{"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\/sq\/blog\/news\/lopnul-li-puzyr-mashinnogo-obucheniya-ili-nachalo-novoj-zari","title":{"rendered":"A shp\u00ebrtheu bula e m\u00ebsimit t\u00eb makinerive, apo \u00ebsht\u00eb fillimi i nj\u00eb agimi t\u00eb ri?","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

S\u00eb fundi doli artikulli<\/a><\/noindex>, e cila tregon mir\u00eb tendencat n\u00eb m\u00ebsimin makinerik t\u00eb viteve t\u00eb fundit. N\u00ebse flasim shkurt: numri i startup-eve n\u00eb fush\u00ebn e m\u00ebsimit makinerik ka r\u00ebn\u00eb ndjesh\u00ebm gjat\u00eb dy viteve t\u00eb fundit.<\/p>\n

\"A
\nE para, le t\u00eb shqyrtojm\u00eb n\u00ebse \u2018burrat e bubullim\u00ebs\u2019 u rrezikuan, \u2018si t\u00eb jetojm\u00eb m\u00eb tej\u2019 dhe t\u00eb flasim p\u00ebr nga erdhi fare kjo situat\u00eb.
\n<\/a><\/noindex>
\nP\u00ebr t\u00eb filluar, le t\u00eb flasim p\u00ebr at\u00eb q\u00eb ishte burimi i k\u00ebsaj curve. Nga erdhi ajo. Sigurisht q\u00eb t\u00eb gjith\u00eb do t\u00eb kujtojn\u00eb fitoren<\/a><\/noindex> e m\u00ebsimit makinerik n\u00eb vitin 2012 n\u00eb konkursin ImageNet. Sepse ky \u00ebsht\u00eb ngjarja e par\u00eb globale! Por n\u00eb t\u00eb v\u00ebrtet\u00eb nuk \u00ebsht\u00eb k\u00ebshtu. Po ashtu, rritja e k\u00ebsaj curve fillon disa vite m\u00eb par\u00eb. Do ta ndaja at\u00eb n\u00eb disa momente.<\/p>\n

    \n
  1. Viti 2008 sh\u00ebnon shfaqjen e termit \u201ct\u00eb dh\u00ebnat e m\u00ebdha\u201d. Produktet reale filluan t\u00eb shfaqen<\/a><\/noindex> nga viti 2010. T\u00eb dh\u00ebnat e m\u00ebdha jan\u00eb ngusht\u00ebsisht t\u00eb lidhura me m\u00ebsimin makinerik. Pa t\u00eb dh\u00ebna t\u00eb m\u00ebdha, funksionimi stabil i algoritmeve q\u00eb ekzistonin at\u00ebher\u00eb nuk ishte i mundur. Dhe k\u00ebto nuk ishin rrjete neuronale. Deri n\u00eb vitin 2012, rrjetet neuronale ishin nj\u00eb privilegj i nj\u00eb pakice margjinale. Megjithat\u00eb, at\u00ebher\u00eb filluan t\u00eb punonin algoritme krejt t\u00eb ndryshme, t\u00eb cilat kishin ekzistuar p\u00ebr vite, madje p\u00ebr dekada: SVM<\/a><\/noindex>(1963, 1993), Random Forest<\/a><\/noindex> (1995), AdaBoost<\/noindex> (2003),... Startupet e viteve ato ishin m\u00eb s\u00eb shumti t\u00eb lidhura me p\u00ebrpunimin automatik t\u00eb t\u00eb dh\u00ebnave t\u00eb strukturuara: kasa, p\u00ebrdoruesit, reklamat, shum\u00eb t\u00eb tjera.\n

    P\u00ebrderisa kjo ishte vala e par\u00eb, nj\u00eb grup framework-e, si XGBoost, CatBoost, LightGBM, etj.\n<\/li>\n

  2. N\u00eb vitet 2011-2012 rrjetet neuronale konvencionale<\/a><\/noindex> fituan disa konkurse p\u00ebr njohjen e imazheve. P\u00ebrdorimi i tyre real zgjati paksa. Do t\u00eb thoja q\u00eb startup-et dhe zgjidhjet e kuptueshme masivisht filluan t\u00eb shfaqeshin nga viti 2014. Nevojiteshin dy vjet p\u00ebr t\u00eb kuptuar se rrjetet neuronale n\u00eb t\u00eb v\u00ebrtet\u00eb funksiononin, p\u00ebr t\u00eb krijuar framework-e komod q\u00eb mund t\u00eb vendoseshin dhe aktivizoheshin brenda nj\u00eb kohe t\u00eb arsyeshme, p\u00ebr t\u00eb zhvilluar metoda q\u00eb do t\u00eb stabilizonin dhe p\u00ebrshpejtonin koh\u00ebn e konvergjenc\u00ebs.\n

    Rrjetet konvencionale lejuan zgjidhjen e problemeve t\u00eb vizionit makinerik: klasifikimi i imazheve dhe objekteve n\u00eb imazh, zbulimi i objekteve, njohja e objekteve dhe njer\u00ebzve, p\u00ebrmir\u00ebsimi i imazheve, etj., etj.<\/li>\n

  3. Viti 2015-2017. Boom i algoritmeve dhe projekteve t\u00eb lidhura me rrjetet recurrence ose variants e tyre (LSTM, GRU, TransformerNet, etj.). U shfaq\u00ebn algoritme q\u00eb funksiononin mir\u00eb p\u00ebr \"folje n\u00eb tekst\", sisteme p\u00ebr p\u00ebrkthimin automatik. Disa prej tyre bazohen n\u00eb rrjetet konvolucionale p\u00ebr t\u00eb nxjerr\u00eb karakteristikat themelore. Disa jan\u00eb rezultat i aft\u00ebsis\u00eb p\u00ebr t\u00eb mbledhur s\u00eb bashku dataset-e reale t\u00eb m\u00ebdha dhe t\u00eb mira. <\/li>\n<\/ol>\n

    \n\"A
    \n
    \n\"A shp\u00ebrtheu flluska? A \u00ebsht\u00eb hype i tep\u00ebrt? A vdiq\u00ebn si blockchain?\"
    \nSigurisht! Nes\u00ebr n\u00eb telefonin tuaj nuk do t\u00eb funksionoj\u00eb Siri, dhe pasnes\u00ebr Tesla nuk do ta dalloj\u00eb kthes\u00ebn nga nj\u00eb kengur.<\/p>\n

    Rrjetet neuronale jan\u00eb tashm\u00eb n\u00eb funksionim. Ato jan\u00eb n\u00eb dhjet\u00ebra pajisje. Realisht ato lejojn\u00eb fitime, ndryshojn\u00eb tregun dhe bot\u00ebn p\u00ebrreth. Hype duket disi ndryshe:<\/p>\n

    \"A
    \n
    \nSimple, rrjetet neuronale kan\u00eb ndaluar s\u00eb qeni di\u00e7ka e re. Po, shum\u00eb njer\u00ebz kan\u00eb pritshm\u00ebri t\u00eb tepruara. Por numri i madh i kompanive ka m\u00ebsuar se si t\u00eb aplikojn\u00eb rrjetet neuronale n\u00eb produktet e tyre. Rrjetet neuronale ofrojn\u00eb funksionalitete t\u00eb reja, reduktojn\u00eb vendet e pun\u00ebs, ulin \u00e7mimin e sh\u00ebrbimeve:<\/p>\n