{"id":1967,"date":"2022-11-13T14:35:04","date_gmt":"2022-11-13T13:35:04","guid":{"rendered":"https:\/\/mikulasskydenik.cz\/?p=1967"},"modified":"2023-04-01T17:09:00","modified_gmt":"2023-04-01T17:09:00","slug":"neuronove-site-jak-vlastne-funguji","status":"publish","type":"post","link":"https:\/\/denik.mikulasske.cz\/?p=1967","title":{"rendered":"Neuronov\u00e9 s\u00edt\u011b, jak vlastn\u011b funguj\u00ed?"},"content":{"rendered":"

Leccos je dnes ovlivn\u011bno neuronov\u00fdmi s\u00edt\u011bmi, na\u0161e aplikace pro komunikaci, r\u016fzn\u00e9 algoritmy personalizace, kter\u00e9 n\u00e1m dok\u00e1\u017eou kdykoliv doporu\u010dit to co zrovna hled\u00e1me, ale i v l\u00e9ka\u0159stv\u00ed, kde neuronov\u00e9 s\u00edt\u011b \u00fasp\u011b\u0161n\u011b detekuj\u00ed n\u00e1dory a dal\u0161\u00ed rakovinotvorn\u00e9 onemocn\u011bn\u00ed. Ale jak tyto algoritmy funguj\u00ed? A jak to \u017ee tak siln\u011b ovlivnily sm\u011b\u0159ov\u00e1n\u00ed modern\u00edho sv\u011bta?<\/p>\n

Pro\u010d je v\u016fbec m\u00e1me?<\/h2>\n

S postupem v\u00fdvoje po\u010d\u00edta\u010d\u016f (na kter\u00e9 jsem tady tak\u00e9 m\u011bl \u010dl\u00e1nek), se i za\u010daly zt\u011b\u017eovat \u00falohy kter\u00e9 m\u011bly samotn\u00e9 po\u010d\u00edta\u010de \u0159e\u0161it. Nap\u0159. jak chceme takov\u00fd po\u010d\u00edta\u010d nau\u010dit p\u0159edpov\u00eddat cenu zlata t\u00edm \u017ee analyzuje p\u0159edchoz\u00ed roky? Toto s na\u0161\u00edm „hloup\u00fdm“ po\u010d\u00edta\u010dem nesvedeme, pokud teda nepou\u017eijeme statistiku. Poj\u010fme si ale nejd\u0159\u00edve uk\u00e1zat n\u011bjak\u00e9 rozd\u011blen\u00ed t\u011bchto algoritm\u016f<\/p>\n

Machine learning, Deep learning a Reinforcement learning<\/h2>\n
\"\"
zdroj: https:\/\/www.edureka.co\/blog\/ai-vs-machine-learning-vs-deep-learning\/<\/figcaption><\/figure>\n

Samotn\u00fd proces tohoto „u\u010den\u00ed“, se d\u011bl\u00ed do n\u011bkolik \u010d\u00e1st\u00ed, a to: extrakce p\u0159\u00edznak\u016f, n\u00e1sledn\u00e9 vytvo\u0159en\u00ed testovac\u00ed a tr\u00e9novac\u00edho a testovac\u00edho datasetu, natr\u00e9nov\u00e1n\u00ed algoritmu a otestov\u00e1n\u00ed jak moc efektivn\u00ed je. Dataset je mno\u017eina dat ze kter\u00fdch se algoritmus u\u010d\u00ed, data obsahuj\u00ed input a output (kter\u00fd by m\u011bl na tr\u00e9novac\u00edch datech vyhodit)<\/p>\n

Deep learning<\/h5>\n
    \n
  • Pot\u0159ebuje mnohem v\u011bt\u0161\u00ed po\u010det prvk\u016f v datasetu<\/li>\n
  • Algoritmus se dok\u00e1\u017ee s\u00e1m u\u010dit<\/li>\n
  • Z\u00e1kladn\u00edm prvkem jsou neuronov\u00e9 s\u00edt\u011b, kter\u00e9 se skl\u00e1daj\u00ed z neuron\u016f<\/li>\n
  • Dok\u00e1\u017ee mnohem l\u00e9pe „pochopit“ data<\/li>\n
  • Hodn\u011b n\u00e1chyln\u00e1 k overfittov\u00e1n\u00ed*<\/li>\n
  • Dok\u00e1\u017ee \u0159e\u0161it probl\u00e9my klasifikace, regrese, clusteringu a dimensionality reduction<\/li>\n
  • D\u00edky n\u00e1ro\u010dn\u00e9mu tr\u00e9nov\u00e1n\u00ed pot\u0159ebujete GPU<\/li>\n<\/ul>\n
    Machine learning<\/h5>\n
      \n
    • M\u016f\u017ee se tr\u00e9novat i na men\u0161\u00edch datasetech<\/li>\n
    • Pot\u0159ebuje lehkou korekturu pr\u016fb\u011bhu tr\u00e9nov\u00e1n\u00ed od \u010dlov\u011bka<\/li>\n
    • Slo\u017eeny ze statistick\u00fdch metod, spole\u010dn\u011b s automatizac\u00ed t\u011bchto metod<\/li>\n
    • Nedok\u00e1\u017ee tak dob\u0159e „pochopit“ data<\/li>\n
    • Hodn\u011b n\u00e1chyln\u00e1 k underfittov\u00e1n\u00ed*<\/li>\n
    • Dok\u00e1\u017ee \u0159e\u0161it probl\u00e9my klasifikace, regrese, clusteringu a redukc\u00ed dimenze<\/li>\n
    • Zde v\u00e1m bohat\u011b sta\u010d\u00ed CPU<\/li>\n<\/ul>\n
      * tyto probl\u00e9my vysv\u011btl\u00edm v dal\u0161\u00edm \u010dl\u00e1nku<\/h5>\n
      Reinforcement learning<\/h5>\n

      Reinforcement learning (v \u010de\u0161tin\u011b zp\u011btnovazebn\u00ed u\u010den\u00ed) je speci\u00e1ln\u00ed druh tr\u00e9nov\u00e1n\u00ed um\u011bl\u00e9 inteligence. Tomuto t\u00e9matu se budu v\u011bnovat v jin\u00e9m \u010dl\u00e1nku, proto\u017ee v\u011b\u0159\u00edm \u017ee zrovna tento speci\u00e1ln\u00ed p\u0159\u00edpad u\u010den\u00ed si to zaslou\u017e\u00ed, tak\u017ee zde pouze nap\u00ed\u0161u stru\u010dn\u00fd \u00favod.<\/p>\n

      Tento styl tr\u00e9nov\u00e1n\u00ed je inspirov\u00e1n psychologi\u00ed, pr\u00e1v\u011b d\u00edky tomu, \u017ee \u010dlov\u011bk za sv\u016fj \u017eivot dostane hodn\u011b kritiky (a\u0165 u\u017e objektivn\u00ed nebo ne), kter\u00e1 dok\u00e1\u017ee ovlivnit jeho chov\u00e1n\u00ed. Na podobn\u00e9m principu funguje zp\u011btnovazebn\u00ed u\u010den\u00ed, d\u00e1me do jak\u00e9hosi virtu\u00e1ln\u00edho prost\u0159ed\u00ed n\u00e1\u0161 algoritmus, a n\u00e1sledn\u011b ho nech\u00e1me aby za\u010dal d\u011blat n\u011bkolik n\u00e1hodn\u00fdch rozhodnut\u00ed, kter\u00e1 mu bu\u010f daj\u00ed pozitivn\u00ed body, nebo negativn\u00ed body. D\u00edky tomuto dok\u00e1\u017eeme modelovat v\u00fdvoj dostate\u010dn\u011b inteligentn\u00edho organismu<\/p>\n

      Supervised vs Unsupervised learning<\/h2>\n

      Supervised<\/h5>\n

      Supervised learning (v \u010de\u0161tin\u011b u\u010den\u00ed s u\u010ditelem) je v p\u0159\u00edpad\u011b probl\u00e9mu klasifikace nebo regrese. Ka\u017ed\u00fd prvek v datasetu m\u00e1 p\u0159edem dan\u00fd vstup a v\u00fdstup aby se dok\u00e1zal algoritmus natr\u00e9novat na t\u011bchto prvc\u00edch a n\u00e1sledn\u011b d\u011blat vlastn\u00ed klasifikace\/predikce. V\u011bt\u0161inou jsou efektivn\u011bj\u0161\u00ed ne\u017e Unsupervised metody, ale zase pot\u0159ebuj\u00ed korekturu od \u010dlov\u011bka.<\/p>\n

      Unsupervised<\/h5>\n

      Unsupervised learning (v \u010de\u0161tin\u011b u\u010den\u00ed bez u\u010ditele) je v p\u0159\u00edpad\u011b probl\u00e9mu clusteringu, redukc\u00ed dimenze. Zde d\u00e1me algoritmu data bez \u017e\u00e1dn\u00fdch dal\u0161\u00edch specifikac\u00edch, p\u0159i\u010dem\u017e algoritmus mus\u00ed s\u00e1m zjistit vztahy mezi dan\u00fdmi hodnotami. Tyto algoritmy nevy\u017eaduj\u00ed korekturu od \u010dlov\u011bka, ale zase nejsou tolik efektivn\u00ed.<\/p>\n

      A te\u010f k Neuronov\u00e9 s\u00edti samotn\u00e9…<\/h2>\n
      \"\"
      Zdroj: https:\/\/towardsdatascience.com\/perceptron-the-artificial-neuron-4d8c70d5cc8d<\/figcaption><\/figure>\n

      Z\u00e1kladn\u00ed stavebn\u00ed prvek neuronov\u00e9 s\u00edt\u011b je neuron (v obr\u00e1zku je Perceptron, av\u0161ak princip je stejn\u00fd), neurony jsou prov\u00e1zan\u00e9 dohromady do neuronov\u00e9 s\u00edt\u011b. Ve vstupu m\u00e1 neuron p\u0159edchoz\u00ed v\u00fdstupy neuron\u016f<\/p>\n

      \"\"
      Zdroj: https:\/\/becominghuman.ai\/multi-layer-perceptron-mlp-models-on-real-world-banking-data-f6dd3d7e998f<\/figcaption><\/figure>\n","protected":false},"excerpt":{"rendered":"

      Leccos je dnes ovlivn\u011bno neuronov\u00fdmi s\u00edt\u011bmi, na\u0161e aplikace pro komunikaci, r\u016fzn\u00e9 algoritmy personalizace, kter\u00e9 n\u00e1m dok\u00e1\u017eou kdykoliv doporu\u010dit to co zrovna hled\u00e1me, ale i v l\u00e9ka\u0159stv\u00ed, kde neuronov\u00e9 s\u00edt\u011b \u00fasp\u011b\u0161n\u011b detekuj\u00ed n\u00e1dory a dal\u0161\u00ed rakovinotvorn\u00e9 onemocn\u011bn\u00ed. Ale jak tyto algoritmy funguj\u00ed? A jak to \u017ee tak siln\u011b ovlivnily sm\u011b\u0159ov\u00e1n\u00ed modern\u00edho sv\u011bta? Pro\u010d je v\u016fbec m\u00e1me?<\/p>\n","protected":false},"author":8,"featured_media":1992,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[61],"tags":[169,202,236],"acf":[],"_links":{"self":[{"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=\/wp\/v2\/posts\/1967"}],"collection":[{"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1967"}],"version-history":[{"count":0,"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=\/wp\/v2\/posts\/1967\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=\/wp\/v2\/media\/1992"}],"wp:attachment":[{"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1967"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1967"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/denik.mikulasske.cz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1967"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}