{"id":54780,"date":"2020-01-04T00:00:00","date_gmt":"2020-01-03T21:00:00","guid":{"rendered":"https:\/\/prohoster.info\/blog\/blog_prohoster\/nejroseti-kuda-eto-vse-dvizhetsya"},"modified":"2020-02-18T14:02:50","modified_gmt":"2020-02-18T11:02:50","slug":"nejroseti-kuda-eto-vse-dvizhetsya","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/nejroseti-kuda-eto-vse-dvizhetsya","title":{"rendered":"Nevros\u00edt\u011b. Kam to v\u0161e sm\u011b\u0159uje","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"<p>\u010cl\u00e1nek se skl\u00e1d\u00e1 ze dvou \u010d\u00e1st\u00ed:<\/p>\n<p><\/p>\n<ol>\n<li>Kr\u00e1tk\u00fd p\u0159ehled n\u011bkolika architektur s\u00edt\u00ed pro rozpozn\u00e1v\u00e1n\u00ed objekt\u016f na obr\u00e1zc\u00edch a segmentaci obr\u00e1zk\u016f s nejjasn\u011bj\u0161\u00edmi odkazy na zdroje, kter\u00e9 pro m\u011b byly srozumiteln\u00e9. Sna\u017eil jsem se vyb\u00edrat videa s vysv\u011btlen\u00edmi, ide\u00e1ln\u011b v ru\u0161tin\u011b.<\/li>\n<li>Druh\u00e1 \u010d\u00e1st se pokou\u0161\u00ed pochopit sm\u011br v\u00fdvoje architektur neuronov\u00fdch s\u00edt\u00ed a technologi\u00ed, kter\u00e9 jsou na nich zalo\u017eeny.<\/li>\n<\/ol>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Nevros\u00edt\u011b. Kam to v\u0161e sm\u011b\u0159uje\" src=\"\/wp-content\/uploads\/2020\/01\/3e0238547dc956dbb069c11241e4534f.png\" style=\"display:block;margin: 0 auto;\" \/><\/p>\n<p><\/p>\n<p>Obr\u00e1zek 1 \u2013 Ch\u00e1pat architektury neur\u00e1ln\u00edch s\u00edt\u00ed nen\u00ed jednoduch\u00e9<\/p>\n<p><\/p>\n<p>V\u0161echno to za\u010dalo t\u00edm, \u017ee jsem vytvo\u0159il dv\u011b demonstra\u010dn\u00ed aplikace pro klasifikaci a rozpozn\u00e1v\u00e1n\u00ed objekt\u016f na telefonu Android:<\/p>\n<p><\/p>\n<ul>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/junkyard\/tree\/master\/object_classifier\">Back-end demo<\/a><\/noindex>, kdy jsou data zpracov\u00e1v\u00e1na na serveru a p\u0159en\u00e1\u0161ena do telefonu. Klasifikace obr\u00e1zk\u016f (image classification) t\u0159\u00ed typ\u016f medv\u011bd\u016f: hn\u011bd\u00e9ho, \u010dern\u00e9ho a ply\u0161ov\u00e9ho.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/android\/tree\/master\/object_detection_demo\">Front-end demo<\/a><\/noindex>, kdy jsou data zpracov\u00e1v\u00e1na p\u0159\u00edmo na telefonu. Rozpozn\u00e1v\u00e1n\u00ed objekt\u016f (object detection) t\u0159\u00ed typ\u016f: l\u00edskov\u00fd o\u0159ech, f\u00edk a datlovn\u00edk.<\/li>\n<\/ul>\n<p><noindex><a rel=\"nofollow\" name=\"habracut\"><\/a><\/noindex><\/p>\n<p>Existuje rozd\u00edl mezi \u00fakoly klasifikace obr\u00e1zk\u016f, rozpozn\u00e1v\u00e1n\u00ed objekt\u016f na obr\u00e1zku a <noindex><a rel=\"nofollow\" href=\"https:\/\/medium.com\/analytics-vidhya\/image-classification-vs-object-detection-vs-image-segmentation-f36db85fe81\">segmentac\u00ed obr\u00e1zk\u016f<\/a><\/noindex>. Proto vznikla pot\u0159eba zjistit, kter\u00e9 architektury neuronov\u00fdch s\u00edt\u00ed rozpozn\u00e1vaj\u00ed objekty na obr\u00e1zc\u00edch a kter\u00e9 mohou segmentovat. Na\u0161el jsem n\u00e1sleduj\u00edc\u00ed p\u0159\u00edklady architektur s nejjasn\u011bj\u0161\u00edmi odkazy na zdroje:<\/p>\n<p><\/p>\n<ul>\n<li>S\u00e9rie architektur na z\u00e1klad\u011b R-CNN (<strong>R<\/strong>egions with <strong>C<\/strong>onvolution <strong>N<\/strong>eural <strong>N<\/strong>etworks features): R-CNN, Fast R-CNN, <noindex><a rel=\"nofollow\" href=\"https:\/\/medium.com\/@smallfishbigsea\/faster-r-cnn-explained-864d4fb7e3f8\">Faster R-CNN<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/0vt05rQqk_I\">Mask R-CNN<\/a><\/noindex>. Pro rozpozn\u00e1v\u00e1n\u00ed objektu na obr\u00e1zku pomoc\u00ed mechanismu Region Proposal Network (RPN) se vymezuj\u00ed omezen\u00e9 regiony (bounding boxes). P\u016fvodn\u011b byl m\u00edsto RPN pou\u017eit pomalej\u0161\u00ed mechanismus Selective Search. Pot\u00e9 jsou vymezen\u00e9 omezen\u00e9 regiony p\u0159ed\u00e1ny na vstup obvykl\u00e9 neuronov\u00e9 s\u00edti pro klasifikaci. V architektu\u0159e R-CNN jsou jasn\u00e9 cykly \u201efor\u201c pro proch\u00e1zen\u00ed omezen\u00fdmi regiony, celkem a\u017e 2000 pr\u016fchod\u016f p\u0159es intern\u00ed s\u00ed\u0165 AlexNet. Vzhledem k jasn\u00fdm cykl\u016fm \u201efor\u201c se zpomaluje rychlost zpracov\u00e1n\u00ed obr\u00e1zk\u016f. Po\u010det jasn\u00fdch cykl\u016f, pr\u016fchod\u016f p\u0159es intern\u00ed neuronovou s\u00ed\u0165, se s ka\u017edou novou verz\u00ed architektury sni\u017euje a tak\u00e9 se prov\u00e1d\u00ed des\u00edtky dal\u0161\u00edch zm\u011bn ke zv\u00fd\u0161en\u00ed rychlosti a k nahrazen\u00ed \u00fakolu rozpozn\u00e1v\u00e1n\u00ed objekt\u016f segmentac\u00ed objekt\u016f v Mask R-CNN.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/L0tzmv--CGY\">YOLO<\/a><\/noindex> (<strong>Y<\/strong>ou <strong>\u00fatoky jsou dnes velmi roz\u0161\u00ed\u0159en\u00e9 ve sv\u011bt\u011b po\u010d\u00edta\u010dov\u00fdch server\u016f. Jejich hlavn\u00edm c\u00edlem je vy\u0159adit servery z provozu pomoc\u00ed mnoha po\u017eadavk\u016f. V d\u016fsledku takov\u00e9ho \u00fatoku jsou servery zahlceny po\u017eadavky a vyp\u00ednaj\u00ed se. Tento typ \u00fatoku je nyn\u00ed velmi popul\u00e1rn\u00ed mezi hackery. Vyd\u00edraj\u00ed pen\u00edze za zastaven\u00ed takov\u00e9ho \u00fatoku. Obvykle je organizov\u00e1n pomoc\u00ed \u201etrojsk\u00fdch kon\u00ed\u201c, kter\u00e9 se do po\u010d\u00edta\u010de dostanou \u010dasto z internetu.<\/strong>nly <strong>L<\/strong>ook <strong>\u00fatoky jsou dnes velmi roz\u0161\u00ed\u0159en\u00e9 ve sv\u011bt\u011b po\u010d\u00edta\u010dov\u00fdch server\u016f. Jejich hlavn\u00edm c\u00edlem je vy\u0159adit servery z provozu pomoc\u00ed mnoha po\u017eadavk\u016f. V d\u016fsledku takov\u00e9ho \u00fatoku jsou servery zahlceny po\u017eadavky a vyp\u00ednaj\u00ed se. Tento typ \u00fatoku je nyn\u00ed velmi popul\u00e1rn\u00ed mezi hackery. Vyd\u00edraj\u00ed pen\u00edze za zastaven\u00ed takov\u00e9ho \u00fatoku. Obvykle je organizov\u00e1n pomoc\u00ed \u201etrojsk\u00fdch kon\u00ed\u201c, kter\u00e9 se do po\u010d\u00edta\u010de dostanou \u010dasto z internetu.<\/strong>nce) \u2013 prvn\u00ed neuronov\u00e1 s\u00ed\u0165, kter\u00e1 rozpozn\u00e1vala objekty v re\u00e1ln\u00e9m \u010dase na mobiln\u00edch za\u0159\u00edzen\u00edch. Hlavn\u00ed charakteristikou je rozli\u0161ov\u00e1n\u00ed objekt\u016f na z\u00e1klad\u011b jednoho pr\u016fchodu (sta\u010d\u00ed se pod\u00edvat jednou). Tedy v architektu\u0159e YOLO nejsou \u017e\u00e1dn\u00e9 explicitn\u00ed cykly \u201efor\u201c, d\u00edky \u010demu\u017e s\u00ed\u0165 pracuje rychle. Nap\u0159\u00edklad analogie: v NumPy p\u0159i operac\u00edch s maticemi tak\u00e9 nejsou \u017e\u00e1dn\u00e9 explicitn\u00ed cykly \u201efor\u201c, kter\u00e9 jsou v NumPy realizov\u00e1ny na ni\u017e\u0161\u00edch \u00farovn\u00edch architektury prost\u0159ednictv\u00edm programovac\u00edho jazyka C. YOLO pou\u017e\u00edv\u00e1 m\u0159\u00ed\u017eku p\u0159edem definovan\u00fdch oken. Aby se stejn\u00fd objekt neidentifikoval opakovan\u011b, pou\u017e\u00edv\u00e1 se koeficient p\u0159ekryt\u00ed oken (IoU, <strong>I<\/strong>ntersection <strong>o<\/strong>ver <strong>U<\/strong>nion). Tato architektura pracuje v \u0161irok\u00e9m rozsahu a m\u00e1 vysokou <noindex><a rel=\"nofollow\" href=\"https:\/\/ru.wikipedia.org\/wiki\/%D0%A0%D0%BE%D0%B1%D0%B0%D1%81%D1%82%D0%BD%D0%BE%D1%81%D1%82%D1%8C\">robustnost<\/a><\/noindex>: model m\u016f\u017ee b\u00fdt tr\u00e9nov\u00e1n na fotografi\u00edch, ale z\u00e1rove\u0148 dob\u0159e funguje na kreslen\u00fdch obrazech.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/P8e-G-Mhx4k\">SSD<\/a><\/noindex> (<strong>S<\/strong>ingle <strong>S<\/strong>hot MultiBox <strong>DDoS<\/strong>etector) \u2013 pou\u017e\u00edvaj\u00ed se nej\u00fasp\u011b\u0161n\u011bj\u0161\u00ed \u201ehacky\u201c architektury YOLO (nap\u0159\u00edklad non-maximum suppression) a p\u0159id\u00e1vaj\u00ed se nov\u00e9, aby neuronov\u00e1 s\u00ed\u0165 rychleji a p\u0159esn\u011bji pracovala. Hlavn\u00ed charakteristikou je rozli\u0161ov\u00e1n\u00ed objekt\u016f na z\u00e1klad\u011b jednoho pr\u016fchodu pomoc\u00ed p\u0159edem definovan\u00e9 m\u0159\u00ed\u017eky oken (default box) na pyramid\u011b obraz\u016f. Pyramida obraz\u016f je zak\u00f3dov\u00e1na v konvolu\u010dn\u00edch tensorech p\u0159i sekven\u010dn\u00edch operac\u00edch konvoluce a poolingu (p\u0159i operaci max-pooling prostorov\u00e1 dimenze kles\u00e1). T\u00edmto zp\u016fsobem jsou ur\u010dov\u00e1ny jak velk\u00e9, tak mal\u00e9 objekty na jednom pr\u016fchodu s\u00edt\u011b.<\/li>\n<li>MobileSSD (<strong>Mobile<\/strong>NetV2 + <strong>SSD<\/strong>) \u2013 kombinace dvou architektur neuronov\u00fdch s\u00edt\u00ed. Prvn\u00ed s\u00ed\u0165 <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/352804\/\">MobileNetV2<\/a><\/noindex> pracuje rychle a zvy\u0161uje p\u0159esnost rozpozn\u00e1v\u00e1n\u00ed. MobileNetV2 se pou\u017e\u00edv\u00e1 m\u00edsto VGG-16, kter\u00e1 byla p\u016fvodn\u011b pou\u017eita v <noindex><a rel=\"nofollow\" href=\"https:\/\/arxiv.org\/abs\/1512.02325\">origin\u00e1ln\u00edm \u010dl\u00e1nku<\/a><\/noindex>. Druh\u00e1 s\u00ed\u0165 SSD ur\u010duje um\u00edst\u011bn\u00ed objekt\u016f v obr\u00e1zku.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/ge_RT5wvHvY\">SqueezeNet<\/a><\/noindex> \u2013 velmi mal\u00e1, ale p\u0159esn\u00e1 neuronov\u00e1 s\u00ed\u0165. Sama o sob\u011b ne\u0159e\u0161\u00ed \u00fakol detekce objekt\u016f. M\u016f\u017ee v\u0161ak b\u00fdt pou\u017eita v kombinaci r\u016fzn\u00fdch architektur. A vyu\u017e\u00edv\u00e1na na mobiln\u00edch za\u0159\u00edzen\u00edch. Hlavn\u00ed charakteristikou je, \u017ee nejprve jsou data stla\u010dena do \u010dty\u0159 1\u00d71 konvolu\u010dn\u00edch filtr\u016f, a pot\u00e9 roz\u0161\u00ed\u0159ena do \u010dty\u0159 1\u00d71 a \u010dty\u0159 3\u00d73 konvolu\u010dn\u00edch filtr\u016f. Jedna takov\u00e1 iterace stla\u010den\u00ed-roz\u0161\u00ed\u0159en\u00ed dat se naz\u00fdv\u00e1 \u201eFire Module\u201c.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/b6jhopSMit8\">DeepLab<\/a><\/noindex> (Semantic Image Segmentation with Deep Convolutional Nets) \u2013 segmentace objekt\u016f na obr\u00e1zku. Rozhoduj\u00edc\u00edm rysem architektury je \u0159\u00eddk\u00e1 (dilated convolution) konvoluce, kter\u00e1 zachov\u00e1v\u00e1 prostorov\u00e9 rozli\u0161en\u00ed. N\u00e1sledn\u011b n\u00e1sleduje f\u00e1ze postprocessing v\u00fdsledk\u016f s vyu\u017eit\u00edm grafick\u00e9ho pravd\u011bpodobnostn\u00edho modelu (conditional random field), co\u017e umo\u017e\u0148uje odstranit drobn\u00e9 \u0161umy v segmentaci a zlep\u0161it kvalitu odd\u011blen\u00e9ho obr\u00e1zku. Za hroziv\u00fdm n\u00e1zvem \u201egrafick\u00fd pravd\u011bpodobnostn\u00ed model\u201c se skr\u00fdv\u00e1 oby\u010dejn\u00fd Gauss\u016fv filtr, kter\u00fd je aproximov\u00e1n na p\u011bti bodech.<\/li>\n<li>Sna\u017eil jsem se pochopit za\u0159\u00edzen\u00ed <noindex><a rel=\"nofollow\" href=\"https:\/\/arxiv.org\/abs\/1711.06897\">RefineDet<\/a><\/noindex> (Single-Shot <strong>Refine<\/strong>ment Neural Network for Object <strong>Det<\/strong>ection), ale moc jsem toho nepochopil.<\/li>\n<li>Pod\u00edval jsem se tak\u00e9, jak funguje technologie \u201epozornost\u201c (attention): <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/W2rWgXJBZhU\">video1<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/iDulhoQ2pro\">video2<\/a><\/noindex>, <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/H6Qiegq_36c\">video3<\/a><\/noindex>. Rozhoduj\u00edc\u00edm rysem architektury \u201epozornost\u201c je automatick\u00e9 vy\u010dle\u0148ov\u00e1n\u00ed oblast\u00ed zv\u00fd\u0161en\u00e9 pozornosti na obr\u00e1zku (RoI, <strong>R<\/strong>egions <strong>o<\/strong>f <strong>I<\/strong>nterest) pomoc\u00ed neuronov\u00e9 s\u00edt\u011b nazvan\u00e9 Attention Unit. Oblasti zv\u00fd\u0161en\u00e9 pozornosti se podobaj\u00ed ohrani\u010den\u00fdm oblastem (bounding boxes), ale na rozd\u00edl od nich nejsou na obr\u00e1zku fixov\u00e1ny a mohou m\u00edt rozmazan\u00e9 okraje. Pot\u00e9 se z oblast\u00ed zv\u00fd\u0161en\u00e9 pozornosti vyb\u00edraj\u00ed znaky (fich), kter\u00e9 jsou \u201ep\u0159ed\u00e1v\u00e1ny\u201c rekurentn\u00edm neuronov\u00fdm s\u00edt\u00edm s architekturami <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/5lUUrREboSk\">LSDM, GRU nebo Vanilla RNN<\/a><\/noindex>. Rekurentn\u00ed neuronov\u00e9 s\u00edt\u011b um\u00ed analyzovat vztahy mezi znaky v sekvenci. Rekurentn\u00ed neuronov\u00e9 s\u00edt\u011b byly p\u016fvodn\u011b pou\u017eity pro p\u0159eklady textu do jin\u00fdch jazyk\u016f, nyn\u00ed i pro p\u0159eklad <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/e-WB4lfg30M\">obr\u00e1zku do textu<\/a><\/noindex> a <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/rAbhypxs1qQ\">textu do obr\u00e1zku<\/a><\/noindex>.<\/li>\n<\/ul>\n<p><\/p>\n<p>Jak jsem studoval tyto architektury <strong>pochopil jsem, \u017ee nic nerozum\u00edm<\/strong>. A nejde o to, \u017ee by moje neuronov\u00e1 s\u00ed\u0165 m\u011bla probl\u00e9my s mechanismem pozornosti. Vytv\u00e1\u0159en\u00ed v\u0161ech t\u011bchto architektur vypad\u00e1 jako n\u011bjak\u00fd obrovsk\u00fd hackathon, kde se auto\u0159i sout\u011b\u017e\u00ed v hackch. Hack (hack) \u2013 rychl\u00e9 \u0159e\u0161en\u00ed slo\u017eit\u00e9 programov\u00e9 \u00falohy. To znamen\u00e1, \u017ee mezi v\u0161emi t\u011bmito architekturami nen\u00ed viditeln\u00e1 a srozumiteln\u00e1 logick\u00e1 vazba. V\u0161e, co je spojuje, je soubor nej\u00fasp\u011b\u0161n\u011bj\u0161\u00edch hack\u016f, kter\u00e9 si navz\u00e1jem p\u0159eb\u00edraj\u00ed, plus spole\u010dn\u00e1 pro v\u0161echny <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/Ilg3gGewQ5U\">operace konvoluce s zp\u011btnou vazbou<\/a><\/noindex> (zp\u011btn\u00e9 \u0161\u00ed\u0159en\u00ed chyby, backpropagation). Neexistuje <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/272473\/\">syst\u00e9mov\u00e9 my\u0161len\u00ed<\/a><\/noindex>! Nen\u00ed jasn\u00e9, co m\u011bnit a jak optimalizovat st\u00e1vaj\u00edc\u00ed \u00fasp\u011bchy.<\/p>\n<p><\/p>\n<p>V d\u016fsledku absence logick\u00e9ho spojen\u00ed mezi hacky je velmi obt\u00ed\u017en\u00e9 je zapamatovat a aplikovat v praxi. Jedn\u00e1 se o fragmentovan\u00e9 znalosti. V nejlep\u0161\u00edm p\u0159\u00edpad\u011b si zapamatujeme p\u00e1r zaj\u00edmav\u00fdch a neo\u010dek\u00e1van\u00fdch moment\u016f, ale v\u011bt\u0161ina ch\u00e1p\u00e1n\u00ed a nepochopen\u00ed zmiz\u00ed z pam\u011bti u\u017e po n\u011bkolika dnech. Bude dob\u0159e, pokud si alespo\u0148 za t\u00fdden vzpomeneme na n\u00e1zev architektury. A p\u0159itom bylo str\u00e1veno n\u011bkolik hodin a dokonce dn\u00ed pracovn\u00edho \u010dasu na \u010dten\u00ed \u010dl\u00e1nk\u016f a sledov\u00e1n\u00ed p\u0159ehledov\u00fdch vide\u00ed!<\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Nevros\u00edt\u011b. Kam to v\u0161e sm\u011b\u0159uje\" src=\"\/wp-content\/uploads\/2020\/01\/86f0f24e3be1d0f1a210f3132897d981.png\" style=\"display:block;margin: 0 auto;\" \/><\/p>\n<p><\/p>\n<p>Obr\u00e1zek 2 \u2013 <noindex><a rel=\"nofollow\" href=\"https:\/\/www.asimovinstitute.org\/neural-network-zoo\/\">ZOO neuronov\u00fdch s\u00edt\u00ed<\/a><\/noindex><\/p>\n<p><\/p>\n<p>V\u011bt\u0161ina autor\u016f v\u011bdeck\u00fdch \u010dl\u00e1nk\u016f, podle m\u00e9ho osobn\u00edho n\u00e1zoru, d\u011bl\u00e1 v\u0161e pro to, aby i tyto fragmentovan\u00e9 znalosti nebyly \u010dten\u00e1\u0159i pochopeny. Ale participi\u00e1ln\u00ed obraty v deseti \u0159\u00e1dkov\u00fdch v\u011bt\u00e1ch s formulacemi, kter\u00e9 jsou vzaty \u201eze stropu\u201c \u2013 to je t\u00e9ma pro samostatn\u00fd \u010dl\u00e1nek (probl\u00e9m <noindex><a rel=\"nofollow\" href=\"https:\/\/en.wikipedia.org\/wiki\/Publish_or_perish\">publish or perish<\/a><\/noindex>).<\/p>\n<p><\/p>\n<p>Z tohoto d\u016fvodu se objevila pot\u0159eba systematizovat informace o neuronov\u00fdch s\u00edt\u00edch a t\u00edm zv\u00fd\u0161it kvalitu ch\u00e1p\u00e1n\u00ed a zapamatov\u00e1n\u00ed. Hlavn\u00edm t\u00e9matem anal\u00fdzy jednotliv\u00fdch technologi\u00ed a architektur um\u011bl\u00fdch neuronov\u00fdch s\u00edt\u00ed se stala n\u00e1sleduj\u00edc\u00ed \u00faloha: <strong>zjistit, kam to v\u0161echno sm\u011b\u0159uje<\/strong>, a nikoli strukturu n\u011bjak\u00e9 konkr\u00e9tn\u00ed neuronov\u00e9 s\u00edt\u011b samostatn\u011b.<\/p>\n<p><\/p>\n<p>Kam to v\u0161echno sm\u011b\u0159uje. Hlavn\u00ed v\u00fdsledky:<\/p>\n<p><\/p>\n<ul>\n<li>Po\u010det startup\u016f v oblasti strojov\u00e9ho u\u010den\u00ed v posledn\u00edch dvou letech <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/recognitor\/blog\/455676\/\">ost\u0159e klesl<\/a><\/noindex>. Mo\u017en\u00fd d\u016fvod: \u201eneuronov\u00e9 s\u00edt\u011b p\u0159estaly b\u00fdt n\u011b\u010d\u00edm nov\u00fdm.\u201c<\/li>\n<li>Ka\u017ed\u00fd si bude moci vytvo\u0159it funguj\u00edc\u00ed neuronovou s\u00ed\u0165 pro \u0159e\u0161en\u00ed jednoduch\u00e9ho \u00fakolu. K tomu vezme hotov\u00fd model z \u201ezoo model\u016f\u201c (model zoo) a natr\u00e9nuje posledn\u00ed vrstvu neuronov\u00e9 s\u00edt\u011b (<noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/yofjFQddwHE\">transfer learning<\/a><\/noindex>) na hotov\u00fdch datech z <noindex><a rel=\"nofollow\" href=\"https:\/\/toolbox.google.com\/datasetsearch\">Google Dataset Search<\/a><\/noindex> nebo z <noindex><a rel=\"nofollow\" href=\"https:\/\/www.kaggle.com\/datasets\">25 tis\u00edc dataset\u016f Kaggle<\/a><\/noindex> v bezplatn\u00e9m <noindex><a rel=\"nofollow\" href=\"https:\/\/www.dataschool.io\/cloud-services-for-jupyter-notebook\/\">cloudu Jupyter Notebook<\/a><\/noindex>.<\/li>\n<li>Velc\u00ed v\u00fdrobci neuronov\u00fdch s\u00edt\u00ed za\u010dali vytv\u00e1\u0159et <strong>\u201ezoo model\u016f\u201c<\/strong> (model zoo). Pomoc\u00ed nich lze rychle vytvo\u0159it komer\u010dn\u00ed aplikaci: <noindex><a rel=\"nofollow\" href=\"https:\/\/tfhub.dev\/\">TF Hub<\/a><\/noindex> pro TensorFlow, <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/open-mmlab\/mmdetection\">MMDetection<\/a><\/noindex> pro PyTorch, <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/facebookresearch\/Detectron\">Detectron<\/a><\/noindex> pro Caffe2, <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/wkentaro\/chainer-modelzoo\">chainer-modelzoo<\/a><\/noindex> pro Chainer a <noindex><a rel=\"nofollow\" href=\"https:\/\/modelzoo.co\/\">dal\u0161\u00ed<\/a><\/noindex>.<\/li>\n<li>Neuronov\u00e9 s\u00edt\u011b pracuj\u00edc\u00ed v <strong>re\u00e1ln\u00e9m \u010dase<\/strong> (real-time) na mobiln\u00edch za\u0159\u00edzen\u00edch. Od 10 do 50 sn\u00edmk\u016f za sekundu.<\/li>\n<li>Pou\u017eit\u00ed neuronov\u00fdch s\u00edt\u00ed v telefonech (TF Lite), v prohl\u00ed\u017ee\u010d\u00edch (TF.js) a v <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/19ZNz2N79u4\">dom\u00e1c\u00edch p\u0159edm\u011btech<\/a><\/noindex> (IoT, <strong>I<\/strong>internet <strong>o<\/strong>f <strong>T<\/strong>hings). Zvl\u00e1\u0161\u0165 v telefonech, kter\u00e9 ji\u017e podporuj\u00ed neuronov\u00e9 s\u00edt\u011b na \u00farovni \u201ehardwaru\u201c (neuroakceler\u00e1tory).<\/li>\n<li>\u201eKa\u017ed\u00e9 za\u0159\u00edzen\u00ed, oble\u010den\u00ed a mo\u017en\u00e1 i j\u00eddlo budou m\u00edt <strong>IP-v6 adresu<\/strong> a budou mezi sebou komunikovat\u201c \u2013 <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/GG7H8Xa4m8I?t=85\">Sebastian Thrun<\/a><\/noindex>.<\/li>\n<li>N\u00e1r\u016fst publikac\u00ed o strojov\u00e9m u\u010den\u00ed za\u010dal <noindex><a rel=\"nofollow\" href=\"http:\/\/data-mining.philippe-fournier-viger.com\/too-many-machine-learning-papers\">p\u0159ekra\u010dovat Moore\u016fv z\u00e1kon<\/a><\/noindex> (zdvojen\u00ed ka\u017ed\u00e9 dva roky) od roku 2015. Je jasn\u00e9, \u017ee pot\u0159ebujeme neuronov\u00e9 s\u00edt\u011b pro anal\u00fdzu \u010dl\u00e1nk\u016f.<\/li>\n<li>N\u00e1sleduj\u00edc\u00ed technologie z\u00edsk\u00e1vaj\u00ed na popularit\u011b:\n<ul>\n<li><strong>PyTorch<\/strong> \u2013 popularita rychle roste a zd\u00e1 se, \u017ee p\u0159ekon\u00e1v\u00e1 TensorFlow.<\/li>\n<li>Automatick\u00fd v\u00fdb\u011br hyperparametr\u016f <strong>AutoML<\/strong> \u2013 popularita roste plynule.<\/li>\n<li>Postupn\u00e9 zmen\u0161ov\u00e1n\u00ed p\u0159esnosti a zvy\u0161ov\u00e1n\u00ed rychlosti v\u00fdpo\u010dt\u016f: <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/rln_kZbYaWc\">fuzzy logika<\/a><\/noindex>, algoritmy <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/MIPkK5ZAsms\">boostingu<\/a><\/noindex>, nep\u0159esn\u00e9 (p\u0159ibli\u017en\u00e9) v\u00fdpo\u010dty, kvantizace (kdy jsou v\u00e1hy neuronov\u00fdch s\u00edt\u00ed p\u0159ev\u00e1d\u011bny na cel\u00e1 \u010d\u00edsla a kvantizov\u00e1ny), neuroakceler\u00e1tory.<\/li>\n<li>P\u0159eklad <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/e-WB4lfg30M\">obr\u00e1zku do textu<\/a><\/noindex> a <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/rAbhypxs1qQ\">textu do obr\u00e1zku<\/a><\/noindex>.<\/li>\n<li>Tvorba <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/OrHLacCDZVQ\">trojrozm\u011brn\u00fdch objekt\u016f z videa<\/a><\/noindex>, nyn\u00ed ji\u017e v re\u00e1ln\u00e9m \u010dase.<\/li>\n<li>Hlavn\u00ed v DL \u2013 to jsou velk\u00e9 data, ale jejich sb\u011br a anotace nejsou jednoduch\u00e9. Proto se vyv\u00edj\u00ed automatizace anotace (<noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/NcKTn4C91Yc\">automated annotation<\/a><\/noindex>) pro neuronov\u00e9 s\u00edt\u011b pomoc\u00ed neuronov\u00fdch s\u00edt\u00ed.<\/li>\n<\/ul>\n<\/li>\n<li>S neuronov\u00fdmi s\u00edt\u011bmi se po\u010d\u00edta\u010dov\u00e9 v\u011bdy n\u00e1hle staly <strong>experiment\u00e1ln\u00ed v\u011bdou<\/strong> a vznikla <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/480348\">krize reprodukovatelnosti<\/a><\/noindex>.<\/li>\n<li>IT-pen\u00edze a popularita neuronov\u00fdch s\u00edt\u00ed vznikly sou\u010dasn\u011b, kdy\u017e se v\u00fdpo\u010dty staly tr\u017en\u00ed hodnotou. Ekonomika ze zlatovl\u00e1knov\u00e9 se st\u00e1v\u00e1 <strong>zlatovl\u00e1knov\u011b-v\u00fdpo\u010dtovou<\/strong>. Pod\u00edvejte se na m\u016fj \u010dl\u00e1nek o <noindex><a rel=\"nofollow\" href=\"https:\/\/ru.wikipedia.org\/wiki\/%D0%AD%D0%BA%D0%BE%D0%BD%D0%BE%D1%84%D0%B8%D0%B7%D0%B8%D0%BA%D0%B0\">ekonofyzice<\/a><\/noindex> a o d\u016fvodu vzniku IT-pen\u011bz.<\/li>\n<\/ul>\n<p><\/p>\n<p>Postupn\u011b se objevuje nov\u00e1 <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/481844\">metodologie programov\u00e1n\u00ed ML\/DL<\/a><\/noindex> (Strojov\u00e9 u\u010den\u00ed &amp; Hlubok\u00e9 u\u010den\u00ed), kter\u00e1 je zalo\u017eena na p\u0159edstav\u011b programu jako souhrnu vytr\u00e9novan\u00fdch model\u016f neuronov\u00fdch s\u00edt\u00ed.<\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Nevros\u00edt\u011b. Kam to v\u0161e sm\u011b\u0159uje\" src=\"\/wp-content\/uploads\/2020\/01\/17df76a613191522145abeba177c2685.png\" style=\"display:block;margin: 0 auto;\" \/><\/p>\n<p><\/p>\n<p>Obr\u00e1zek 3 \u2013 ML\/DL jako nov\u00e1 metodologie programov\u00e1n\u00ed<\/p>\n<p><\/p>\n<p>Nicm\u00e9n\u011b st\u00e1le neexistuje <strong>\u201eteorie neuronov\u00fdch s\u00edt\u00ed\u201c<\/strong>, ve kter\u00e9 by bylo mo\u017en\u00e9 myslet a pracovat syst\u00e9mov\u011b. To, co je nyn\u00ed naz\u00fdv\u00e1no \u201eteori\u00ed\u201c, jsou ve skute\u010dnosti experiment\u00e1ln\u00ed, heuristick\u00e9 algoritmy.<\/p>\n<p><\/p>\n<p>Odkazy na moje a jin\u00e9 zdroje:<\/p>\n<p><\/p>\n<ul>\n<li>Zpravodajstv\u00ed o Data Science. Hlavn\u011b o zpracov\u00e1n\u00ed obr\u00e1zk\u016f. Kdo chce dost\u00e1vat, a\u0165 po\u0161le e-mail (foobar167&lt;h\u00e1\u010dek-h\u00e1\u010dek&gt;gmail&lt;te\u010dka&gt;com). Odkazy na \u010dl\u00e1nky a videa pos\u00edl\u00e1m podle toho, jak shroma\u017e\u010fuji materi\u00e1l.<\/li>\n<li>Celkov\u00fd <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/articles\/blob\/master\/Machine_Learning\/courses_on_machine_learning.md\">seznam kurz\u016f a \u010dl\u00e1nk\u016f<\/a><\/noindex>, kter\u00e9 jsem absolvoval a kter\u00e9 bych r\u00e1d absolvoval.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/articles\/blob\/master\/Ubuntu\/13_Keras_and_TensorFlow_how-tos.md#exercises\">Kurzy a videa pro za\u010d\u00e1te\u010dn\u00edky<\/a><\/noindex>, ze kter\u00fdch byste m\u011bli za\u010d\u00edt studovat neuronov\u00e9 s\u00edt\u011b. Plus bro\u017eura <noindex><a rel=\"nofollow\" href=\"https:\/\/foobar167.github.io\/page\/vvedeniye-v-mashinnoye-obucheniye-i-iskusstvennyye-neyronnyye-seti.html\">\u201e\u00davod do strojov\u00e9ho u\u010den\u00ed a um\u011bl\u00fdch neuronov\u00fdch s\u00edt\u00ed\u201c<\/a><\/noindex>.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/articles\/blob\/master\/Ubuntu\/13_Keras_and_TensorFlow_how-tos.md#tools\">U\u017eite\u010dn\u00e9 n\u00e1stroje<\/a><\/noindex>, kde si ka\u017ed\u00fd najde n\u011bco zaj\u00edmav\u00e9ho pro sebe.<\/li>\n<li>Velmi u\u017eite\u010dn\u00e9 se uk\u00e1zaly <strong>videokan\u00e1ly zam\u011b\u0159en\u00e9 na anal\u00fdzu v\u011bdeck\u00fdch \u010dl\u00e1nk\u016f<\/strong> v oblasti Data Science. Vyhled\u00e1vejte je, p\u0159ihlaste se k odb\u011bru a sd\u00edlejte odkazy se sv\u00fdmi kolegy, a tak\u00e9 se mnou. P\u0159\u00edklady:\n<ul>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/www.youtube.com\/user\/keeroyz\">Two Minute Papers<\/a><\/noindex><\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/www.youtube.com\/channel\/UCHB9VepY6kYvZjj0Bgxnpbw\">Henry AI Labs<\/a><\/noindex><\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/www.youtube.com\/channel\/UCZHmQk67mSJgfCCTn7xBfew\">Yannic Kilcher<\/a><\/noindex><\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/www.youtube.com\/channel\/UC5_6ZD6s8klmMu9TXEB_1IA\">CodeEmporium<\/a><\/noindex><\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/www.dlology.com\">Blog Chengwei Zhang<\/a><\/noindex> aka <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/Tony607\">Tony607<\/a><\/noindex> s podrobn\u00fdmi pokyny a otev\u0159en\u00fdm zdrojov\u00fdm k\u00f3dem.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p><\/p>\n<p>D\u011bkujeme za pozornost!<\/p>\n<p>Zdroj: <a content=\"nofollow\" rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/482794\/\">habr.com<\/a><\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"<p>\u0421\u0442\u0430\u0442\u044c\u044f \u0441\u043e\u0441\u0442\u043e\u0438\u0442 \u0438\u0437 \u0434\u0432\u0443\u0445 \u0447\u0430\u0441\u0442\u0435\u0439: \u041a\u0440\u0430\u0442\u043a\u043e\u0435 \u043e\u043f\u0438\u0441\u0430\u043d\u0438\u0435 \u043d\u0435\u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440 \u0441\u0435\u0442\u0435\u0439 \u043f\u043e \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0435\u043d\u0438\u044e \u043e\u0431\u044a\u0435\u043a\u0442\u043e\u0432 \u043d\u0430 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0438 \u0438 \u0441\u0435\u0433\u043c\u0435\u043d\u0442\u0430\u0446\u0438\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0441 \u0441\u0430\u043c\u044b\u043c\u0438 \u043f\u043e\u043d\u044f\u0442\u043d\u044b\u043c\u0438 \u0434\u043b\u044f \u043c\u0435\u043d\u044f \u0441\u0441\u044b\u043b\u043a\u0430\u043c\u0438 \u043d\u0430 \u0440\u0435\u0441\u0443\u0440\u0441\u044b. \u0421\u0442\u0430\u0440\u0430\u043b\u0441\u044f \u0432\u044b\u0431\u0438\u0440\u0430\u0442\u044c \u0432\u0438\u0434\u0435\u043e \u043f\u043e\u044f\u0441\u043d\u0435\u043d\u0438\u044f \u0438 \u0436\u0435\u043b\u0430\u0442\u0435\u043b\u044c\u043d\u043e \u043d\u0430 \u0440\u0443\u0441\u0441\u043a\u043e\u043c \u044f\u0437\u044b\u043a\u0435. \u0412\u0442\u043e\u0440\u0430\u044f \u0447\u0430\u0441\u0442\u044c \u0441\u043e\u0441\u0442\u043e\u0438\u0442 \u0432 \u043f\u043e\u043f\u044b\u0442\u043a\u0435 \u043e\u0441\u043e\u0437\u043d\u0430\u0442\u044c \u043d\u0430\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u0435 \u0440\u0430\u0437\u0432\u0438\u0442\u0438\u044f \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440 \u043d\u0435\u0439\u0440\u043e\u043d\u043d\u044b\u0445 \u0441\u0435\u0442\u0435\u0439. \u0418 \u0442\u0435\u0445\u043d\u043e\u043b\u043e\u0433\u0438\u0439 \u043d\u0430 \u0438\u0445 \u043e\u0441\u043d\u043e\u0432\u0435. \u0420\u0438\u0441\u0443\u043d\u043e\u043a 1 \u2013 \u041f\u043e\u043d\u0438\u043c\u0430\u0442\u044c [&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-54780","post","type-post","status-publish","format-standard","hentry","category-news"],"aioseo_notices":[],"aioseo_head":"\n\t\t<!-- All in One SEO 5.0.3 - aioseo.com -->\n\t<meta name=\"description\" content=\"\u0421\u0442\u0430\u0442\u044c\u044f \u0441\u043e\u0441\u0442\u043e\u0438\u0442 \u0438\u0437 \u0434\u0432\u0443\u0445 \u0447\u0430\u0441\u0442\u0435\u0439: \u041a\u0440\u0430\u0442\u043a\u043e\u0435 \u043e\u043f\u0438\u0441\u0430\u043d\u0438\u0435 \u043d\u0435\u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440 \u0441\u0435\u0442\u0435\u0439 \u043f\u043e \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0435\u043d\u0438\u044e \u043e\u0431\u044a\u0435\u043a\u0442\u043e\u0432 \u043d\u0430 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0438 \u0438 \u0441\u0435\u0433\u043c\u0435\u043d\u0442\u0430\u0446\u0438\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0441 \u0441\u0430\u043c\u044b\u043c\u0438 \u043f\u043e\u043d\u044f\u0442\u043d\u044b\u043c\u0438 \u0434\u043b\u044f \u043c\u0435\u043d\u044f \u0441\u0441\u044b\u043b\u043a\u0430\u043c\u0438 \u043d\u0430 \u0440\u0435\u0441\u0443\u0440\u0441\u044b.\" \/>\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\/cs\/blog\/news\/nejroseti-kuda-eto-vse-dvizhetsya\" \/>\n\t\t<meta name=\"generator\" content=\"All in One SEO (AIOSEO) 5.0.3\" \/>\n\t\t<meta property=\"og:locale\" content=\"cs_CZ\" \/>\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\u041d\u0435\u0439\u0440\u043e\u0441\u0435\u0442\u0438. \u041a\u0443\u0434\u0430 \u044d\u0442\u043e \u0432\u0441\u0435 \u0434\u0432\u0438\u0436\u0435\u0442\u0441\u044f | ProHoster\" \/>\n\t\t<meta property=\"og:description\" content=\"\u0421\u0442\u0430\u0442\u044c\u044f \u0441\u043e\u0441\u0442\u043e\u0438\u0442 \u0438\u0437 \u0434\u0432\u0443\u0445 \u0447\u0430\u0441\u0442\u0435\u0439: \u041a\u0440\u0430\u0442\u043a\u043e\u0435 \u043e\u043f\u0438\u0441\u0430\u043d\u0438\u0435 \u043d\u0435\u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440 \u0441\u0435\u0442\u0435\u0439 \u043f\u043e \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0435\u043d\u0438\u044e \u043e\u0431\u044a\u0435\u043a\u0442\u043e\u0432 \u043d\u0430 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0438 \u0438 \u0441\u0435\u0433\u043c\u0435\u043d\u0442\u0430\u0446\u0438\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0441 \u0441\u0430\u043c\u044b\u043c\u0438 \u043f\u043e\u043d\u044f\u0442\u043d\u044b\u043c\u0438 \u0434\u043b\u044f \u043c\u0435\u043d\u044f \u0441\u0441\u044b\u043b\u043a\u0430\u043c\u0438 \u043d\u0430 \u0440\u0435\u0441\u0443\u0440\u0441\u044b.\" \/>\n\t\t<meta property=\"og:url\" content=\"https:\/\/prohoster.info\/cs\/blog\/news\/nejroseti-kuda-eto-vse-dvizhetsya\" \/>\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=\"2020-01-03T21:00:00+00:00\" \/>\n\t\t<meta property=\"article:modified_time\" content=\"2020-02-18T11:02:50+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\udd47Neur\u00e1ln\u00ed s\u00edt\u011b. Kam to v\u0161e sm\u011b\u0159uje | ProHoster","description":"\u010cl\u00e1nek se skl\u00e1d\u00e1 ze dvou \u010d\u00e1st\u00ed: Stru\u010dn\u00fd popis n\u011bkter\u00fdch architektur s\u00edt\u00ed pro detekci objekt\u016f na obr\u00e1zku a segmentaci obr\u00e1zk\u016f s nejjasn\u011bj\u0161\u00edmi odkazy na zdroje, kter\u00e9 jsem na\u0161el.","canonical_url":"https:\/\/prohoster.info\/cs\/blog\/news\/nejroseti-kuda-eto-vse-dvizhetsya","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"cs_CZ","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\u041d\u0435\u0439\u0440\u043e\u0441\u0435\u0442\u0438. \u041a\u0443\u0434\u0430 \u044d\u0442\u043e \u0432\u0441\u0435 \u0434\u0432\u0438\u0436\u0435\u0442\u0441\u044f | ProHoster","og:description":"\u0421\u0442\u0430\u0442\u044c\u044f \u0441\u043e\u0441\u0442\u043e\u0438\u0442 \u0438\u0437 \u0434\u0432\u0443\u0445 \u0447\u0430\u0441\u0442\u0435\u0439: \u041a\u0440\u0430\u0442\u043a\u043e\u0435 \u043e\u043f\u0438\u0441\u0430\u043d\u0438\u0435 \u043d\u0435\u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440 \u0441\u0435\u0442\u0435\u0439 \u043f\u043e \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0435\u043d\u0438\u044e \u043e\u0431\u044a\u0435\u043a\u0442\u043e\u0432 \u043d\u0430 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0438 \u0438 \u0441\u0435\u0433\u043c\u0435\u043d\u0442\u0430\u0446\u0438\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0441 \u0441\u0430\u043c\u044b\u043c\u0438 \u043f\u043e\u043d\u044f\u0442\u043d\u044b\u043c\u0438 \u0434\u043b\u044f \u043c\u0435\u043d\u044f \u0441\u0441\u044b\u043b\u043a\u0430\u043c\u0438 \u043d\u0430 \u0440\u0435\u0441\u0443\u0440\u0441\u044b.","og:url":"https:\/\/prohoster.info\/cs\/blog\/news\/nejroseti-kuda-eto-vse-dvizhetsya","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":"2020-01-03T21:00:00+00:00","article:modified_time":"2020-02-18T11:02:50+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"54780","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":"2026-01-24 12:43:20","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-02-28 19:57:38","updated":"2026-01-24 12:43:20","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts\/54780","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/comments?post=54780"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts\/54780\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/media?parent=54780"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/categories?post=54780"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/tags?post=54780"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}