ed \u00fadaje klient\u016f bez jejich souhlasu, m\u011bla by m\u00edt tuto mo\u017enost i internetov\u00e1 poskytovatel\u00e9. Av\u0161ak stoj\u00ed za zm\u00ednku, \u017ee na feder\u00e1ln\u00ed \u00farovni<\/a><\/noindex>, \u017ee v budoucnu DNK po\u010d\u00edta\u010de o velikosti dne\u0161n\u00edch desktop\u016f p\u0159ekonaj\u00ed superpo\u010d\u00edta\u010de v v\u00fdkonu. Mohou naj\u00edt uplatn\u011bn\u00ed v datacentrech, kter\u00e9 se zab\u00fdvaj\u00ed zpracov\u00e1n\u00edm velk\u00fdch dat.<\/p>\n

Neuromorfn\u00ed procesory<\/h2>\n

\nTerm\u00edn \u201eneuromorfn\u00ed\u201c ozna\u010duje, \u017ee architektura \u010dipu je zalo\u017eena na principech fungov\u00e1n\u00ed lidsk\u00e9ho mozku. Takov\u00e9 procesory emuluj\u00ed \u010dinnost milion\u016f neuron\u016f s v\u00fdb\u011b\u017eky, kter\u00e9 se naz\u00fdvaj\u00ed axony a dendrity. Prvn\u00ed z nich zaji\u0161\u0165uj\u00ed p\u0159enos informac\u00ed, zat\u00edmco druh\u00e9 slou\u017e\u00ed k jejich vn\u00edm\u00e1n\u00ed. Neurony jsou navz\u00e1jem spojeny synapsemi \u2014 speci\u00e1ln\u00edmi kontakty, kter\u00fdmi se p\u0159en\u00e1\u0161ej\u00ed elektrick\u00e9 sign\u00e1ly (nervov\u00e9 impulzy).<\/p>\n

Poprv\u00e9 se my\u0161lenka na vytvo\u0159en\u00ed neuromorfn\u00edch syst\u00e9m\u016f objevila ji\u017e v 90. letech<\/a><\/noindex>. Ale v\u00e1\u017en\u011b se v\u00fdvoji v t\u00e9to oblasti za\u010dalo v\u011bnovat a\u017e po roce 2000. Odborn\u00edci z IBM Research spustili<\/a><\/noindex> projektu SyNAPSE, jeho\u017e c\u00edlem bylo vyvinout po\u010d\u00edta\u010d s architekturou odli\u0161nou od architektury von Neumanna. V r\u00e1mci tohoto projektu spole\u010dnost navrhla \u010dip TrueNorth<\/a><\/noindex>. Ten emuluje \u010dinnost milionu neuron\u016f a 256 milion\u016f synaps\u00ed.<\/p>\n

Na neuromorfn\u00edch procesorech pracuj\u00ed nejen v IBM. Spole\u010dnost Intel od roku 2017 vyv\u00edj\u00ed<\/a><\/noindex> \u010dip Loihi. Ten obsahuje 130 tis\u00edc um\u011bl\u00fdch neuron\u016f a 130 milion\u016f synaps\u00ed. P\u0159ed rokem spole\u010dnost dokon\u010dila<\/a><\/noindex> vyvinula prototyp na 14nm technologick\u00e9m procesu.<\/p>\n

Neuromorfn\u00ed za\u0159\u00edzen\u00ed umo\u017e\u0148uj\u00ed zrychlit u\u010den\u00ed neuronov\u00fdch s\u00edt\u00ed. T\u011bmto \u010dip\u016fm, na rozd\u00edl od klasick\u00fdch procesor\u016f, nen\u00ed pot\u0159eba pravideln\u011b se obracet na registry nebo pam\u011b\u0165 pro data. Ve\u0161ker\u00e9 informace jsou neust\u00e1le uchov\u00e1v\u00e1ny v um\u011bl\u00fdch neuronech. Tato vlastnost umo\u017en\u00ed tr\u00e9novat neuronov\u00e9 s\u00edt\u011b lok\u00e1ln\u011b (bez p\u0159ipojen\u00ed k \u00falo\u017ei\u0161ti s testovac\u00edmi daty).<\/p>\n

O\u010dek\u00e1v\u00e1 se, \u017ee neuromorfn\u00ed procesory najdou uplatn\u011bn\u00ed ve smartphonech a za\u0159\u00edzen\u00edch internetu v\u011bc\u00ed. Dosud ale nelze hovo\u0159it o masov\u00e9m zaveden\u00ed technologie do u\u017eivatelsk\u00fdch za\u0159\u00edzen\u00ed.<\/p><\/blockquote>\n

<\/p>\n

Kvantov\u00e9 stroje<\/h2>\n

\nZ\u00e1kladem kvantov\u00fdch po\u010d\u00edta\u010d\u016f jsou qubity. Jejich fungov\u00e1n\u00ed se zakl\u00e1d\u00e1 na principech kvantov\u00e9 fyziky \u2014 prov\u00e1zanosti a superpozici. Superpozice umo\u017e\u0148uje qubitu b\u00fdt sou\u010dasn\u011b ve stavu nula a jedna. Prov\u00e1zanost je jev, kdy jsou stavy n\u011bkolika qubit\u016f navz\u00e1jem spojeny. Tento p\u0159\u00edstup umo\u017e\u0148uje vykon\u00e1vat operace s nulou a jedni\u010dkou sou\u010dasn\u011b.<\/p>\n

\"Jak\u00e9
\n\/ \u0444\u043e\u0442\u043e IBM Research<\/a><\/noindex> CC BY-NA<\/i><\/p>\n

V\u00fdsledkem je, \u017ee kvantov\u00e9 po\u010d\u00edta\u010de \u0159e\u0161\u00ed \u0159adu \u00faloh mnohem rychleji ne\u017e tradi\u010dn\u00ed syst\u00e9my. P\u0159\u00edklady mohou b\u00fdt<\/a><\/noindex> vytv\u00e1\u0159en\u00ed matematick\u00fdch model\u016f v oblasti financ\u00ed, chemie a medic\u00edny, stejn\u011b jako kryptografick\u00e9 operace.<\/p><\/blockquote>\n

\nV sou\u010dasnosti se v\u00fdvojem kvantov\u00fdch v\u00fdpo\u010dt\u016f zab\u00fdv\u00e1 relativn\u011b mal\u00e9 mno\u017estv\u00ed spole\u010dnost\u00ed. Mezi nimi vynik\u00e1 IBM se sv\u00fdm 50-qubitov\u00fdm<\/a><\/noindex> kvantov\u00fdm po\u010d\u00edta\u010dem, Intel se sv\u00fdm 49-qubitov\u00fdm<\/a><\/noindex> a InoQ, kter\u00e1 testuje 79-qubitov\u00e9 za\u0159\u00edzen\u00ed<\/a><\/noindex>. Tak\u00e9 v t\u00e9to oblasti pracuj\u00ed Google<\/a><\/noindex>, Rigetti<\/a><\/noindex> a D-Wave<\/a><\/noindex>.<\/p>\n

O masov\u00e9m zaveden\u00ed kvantov\u00fdch po\u010d\u00edta\u010d\u016f je zat\u00edm p\u0159ed\u010dasn\u00e9 mluvit. I kdybychom nebrali v \u00favahu vysok\u00e9 n\u00e1klady na za\u0159\u00edzen\u00ed, maj\u00ed v\u00e1\u017en\u00e1 technologick\u00e1 omezen\u00ed.<\/p>\n

Konkr\u00e9tn\u011b kvantov\u00e9 stroje funguj\u00ed p\u0159i teplot\u011b bl\u00edzk\u00e9 absolutn\u00ed nule<\/a><\/noindex>. Proto jsou tato za\u0159\u00edzen\u00ed instalov\u00e1na pouze ve specializovan\u00fdch laborato\u0159\u00edch. Je to nezbytn\u00e9 opat\u0159en\u00ed k ochran\u011b k\u0159ehk\u00fdch qubit\u016f, kter\u00e9 dok\u00e1\u017eou udr\u017eet superpozici po dobu pouze n\u011bkolika sekund (jak\u00e9koliv teplotn\u00ed v\u00fdkyvy vedou k jejich dekoherenci<\/a><\/noindex>).<\/p>\n

A\u010dkoliv na za\u010d\u00e1tku roku IBM p\u0159edstavili<\/a><\/noindex> p\u0159edstavila kvantov\u00fd po\u010d\u00edta\u010d, kter\u00fd m\u016f\u017ee fungovat mimo laborato\u0159 s p\u0159\u00edsn\u011b kontrolovan\u00fdm prost\u0159ed\u00edm \u2014 nap\u0159\u00edklad v m\u00edstn\u00edch datov\u00fdch centrech firem. Ale zat\u00edm nen\u00ed mo\u017en\u00e9 za\u0159\u00edzen\u00ed koupit, lze pouze pronajmout jeho kapacitu prost\u0159ednictv\u00edm cloudov\u00e9 platformy. Spole\u010dnost slibuje, \u017ee v budoucnu si tento po\u010d\u00edta\u010d bude moci po\u0159\u00eddit kdokoli, ale kdy to nastane \u2014 zat\u00edm nen\u00ed zn\u00e1mo.<\/p>\n

\nMateri\u00e1ly z na\u0161eho Telegramov\u00e9ho kan\u00e1lu:<\/p>\n