{"id":35293,"date":"2019-10-31T22:03:28","date_gmt":"2019-10-31T19:03:28","guid":{"rendered":"https:\/\/prohoster.info\/blog\/lopnul-li-puzyr-mashinnogo-obucheniya-ili-nachalo-novoj-zari\/"},"modified":"2019-10-31T22:03:28","modified_gmt":"2019-10-31T19:03:28","slug":"lopnul-li-puzyr-mashinnogo-obucheniya-ili-nachalo-novoj-zari","status":"publish","type":"post","link":"https:\/\/prohoster.info\/ro\/blog\/news\/lopnul-li-puzyr-mashinnogo-obucheniya-ili-nachalo-novoj-zari","title":{"rendered":"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"<p>Recentemente a ap\u0103rut <noindex><a rel=\"nofollow\" href=\"https:\/\/www.getrevue.co\/profile\/peterzhegin\/issues\/ai-investment-activity-trends-of-2018-issue-8-150825?fbclid=IwAR0tU3l4WSotv7pQpYm8PmyKgVbgUgfMLeue_IiV78lXXApH-cy9EcG2kDc\">articol<\/a><\/noindex>, care arat\u0103 bine tendin\u021bele \u00een \u00eenv\u0103\u021barea automat\u0103 din ultimii ani. Pe scurt: num\u0103rul startup-urilor \u00een domeniul \u00eenv\u0103\u021b\u0103rii automate a sc\u0103zut brusc \u00een ultimii doi ani.<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/68b58feab2da46b7bb6f412e088313c1.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\nEi bine. S\u0103 discut\u0103m \u201edac\u0103 bula a explodat\u201d, \u201ecum s\u0103 tr\u0103im mai departe\u201d \u0219i s\u0103 vorbim despre de unde a ap\u0103rut aceast\u0103 situa\u021bie.<br \/>\n<noindex><a rel=\"nofollow\" name=\"habracut\"><\/a><\/noindex><br \/>\nLa \u00eenceput, s\u0103 discut\u0103m despre ce a fost booster-ul acestei curbe. De unde a ap\u0103rut. Probabil c\u0103 toat\u0103 lumea \u00ee\u0219i aminte\u0219te <noindex><a rel=\"nofollow\" href=\"https:\/\/papers.nips.cc\/paper\/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf\">victoria<\/a><\/noindex> \u00een \u00eenv\u0103\u021barea automat\u0103 \u00een 2012 la competi\u021bia ImageNet. Pentru c\u0103 acesta a fost primul eveniment global! Dar \u00een realitate nu este a\u0219a. De asemenea, cre\u0219terea curbei \u00eencepe cu c\u00e2\u021biva ani mai devreme. A\u0219 \u00eemp\u0103r\u021bi-o \u00een c\u00e2teva momente.<\/p>\n<ol>\n<li>Anul 2008 este apari\u021bia termenului \u201edate mari\u201d. Produsele reale au \u00eenceput <noindex><a rel=\"nofollow\" href=\"https:\/\/ru.wikipedia.org\/wiki\/%D0%91%D0%BE%D0%BB%D1%8C%D1%88%D0%B8%D0%B5_%D0%B4%D0%B0%D0%BD%D0%BD%D1%8B%D0%B5\">s\u0103 apar\u0103<\/a><\/noindex> \u00eencep\u00e2nd cu anul 2010. Datele mari sunt str\u00e2ns legate de \u00eenv\u0103\u021barea automat\u0103. F\u0103r\u0103 date mari, nu este posibil\u0103 func\u021bionarea stabil\u0103 a algoritmilor care existau pe atunci. \u0218i aceasta nu sunt re\u021bele neuronale. P\u00e2n\u0103 \u00een 2012, re\u021belele neuronale erau domeniul unei minorit\u0103\u021bi marginale. \u00cens\u0103 atunci au \u00eenceput s\u0103 func\u021bioneze algoritmi complet diferi\u021bi, care existau deja de ani sau chiar decenii: <noindex><a rel=\"nofollow\" href=\"https:\/\/ru.wikipedia.org\/wiki\/%D0%9C%D0%B5%D1%82%D0%BE%D0%B4_%D0%BE%D0%BF%D0%BE%D1%80%D0%BD%D1%8B%D1%85_%D0%B2%D0%B5%D0%BA%D1%82%D0%BE%D1%80%D0%BE%D0%B2\">SVM<\/a><\/noindex>(1963, 1993), <noindex><a rel=\"nofollow\" href=\"https:\/\/ru.wikipedia.org\/wiki\/Random_forest\">Random Forest<\/a><\/noindex> (1995), <noindex>AdaBoost<\/noindex> (2003),\u2026 Startup-urile acelor ani se leag\u0103 \u00een principal de procesarea automat\u0103 a datelor structurate: case de marcat, utilizatori, publicitate, multe altele.\n<p>Derivata acestei prime unde este un set de cadre de lucru, cum ar fi XGBoost, CatBoost, LightGBM etc.\n<\/li>\n<li>\u00centre 2011-2012 <noindex><a rel=\"nofollow\" href=\"https:\/\/en.wikipedia.org\/wiki\/Convolutional_neural_network\">re\u021bele neuronale convolu\u021bionale<\/a><\/noindex> au c\u00e2\u0219tigat o serie de competi\u021bii de recunoa\u0219tere a imaginilor. Utilizarea lor real\u0103 s-a \u00eent\u00e2rziat pu\u021bin. A\u0219 spune c\u0103 startup-urile \u0219i solu\u021biile \u00een\u021belese pe larg au \u00eenceput s\u0103 apar\u0103 din 2014. Au fost necesari doi ani pentru a procesa c\u0103 re\u021belele neuronale func\u021bioneaz\u0103, pentru a crea cadre de lucru convenabile care puteau fi instalate \u0219i rulate \u00eentr-un timp rezonabil, pentru a dezvolta metode care s\u0103 stabilizeze \u0219i s\u0103 accelereze timpul de convergen\u021b\u0103.\n<p>Re\u021belele convolu\u021bionale au permis rezolvarea problemelor de viziune computerizat\u0103: clasificarea imaginilor \u0219i a obiectelor din imagini, detectarea obiectelor, recunoa\u0219terea obiectelor \u0219i a oamenilor, \u00eembun\u0103t\u0103\u021birea imaginilor etc.<\/li>\n<li>Anii 2015-2017. Boom-ul algoritmilor \u0219i proiectelor legate de re\u021belele recurente sau echivalentele lor (LSTM, GRU, TransformerNet etc.). Au ap\u0103rut algoritmi bine func\u021bionali pentru \u201evorbire-\u00een-text\u201d, sisteme de traducere automat\u0103. Par\u021bial, acestea se bazeaz\u0103 pe re\u021bele convolu\u021bionale pentru extragerea caracteristicilor de baz\u0103. Par\u021bial, pe faptul c\u0103 am \u00eenv\u0103\u021bat s\u0103 colect\u0103m seturi de date real mari \u0219i de calitate bun\u0103. <\/li>\n<\/ol>\n<p>\n<img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/d4b3a1dd2483ef1300862f5f61db645a.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\n\u201eBula a explodat? Hype-ul este supra\u00eenc\u0103lzit? Au murit ca blockchainul?\u201d<br \/>\nAsta-i sigur! M\u00e2ine, Siri va \u00eenceta s\u0103 func\u021bioneze pe telefonul vostru, iar poim\u00e2ine Tesla nu va distinge o curb\u0103 de un cangur.<\/p>\n<p>Re\u021belele neuronale deja func\u021bioneaz\u0103. Ele se afl\u0103 \u00een zeci de dispozitive. Ele permit cu adev\u0103rat c\u00e2\u0219tiguri, schimb\u0103 pia\u021ba \u0219i mediul \u00eenconjur\u0103tor. Hype-ul arat\u0103 un pic altfel:<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/a2b271c8eb1cf54fe59389d10cf8e17e.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nPur \u0219i simplu, re\u021belele neuronale au \u00eencetat s\u0103 mai fie ceva nou. Da, mul\u021bi oameni au a\u0219tept\u0103ri exagerate. Dar un num\u0103r mare de companii au \u00eenv\u0103\u021bat s\u0103 aplice re\u021bele neuronale \u0219i s\u0103 dezvolte produse pe baza lor. Re\u021belele neuronale ofer\u0103 func\u021bionalit\u0103\u021bi noi, permit reducerea locurilor de munc\u0103, sc\u0103derea pre\u021bului serviciilor:<\/p>\n<ul>\n<li>Companiile de produc\u021bie integreaz\u0103 algoritmi pentru analiza defectelor pe linia de asamblare. <\/li>\n<li>\u00centreprinderile de cre\u0219tere a animalelor cump\u0103r\u0103 sisteme pentru monitorizarea vacilor. <\/li>\n<li> Combain\u0103 automate. <\/li>\n<li>Centre de apel automatizate.<\/li>\n<li>Filtre \u00een SnapChat. (cel pu\u021bin un lucru folositor!)<\/li>\n<\/ul>\n<p>\n\u00cens\u0103, cel mai important \u0219i cel mai pu\u021bin evident: \u201eNu mai sunt idei noi, sau ele nu vor aduce capital instantaneu\u201d. Re\u021belele neuronale au rezolvat zeci de probleme. \u0218i vor rezolva \u0219i mai multe. Toate ideile evidente care au existat au generat o mul\u021bime de startup-uri. Dar tot ce era la suprafa\u021b\u0103 a fost deja adunat. \u00cen ultimii doi ani, nu am \u00eent\u00e2lnit nicio idee nou\u0103 pentru aplicarea re\u021belelor neuronale. Nici o abordare nou\u0103 (ok, sunt c\u00e2teva complica\u021bii cu GAN-urile).<\/p>\n<p>\u0218i fiecare startup urm\u0103tor devine tot mai complex. Acesta nu mai necesit\u0103 doar doi b\u0103ie\u021bi care s\u0103 antreneze re\u021beaua neuronal\u0103 pe date deschise. Acesta necesit\u0103 programatori, servere, o echip\u0103 de eticheta\u021bi, suport complex etc.<\/p>\n<p>Ca rezultat \u2014 num\u0103rul startup-urilor scade. Dar produc\u021bia cre\u0219te. Trebuie s\u0103 ad\u0103uga\u021bi recunoa\u0219terea numerelor de ma\u0219in\u0103? Pe pia\u021b\u0103 sunt sute de speciali\u0219ti cu experien\u021b\u0103 relevant\u0103. Pute\u021bi angaja \u0219i, \u00een c\u00e2teva luni, angajatul dvs. va crea sistemul. Sau s\u0103 cump\u0103ra\u021bi unul gata f\u0103cut. Dar s\u0103 crea\u021bi un nou startup?.. Nebunie!<\/p>\n<p>Este nevoie s\u0103 realiz\u0103m un sistem de urm\u0103rire a vizitatorilor - de ce s\u0103 pl\u0103tim pentru o mul\u021bime de licen\u021be c\u00e2nd putem construi unul personalizat pentru afacerea noastr\u0103 \u00een 3-4 luni. <\/p>\n<p>Acum, re\u021belele neuronale urmeaz\u0103 acela\u0219i parcurs pe care l-au avut zeci de alte tehnologii. <\/p>\n<p>\u00ce\u021bi aminte\u0219ti cum s-a schimbat no\u021biunea de \u00abdezvoltator de site-uri\u00bb din 1995 p\u00e2n\u0103 acum? \u00cen timp ce pia\u021ba nu este saturat\u0103 de speciali\u0219ti. Sunt foarte pu\u021bini profesioni\u0219ti. Dar pot sus\u021bine c\u0103, \u00een 5-10 ani, nu va mai exista o distinc\u021bie semnificativ\u0103 \u00eentre un programator Java \u0219i un dezvoltator de re\u021bele neuronale. Vor fi destui speciali\u0219ti \u00een ambele domenii pe pia\u021b\u0103.<\/p>\n<p>Pur \u0219i simplu va exista o clas\u0103 de sarcini pentru care se vor utiliza re\u021bele neuronale. Dac\u0103 apare o sarcin\u0103 - angajezi un specialist.<\/p>\n<p><b>\u201e\u0218i ce urmeaz\u0103? Unde este inteligen\u021ba artificial\u0103 promis\u0103?\u201d<\/b><\/p>\n<p>Aici exist\u0103 o mic\u0103, dar interesant\u0103 neclaritate :)<\/p>\n<p>Tehnologiile de ast\u0103zi, se pare, nu ne conduc spre inteligen\u021ba artificial\u0103. Ideile \u0219i noutatea lor s-au epuizat \u00een mare parte. Hai s\u0103 discut\u0103m despre ceea ce men\u021bine actualul nivel de dezvoltare.<\/p>\n<h3>Limit\u0103ri<\/h3>\n<p>\nS\u0103 \u00eencepem cu ma\u0219inile autonome. Pare destul de clar c\u0103 realizarea de automobile complet autonome cu tehnologiile actuale este posibil\u0103. Dar \u00een c\u00e2t timp se va \u00eent\u00e2mpla asta - nu este clar. Tesla crede c\u0103 se va \u00eent\u00e2mpla \u00een c\u00e2\u021biva ani - <\/p>\n<p><center><div class=\"youtube-placeholder\" data-id=\"Ucp0TTmvqOE\" onclick=\"loadVideo(this)\">\r\n        <img decoding=\"async\" src=\"https:\/\/img.youtube.com\/vi\/Ucp0TTmvqOE\/hqdefault.jpg\" alt=\"Reda\u021bi video\" loading=\"lazy\" width=\"480\" height=\"360\" style=\"width:100%;height:auto;\">\r\n        <div class=\"play-button\"><\/div>\r\n    <\/div><\/center><br \/>\nExist\u0103 mul\u021bi al\u021bii <noindex><a rel=\"nofollow\" href=\"https:\/\/beth.technology\/truths-autonomous-vehicles\/\">speciali\u0219ti<\/a><\/noindex>, care estimeaz\u0103 c\u0103 va dura \u00eentre 5 \u0219i 10 ani. <\/p>\n<p>Din punctul meu de vedere, cel mai probabil, \u00een aproximativ 15 ani infrastructura ora\u0219elor se va schimba astfel \u00eenc\u00e2t apari\u021bia automobilelor autonome va deveni inevitabil\u0103, va deveni o extensie a acesteia. Dar asta nu poate fi considerat inteligen\u021b\u0103. Tesla modern\u0103 este un conveior foarte complex pentru filtrarea, c\u0103utarea \u0219i re\u00eenv\u0103\u021barea datelor. Acestea sunt reguli - reguli - reguli, colectare de date \u0219i filtre aplicate acestora (iat\u0103 <noindex><a rel=\"nofollow\" href=\"http:\/\/cv-blog.ru\/?p=279\">aici<\/a><\/noindex> am scris mai \u00een detaliu despre asta, sau po\u021bi viziona de la <noindex><a rel=\"nofollow\" href=\"https:\/\/www.youtube.com\/watch?time_continue=7614&amp;v=Ucp0TTmvqOE\">aceast\u0103<\/a><\/noindex> marcaj).<\/p>\n<h3>Prima problem\u0103<\/h3>\n<p>\n\u0218i aici vedem <b>prima problem\u0103 fundamental\u0103<\/b>. Big Data. This is precisely what has generated the current wave of neural networks and machine learning. Today, to accomplish something complex and automated, a vast amount of data is required. Not just a lot, but an overwhelming amount. We need automated algorithms for collection, labeling, and usage. If we want the machine to see trucks against the sun, we first need to gather a sufficient number of them. If we want the machine not to go crazy from a bicycle attached to the trunk, we need more samples.<\/p>\n<p>Moreover, one example is not enough. Hundreds? Thousands? <\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/923ef975804234f1b3dcbfee0143f2b4.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<\/p>\n<h3>The second problem<\/h3>\n<p>\n<b>The second problem <\/b> \u2014 visualizing what our neural network has understood. This is a very non-trivial task. Until now, few understand how to visualize this. These articles are quite recent, and they offer just a few examples, however distant:<br \/>\n<noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/ods\/blog\/453788\/\">Vizualizare<\/a><\/noindex> the fixation on textures. It effectively shows what the neural network tends to focus on + what it perceives as initial information.<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/0eab41c3aec71dc2b04794559dd4a651.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<noindex><a rel=\"nofollow\" href=\"http:\/\/jalammar.github.io\/visualizing-neural-machine-translation-mechanics-of-seq2seq-models-with-attention\/\">Vizualizare<\/a><\/noindex> attention during <noindex><a rel=\"nofollow\" href=\"http:\/\/www.wildml.com\/2016\/01\/attention-and-memory-in-deep-learning-and-nlp\/\">translations<\/a><\/noindex>. In fact, attention can often be used precisely to demonstrate what caused such a reaction from the network. I have come across such tools for debugging and for product solutions. There are many articles on this topic. However, the more complex the data, the harder it is to achieve stable visualization.<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/e0c370724115f602e5bd35b20b56f6eb.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nAnd yes, the old reliable set of 'look at what the network has inside in <noindex><a rel=\"nofollow\" href=\"https:\/\/towardsdatascience.com\/how-to-visualize-convolutional-features-in-40-lines-of-code-70b7d87b0030\">filters<\/a><\/noindex>'. These images were popular about 3-4 years ago, but everyone quickly realized that while the images are pretty, they lack meaning.<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/87ace90924d5b900f9382ecc6ceef6d0.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nI haven\u2019t mentioned dozens of other gimmicks, methods, hacks, and research on how to display the internals of the network. Do these tools work? Do they help quickly understand what the problem is and debug the network? Do they extract those last few percentages? Well, it\u2019s about the same:<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/6da71648c300ee3bc8673d08287b77e3.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nYou can check any competition on Kaggle. And the descriptions of how people arrive at final solutions. We stacked 100-500-800 million models, and it worked!<\/p>\n<p>Of course, I\u2019m exaggerating. But these approaches do not provide quick and direct answers.<\/p>\n<p>With sufficient experience, by testing various options, one can make a verdict on why your system made such a decision. However, correcting the system's behavior will be challenging. It\u2019s possible to implement a workaround, adjust the threshold, add a dataset, or take another backend network.<\/p>\n<h3>The third problem<\/h3>\n<p>\n<b>The third fundamental problem <\/b> \u2014 re\u021belele \u00eenva\u021b\u0103 nu logica, ci statistica. Statistically this <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/417405\/\">fa\u021b\u0103<\/a><\/noindex>:<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/def14bbc2f40e4e26656a2d8032b09c1.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nLogic nu foarte asem\u0103n\u0103toare. Re\u021belele neuronale nu \u00eenva\u021b\u0103 ceva complex, dac\u0103 nu sunt for\u021bate. Ele \u00eenva\u021b\u0103 \u00eentotdeauna cele mai simple caracteristici. Exist\u0103 ochi, nas, cap? Atunci este o fa\u021b\u0103! Ori adu un exemplu \u00een care ochii s\u0103 nu \u00eensemne fa\u021b\u0103. \u0218i din nou \u2014 milioane de exemple.<\/p>\n<h3>Exist\u0103 mult spa\u021biu \u00een partea de jos<\/h3>\n<p>\nA\u0219 spune c\u0103 aceste trei probleme globale limiteaz\u0103 \u00een prezent dezvoltarea re\u021belelor neuronale \u0219i a \u00eenv\u0103\u021b\u0103rii automate. Iar \u00een locurile unde aceste probleme nu limiteaz\u0103 \u2014 deja sunt utilizate activ.<\/p>\n<p><b>Este acesta sf\u00e2r\u0219itul? Re\u021belele neuronale s-au oprit?<\/b><\/p>\n<p>Nu se \u0219tie. Dar, desigur, toat\u0103 lumea sper\u0103 c\u0103 nu. <\/p>\n<p>Exist\u0103 multe abord\u0103ri \u0219i direc\u021bii pentru a rezolva acele probleme fundamentale pe care le-am men\u021bionat mai sus. Dar p\u00e2n\u0103 acum niciuna dintre aceste abord\u0103ri nu a reu\u0219it s\u0103 creeze ceva cu adev\u0103rat nou, s\u0103 rezolve ceva ce nu a fost rezolvat p\u00e2n\u0103 acum. P\u00e2n\u0103 acum toate proiectele fundamentale se realizeaz\u0103 pe baza unor abord\u0103ri stabile (Tesla), sau r\u0103m\u00e2n proiecte de testare ale institu\u021biilor sau corpora\u021biilor (Google Brain, OpenAI).<\/p>\n<p>Dac\u0103 vorbim \u00een termeni foarte generali, principala direc\u021bie este crearea unei reprezent\u0103ri de nivel \u00eenalt a datelor de intrare. \u00centr-un fel, \u201ememorie\u201d. Cel mai simplu exemplu de memorie sunt diferitele \u201eEmbedding\u201d \u2014 reprezent\u0103ri ale imaginilor. De exemplu, toate sistemele de recunoa\u0219tere facial\u0103. Re\u021beaua \u00eenva\u021b\u0103 s\u0103 ob\u021bin\u0103 dintr-o fa\u021b\u0103 o reprezentare stabil\u0103 care nu depinde de unghi, iluminare, rezolu\u021bie. Practic, re\u021beaua minimizeaz\u0103 metrica \u201efe\u021be diferite \u2014 departe\u201d \u0219i \u201eidentice \u2014 aproape\u201d.<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/5e9d871fe096b7a77dbe11a6e315c5e4.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nPentru un astfel de \u00eenv\u0103\u021b\u0103m\u00e2nt sunt necesare zeci \u0219i sute de mii de exemple. Dar rezultatul aduce unele elemente de \u201eOne-shot Learning\u201d. Acum nu avem nevoie de sute de fe\u021be pentru a memoriza o persoan\u0103. Doar o fa\u021b\u0103, \u0219i totul \u2014 noi <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/davidsandberg\/facenet\">recunoa\u0219tem<\/a><\/noindex>!<br \/>\nDar exist\u0103 o problem\u0103 ... Re\u021beaua poate \u00eenv\u0103\u021ba doar obiecte destul de simple. C\u00e2nd \u00eencerc\u0103m s\u0103 distincem nu fe\u021bele, ci, de exemplu, \u201eoamenii dup\u0103 \u00eembr\u0103c\u0103minte\u201d (sarcin\u0103 <noindex><a rel=\"nofollow\" href=\"https:\/\/medium.com\/@alitech_2017\/reforming-person-re-identification-with-local-convolutional-neural-networks-17148f11f17b\">Re-identification<\/a><\/noindex>) \u2014 calitatea scade cu mult. \u0218i re\u021beaua nu mai poate \u00eenv\u0103\u021ba schimb\u0103ri de unghiuri suficient de evidente.<\/p>\n<p>Dar a \u00eenv\u0103\u021ba din milioane de exemple \u2014 de asemenea, nu este exact o distrac\u021bie pl\u0103cut\u0103. <\/p>\n<p>Exist\u0103 lucr\u0103ri care reduc semnificativ selec\u021bia. De exemplu, una dintre primele lucr\u0103ri despre <b>OneShot Learning<\/b> <noindex><a rel=\"nofollow\" href=\"https:\/\/arxiv.org\/pdf\/1605.06065v1.pdf\">de la Google<\/a><\/noindex>:<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/da83c6f290248f2bcf963c3053a26688.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nSunt multe astfel de lucr\u0103ri, de exemplu <noindex><a rel=\"nofollow\" href=\"https:\/\/pdfs.semanticscholar.org\/d1c4\/c4c7989102e85b5248cebfcb0cb000c3b837.pdf\">1<\/a><\/noindex> sau <noindex><a rel=\"nofollow\" href=\"https:\/\/www.cs.cmu.edu\/~rsalakhu\/papers\/oneshot1.pdf\">2<\/a><\/noindex> sau <noindex><a rel=\"nofollow\" href=\"http:\/\/www.robots.ox.ac.uk\/~tvg\/publications\/2018\/0431.pdf\">3<\/a><\/noindex>.<\/p>\n<p>Un dezavantaj este c\u0103, de obicei, \u00eenv\u0103\u021barea func\u021bioneaz\u0103 bine pentru exemple simple, precum cele din setul MNIST. Dar, atunci c\u00e2nd se trece la sarcini mai complexe, este nevoie de o baz\u0103 mare, un model de obiecte sau de un oarecare tip de magie.<br \/>\n\u00cen general, lucr\u0103rile despre \u00eenv\u0103\u021barea One-Shot sunt un subiect foarte interesant. G\u0103se\u0219ti multe idei. Dar, \u00een mare parte, cele dou\u0103 probleme pe care le-am enumerat (pre\u00eenv\u0103\u021barea pe un set de date uria\u0219 \/ instabilitatea pe date complexe) \u00eengreuneaz\u0103 \u00eenv\u0103\u021barea.<\/p>\n<p>Pe de alt\u0103 parte, la tema Embedding se potrivesc GAN-urile - re\u021bele generative antagoniste. Cu siguran\u021b\u0103 ai citit o mul\u021bime de articole pe aceast\u0103 tem\u0103 pe Habr.<noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/ods\/blog\/340154\/\">1<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/itsumma\/blog\/447896\/\">2<\/a><\/noindex>,<noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/ods\/blog\/322514\/\">3<\/a><\/noindex>)<br \/>\nO caracteristic\u0103 a GAN-urilor este formarea unui anumit spa\u021biu intern de st\u0103ri (de fapt, acela\u0219i Embedding), care permite generarea unei imagini. Acestea pot fi <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/shaoanlu\/faceswap-GAN\">fe\u021be<\/a><\/noindex>, pot fi <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/sergeytulyakov\/mocogan\">ac\u021biuni<\/a><\/noindex>. <\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/8c25c375dae3559b6895d6c4eb3f6cfd.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nProblema GAN-urilor este c\u0103, cu c\u00e2t obiectul generat este mai complex, cu at\u00e2t este mai greu s\u0103 fie descris \u00een logica \u201egenerator-discriminator\u201d. Ca urmare, din aplica\u021biile reale ale GAN-urilor, care sunt cunoscute, avem doar DeepFake, care manipuleaz\u0103 din nou imaginile fe\u021belor (pentru care exist\u0103 o baz\u0103 uria\u0219\u0103).<\/p>\n<p>Am \u00eent\u00e2lnit foarte pu\u021bine aplica\u021bii utile. De obicei, sunt doar gadgeturi cu ad\u0103ugarea de detalii imaginilor.<\/p>\n<p>\u0218i din nou, nimeni nu are o \u00een\u021belegere clar\u0103 a modului \u00een care acest lucru ne va ajuta s\u0103 ne \u00eendrept\u0103m spre un viitor luminos. Reprezentarea logicii\/spa\u021biului \u00eentr-o re\u021bea neuronal\u0103 este bun\u0103. Dar avem nevoie de un num\u0103r uria\u0219 de exemple, nu \u0219tim cum percepe re\u021beaua neuronal\u0103 aceste aspecte, nu \u0219tim cum s\u0103 facem re\u021beaua neuronal\u0103 s\u0103 re\u021bin\u0103 o reprezentare cu adev\u0103rat complex\u0103.<\/p>\n<p><b>\u00cenv\u0103\u021barea prin \u00eent\u0103rire<\/b> este o abordare complet diferit\u0103. Cu siguran\u021b\u0103 \u00ee\u021bi aminte\u0219ti cum Google a \u00eenvins pe toat\u0103 lumea la Go. Victoriile recente \u00een Starcraft \u0219i Dota. Dar aici lucrurile nu sunt deloc at\u00e2t de roz sau promi\u021b\u0103toare. Cel mai bine vorbe\u0219te despre RL \u0219i dificult\u0103\u021bile sale <noindex><a rel=\"nofollow\" href=\"https:\/\/www.alexirpan.com\/2018\/02\/14\/rl-hard.html\">acest articol<\/a><\/noindex>.<\/p>\n<p>Dac\u0103 ar fi s\u0103 rezum\u0103m ce a scris autorul:<\/p>\n<ul>\n<li>Modelele din cutie nu sunt potrivite \/ func\u021bioneaz\u0103 prost \u00een majoritatea cazurilor.<\/li>\n<li>Sarcinile practice sunt mai u\u0219or de rezolvat cu alte metode. Boston Dynamics nu folose\u0219te RL din cauza dificult\u0103\u021bii \/ imprevizibilit\u0103\u021bii \/ complexit\u0103\u021bii calculelor.<\/li>\n<li>Pentru ca RL s\u0103 func\u021bioneze, este nevoie de o func\u021bie complex\u0103. Adesea, este dificil de creat \/ scris.<\/li>\n<li>Este greu s\u0103 antrenezi modele. Trebuie s\u0103 petreci o mul\u021bime de timp pentru a le aduce \u00een form\u0103 \u0219i a ie\u0219i din optimumele locale.<\/li>\n<li>Ca urmare, este dificil s\u0103 replici modelul, instabilitatea modelului la cele mai mici modific\u0103ri.<\/li>\n<li>Adesea se overfit la anumite regularit\u0103\u021bi aleatoare, chiar \u0219i p\u00e2n\u0103 la generatorul de numere aleatoare.<\/li>\n<\/ul>\n<p>\nPunctul cheie este c\u0103 RL (\u00eenv\u0103\u021barea prin \u00eent\u0103rire) \u00eenc\u0103 nu func\u021bioneaz\u0103 \u00een produc\u021bie. Google are unele experimente ( <noindex><a rel=\"nofollow\" href=\"https:\/\/ai.google\/research\/teams\/brain\/robotics\/\">1<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/ai.googleblog.com\/2018\/06\/scalable-deep-reinforcement-learning.html\">2<\/a><\/noindex> ). Dar nu am v\u0103zut nicio sistem productiv.<\/p>\n<p><b>Memorie<\/b>. Dezavantajul tuturor celor descrise mai sus este lipsa de structur\u0103. Una dintre metodele prin care se \u00eencearc\u0103 s\u0103 se rezolve acest aspect este de a oferi re\u021belei neuronale acces la o memorie separat\u0103. Astfel, ea ar putea \u00eenregistra \u0219i rescrie rezultatele pa\u0219ilor s\u0103i. Atunci re\u021beaua neuronal\u0103 poate fi definit\u0103 de starea actual\u0103 a memoriei. Acest lucru sem\u0103n\u0103 foarte mult cu procesoarele \u0219i computerele clasice.<\/p>\n<p>Cea mai cunoscut\u0103 \u0219i popular\u0103 <noindex>articol <\/noindex> \u2014 de la DeepMind:<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/acc6bcd86c8c071fbd9776a91f990752.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nSe pare c\u0103 acesta este cheia \u00een\u021belegerii inteligen\u021bei? Dar mai degrab\u0103 nu. Sistemul are nevoie, oricum, de un volum uria\u0219 de date pentru antrenament. \u0218i func\u021bioneaz\u0103 \u00een principal cu date structurate \u00een form\u0103 de tabel. Totodat\u0103, c\u00e2nd Facebook <noindex><a rel=\"nofollow\" href=\"https:\/\/embodiedqa.org\/\">a abordat <\/a><\/noindex>o problem\u0103 similar\u0103, au optat pentru \u201enu avem nevoie de memorie, doar s\u0103 facem re\u021beaua neuronal\u0103 mai complex\u0103, plus mai multe exemple \u2014 \u0219i ea se va \u00eenv\u0103\u021ba singur\u0103\u201d.<\/p>\n<p><b>Descompunerea<\/b>. O alt\u0103 metod\u0103 de a crea o memorie semnificativ\u0103 este de a lua acelea\u0219i embedding-uri, dar \u00een timpul \u00eenv\u0103\u021b\u0103rii s\u0103 introduc\u0103 criterii suplimentare care s\u0103 permit\u0103 separarea \u201esensurilor\u201d. De exemplu, dac\u0103 dorim s\u0103 \u00eenv\u0103\u021b\u0103m re\u021beaua neuronal\u0103 s\u0103 disting\u0103 comportamentul unei persoane \u00een magazin. Dac\u0103 am urma calea standard \u2014 ar trebui s\u0103 facem o duzin\u0103 de re\u021bele. Una caut\u0103 persoana, a doua determin\u0103 ce face, a treia \u00eei stabile\u0219te v\u00e2rsta, a patra \u2014 sexul. O logic\u0103 separat\u0103 analizeaz\u0103 partea magazinului unde \u00eenva\u021b\u0103. A treia determin\u0103 traiectoria sa, etc.<\/p>\n<p>Sau, dac\u0103 ar exista o cantitate infinit\u0103 de date, atunci am putea antrena o singur\u0103 re\u021bea pe toate rezultatele posibile (evident, un astfel de volum de date este imposibil de adunat).<\/p>\n<p>Abordarea disentan\u021b\u0103r\u0103rii ne spune c\u0103 ar trebui s\u0103 \u00eenv\u0103\u021b\u0103m re\u021beaua astfel \u00eenc\u00e2t s\u0103 poat\u0103 distinge conceptele. Astfel, s\u0103 poat\u0103 genera un embedding din video, unde o zon\u0103 s\u0103 defineasc\u0103 ac\u021biunea, alta \u2014 pozi\u021bia pe podea \u00een timp, una \u2014 \u00een\u0103l\u021bimea unei persoane, iar alta \u2014 sexul acesteia. \u00cen acela\u0219i timp, \u00een timpul \u00eenv\u0103\u021b\u0103rii, ar fi bine s\u0103 evit\u0103m s\u0103 oferim re\u021belei indicii prea clare despre concepte cheie, ci s\u0103 o l\u0103s\u0103m s\u0103 eviden\u021bieze \u0219i s\u0103 grupeze zonele singur\u0103. Exist\u0103 destul de pu\u021bine articole de acest tip (unele dintre ele <noindex><a rel=\"nofollow\" href=\"https:\/\/ai.googleblog.com\/2019\/04\/evaluating-unsupervised-learning-of.html\">1<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"http:\/\/papers.nips.cc\/paper\/5851-deep-convolutional-inverse-graphics-network.pdf\">2<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/arxiv.org\/pdf\/1812.02230.pdf\">3<\/a><\/noindex>) \u0219i, \u00een general, sunt destul de teoretice. <\/p>\n<p>Dar aceast\u0103 direc\u021bie, m\u0103car teoretic, ar trebui s\u0103 rezolve problemele enumerate la \u00eenceput.<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/b36046dc3cbe3b849e1d3a60f56ec3a2.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nDezintegrarea imaginii pe parametrii \"culoarea pere\u021bilor\/culoarea podelei\/forma obiectului\/culoarea obiectului\/etc.\"<\/p>\n<p><img decoding=\"async\" alt=\"A explodat bula \u00eenv\u0103\u021b\u0103rii automate sau este \u00eenceputul unei noi ere?\" src=\"\/wp-content\/uploads\/4d5ac8dfc43f76b35f3490c26049c7db.png\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nDezintegrarea fe\u021bei pe parametrii \"dimensiune, spr\u00e2ncene, orientare, culoarea pielii, etc.\"<\/p>\n<h3>Altele<\/h3>\n<p>\nExist\u0103 multe alte direc\u021bii, nu at\u00e2t de globale, care permit cumva reducerea bazelor de date, lucr\u00e2nd cu date mai diverse, etc.<\/p>\n<p><b>Attention<\/b>. Probabil, nu are sens s\u0103 eviden\u021biem acest lucru ca pe o metod\u0103 separat\u0103. Este pur \u0219i simplu o abordare care \u00eent\u0103re\u0219te altele. I-au fost dedicate multe articole (<noindex><a rel=\"nofollow\" href=\"http:\/\/www.wildml.com\/2016\/01\/attention-and-memory-in-deep-learning-and-nlp\/\">1<\/a><\/noindex>,<noindex><a rel=\"nofollow\" href=\"http:\/\/jalammar.github.io\/visualizing-neural-machine-translation-mechanics-of-seq2seq-models-with-attention\/\">2<\/a><\/noindex>,<noindex><a rel=\"nofollow\" href=\"https:\/\/arxiv.org\/abs\/1706.03762\">3<\/a><\/noindex>). Sensul Attention este acela de a \u00eent\u0103ri reac\u021bia re\u021belei fa\u021b\u0103 de obiecte semnificative \u00een timpul \u00eenv\u0103\u021b\u0103rii. Adesea printr-o anumit\u0103 direc\u021bie extern\u0103, sau printr-o mic\u0103 re\u021bea extern\u0103.<\/p>\n<p><b>Simularea 3D<\/b>. Dac\u0103 se creeaz\u0103 un motor 3D bun, acesta poate acoperi adesea 90% dintre datele de instruire (am v\u0103zut chiar un exemplu \u00een care aproape 99% din date erau acoperite de un motor bun). Exist\u0103 multe idei \u0219i hacks pentru a face re\u021beaua antrenat\u0103 pe un motor 3D s\u0103 func\u021bioneze cu date reale (Fine tuning, transfer de stil, etc.). Dar de obicei s\u0103 creezi un motor bun este cu c\u00e2teva ordine de m\u0103rime mai complicat dec\u00e2t s\u0103 aduni date. Exemple de c\u00e2nd s-au creat motoare:<br \/>\n\u00cenv\u0103\u021barea robo\u021bilor (<noindex><a rel=\"nofollow\" href=\"https:\/\/ai.googleblog.com\/2018\/06\/teaching-uncalibrated-robots-to_22.html\">google<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/VZcmogKXC18\">braingarden<\/a><\/noindex>)<br \/>\nFormare <noindex><a rel=\"nofollow\" href=\"https:\/\/neuromation.io\/\">recunoa\u0219terii<\/a><\/noindex> produselor \u00een magazin (dar \u00een dou\u0103 proiecte pe care le-am f\u0103cut, ne-am descurcat f\u0103r\u0103 aceasta).<br \/>\n\u00cenv\u0103\u021barea la Tesla (din nou, acel video pe care l-am men\u021bionat mai sus).<\/p>\n<h2>Conclusions<\/h2>\n<p>\n\u00centreaga articol\u0103 este, \u00eentr-un anumit sens, concluzii. Probabil, mesajul principal pe care am vrut s\u0103-l transmit este \u2014 \u201ebonusul s-a terminat, neuralele nu mai ofer\u0103 solu\u021bii simple\u201d. Acum trebuie s\u0103 muncim din greu construind solu\u021bii complexe. Sau s\u0103 muncim f\u0103c\u00e2nd cercet\u0103ri \u0219tiin\u021bifice complexe.<\/p>\n<p>\u00cen general, subiectul este discutabil. Poate c\u0103 cititorii au exemple mai interesante?<br \/>\n<br \/>Sursa: <a content=\"nofollow\" rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/recognitor\/blog\/455676\/\">habr.com<\/a><\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"<p>\u041d\u0435\u0434\u0430\u0432\u043d\u043e \u0432\u044b\u0448\u043b\u0430 \u0441\u0442\u0430\u0442\u044c\u044f, \u043a\u043e\u0442\u043e\u0440\u0430\u044f \u043d\u0435\u043f\u043b\u043e\u0445\u043e \u043f\u043e\u043a\u0430\u0437\u044b\u0432\u0430\u0435\u0442 \u0442\u0435\u043d\u0434\u0435\u043d\u0446\u0438\u044e \u0432 \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u043c \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0438 \u043f\u043e\u0441\u043b\u0435\u0434\u043d\u0438\u0445 \u043b\u0435\u0442. \u0415\u0441\u043b\u0438 \u043a\u043e\u0440\u043e\u0442\u043a\u043e: \u0447\u0438\u0441\u043b\u043e \u0441\u0442\u0430\u0440\u0442\u0430\u043f\u043e\u0432 \u0432 \u043e\u0431\u043b\u0430\u0441\u0442\u0438 \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u0432 \u043f\u043e\u0441\u043b\u0435\u0434\u043d\u0438\u0435 \u0434\u0432\u0430 \u0433\u043e\u0434\u0430 \u0440\u0435\u0437\u043a\u043e \u0443\u043f\u0430\u043b\u043e. \u041d\u0443 \u0447\u0442\u043e. \u0420\u0430\u0437\u0431\u0435\u0440\u0451\u043c \u00ab\u043b\u043e\u043f\u043d\u0443\u043b \u043b\u0438 \u043f\u0443\u0437\u044b\u0440\u044c\u00bb, \u00ab\u043a\u0430\u043a \u0434\u0430\u043b\u044c\u0448\u0435 \u0436\u0438\u0442\u044c\u00bb \u0438 \u043f\u043e\u0433\u043e\u0432\u043e\u0440\u0438\u043c \u043e\u0442\u043a\u0443\u0434\u0430 \u0432\u043e\u043e\u0431\u0449\u0435 \u0442\u0430\u043a\u0430\u044f \u0437\u0430\u0433\u043e\u0433\u0443\u043b\u0438\u043d\u0430. \u0414\u043b\u044f \u043d\u0430\u0447\u0430\u043b\u0430 \u043f\u043e\u0433\u043e\u0432\u043e\u0440\u0438\u043c \u0447\u0442\u043e \u0431\u044b\u043b\u043e \u0431\u0443\u0441\u0442\u0435\u0440\u043e\u043c \u044d\u0442\u043e\u0439 \u043a\u0440\u0438\u0432\u043e\u0439. \u041e\u0442\u043a\u0443\u0434\u0430 \u043e\u043d\u0430 \u0432\u0437\u044f\u043b\u0430\u0441\u044c. \u041d\u0430\u0432\u0435\u0440\u043d\u043e\u0435 \u0432\u0441\u0451 \u0432\u0441\u043f\u043e\u043c\u043d\u044f\u0442 [&hellip;]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[702],"tags":[],"class_list":["post-35293","post","type-post","status-publish","format-standard","hentry","category-news"],"aioseo_notices":[],"aioseo_head":"\n\t\t<!-- All in One SEO 5.0.2.1 - aioseo.com -->\n\t<meta name=\"description\" content=\"\u041d\u0435\u0434\u0430\u0432\u043d\u043e \u0432\u044b\u0448\u043b\u0430\" \/>\n\t<meta name=\"robots\" content=\"max-image-preview:large\" \/>\n\t<meta name=\"author\" content=\"Yuri Gagarin\"\/>\n\t<link rel=\"canonical\" href=\"https:\/\/prohoster.info\/ro\/blog\/news\/lopnul-li-puzyr-mashinnogo-obucheniya-ili-nachalo-novoj-zari\" \/>\n\t<meta name=\"generator\" content=\"All in One SEO (AIOSEO) 5.0.2.1\" \/>\n\t\t<meta property=\"og:locale\" content=\"ro_RO\" \/>\n\t\t<meta property=\"og:site_name\" content=\"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b\" \/>\n\t\t<meta property=\"og:type\" content=\"article\" \/>\n\t\t<meta property=\"og:title\" content=\"\ud83e\udd47\u041b\u043e\u043f\u043d\u0443\u043b \u043b\u0438 \u043f\u0443\u0437\u044b\u0440\u044c \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f, \u0438\u043b\u0438 \u043d\u0430\u0447\u0430\u043b\u043e \u043d\u043e\u0432\u043e\u0439 \u0437\u0430\u0440\u0438 | ProHoster\" \/>\n\t\t<meta property=\"og:description\" content=\"\u041d\u0435\u0434\u0430\u0432\u043d\u043e \u0432\u044b\u0448\u043b\u0430\" \/>\n\t\t<meta property=\"og:url\" content=\"https:\/\/prohoster.info\/ro\/blog\/news\/lopnul-li-puzyr-mashinnogo-obucheniya-ili-nachalo-novoj-zari\" \/>\n\t\t<meta property=\"og:image\" content=\"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg\" \/>\n\t\t<meta property=\"og:image:secure_url\" content=\"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg\" \/>\n\t\t<meta property=\"og:image:width\" content=\"350\" \/>\n\t\t<meta property=\"og:image:height\" content=\"350\" \/>\n\t\t<meta property=\"article:published_time\" content=\"2019-10-31T19:03:28+00:00\" \/>\n\t\t<meta property=\"article:modified_time\" content=\"2019-10-31T19:03:28+00:00\" \/>\n\t\t<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/prohoster\" \/>\n\t\t<meta property=\"article:author\" content=\"https:\/\/www.facebook.com\/prohoster\" \/>\n\t\t<!-- All in One SEO -->\n\n","aioseo_head_json":{"title":"\ud83e\udd47A explodat bula \u00eenv\u0103\u021b\u0103rii automate, sau \u00eenceputul unei noi ere | ProHoster","description":"Recentemente a ap\u0103rut","canonical_url":"https:\/\/prohoster.info\/ro\/blog\/news\/lopnul-li-puzyr-mashinnogo-obucheniya-ili-nachalo-novoj-zari","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"ro_RO","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47\u041b\u043e\u043f\u043d\u0443\u043b \u043b\u0438 \u043f\u0443\u0437\u044b\u0440\u044c \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f, \u0438\u043b\u0438 \u043d\u0430\u0447\u0430\u043b\u043e \u043d\u043e\u0432\u043e\u0439 \u0437\u0430\u0440\u0438 | ProHoster","og:description":"\u041d\u0435\u0434\u0430\u0432\u043d\u043e \u0432\u044b\u0448\u043b\u0430","og:url":"https:\/\/prohoster.info\/ro\/blog\/news\/lopnul-li-puzyr-mashinnogo-obucheniya-ili-nachalo-novoj-zari","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2019-10-31T19:03:28+00:00","article:modified_time":"2019-10-31T19:03:28+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"35293","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"","isEnabled":true},"graphs":[]},"schema_type":null,"schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":"2026-01-21 22:41:19","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-03-01 02:06:24","updated":"2026-01-21 22:41:19","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/35293","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/comments?post=35293"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/35293\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media?parent=35293"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/categories?post=35293"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/tags?post=35293"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}