{"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\/ro\/blog\/news\/nejroseti-kuda-eto-vse-dvizhetsya","title":{"rendered":"Re\u021bele neurale. \u00cencotro se \u00eendreapt\u0103 toate acestea","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"<p>Articolul este alc\u0103tuit din dou\u0103 p\u0103r\u021bi:<\/p>\n<p><\/p>\n<ol>\n<li>O scurt\u0103 descriere a unor arhitecturi de re\u021bea pentru detectarea obiectelor \u00een imagini \u0219i segmentarea imaginilor, cu cele mai u\u0219or de \u00een\u021beles linkuri c\u0103tre resurse. Am \u00eencercat s\u0103 selectez explica\u021bii video, preferabil \u00een limba rom\u00e2n\u0103.<\/li>\n<li>A doua parte const\u0103 \u00een \u00eencercarea de a \u00een\u021belege direc\u021bia de dezvoltare a arhitecturilor re\u021belelor neuronale \u0219i a tehnologiilor bazate pe acestea.<\/li>\n<\/ol>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Re\u021bele neurale. \u00cencotro se \u00eendreapt\u0103 toate acestea\" src=\"\/wp-content\/uploads\/2020\/01\/3e0238547dc956dbb069c11241e4534f.png\" style=\"display:block;margin: 0 auto;\" \/><\/p>\n<p><\/p>\n<p>Figura 1 \u2013 \u00cen\u021belegerea arhitecturilor re\u021belelor neuronale nu este simpl\u0103<\/p>\n<p><\/p>\n<p>Totul a \u00eenceput cu realizarea a dou\u0103 aplica\u021bii demonstrative pentru clasificarea \u0219i detectarea obiectelor pe telefonul Android:<\/p>\n<p><\/p>\n<ul>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/junkyard\/tree\/master\/object_classifier\">Demo back-end<\/a><\/noindex>, c\u00e2nd datele sunt procesate pe server \u0219i transmise pe telefon. Clasificarea imaginilor (image classification) pentru trei tipuri de ur\u0219i: brun, negru \u0219i de plu\u0219.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/android\/tree\/master\/object_detection_demo\">Demo front-end<\/a><\/noindex>, c\u00e2nd datele sunt procesate direct pe telefon. Detectarea obiectelor (object detection) pentru trei tipuri: alune, smochine \u0219i curmale.<\/li>\n<\/ul>\n<p><noindex><a rel=\"nofollow\" name=\"habracut\"><\/a><\/noindex><\/p>\n<p>Exist\u0103 o diferen\u021b\u0103 \u00eentre sarcinile de clasificare a imaginilor, detectarea obiectelor \u00een imagini \u0219i <noindex><a rel=\"nofollow\" href=\"https:\/\/medium.com\/analytics-vidhya\/image-classification-vs-object-detection-vs-image-segmentation-f36db85fe81\">segmentarea imaginilor<\/a><\/noindex>. De aceea, a ap\u0103rut necesitatea de a afla care arhitecturi de re\u021bele neuronale pot detecta obiecte \u00een imagini \u0219i care pot segmenta. Am g\u0103sit urm\u0103toarele exemple de arhitecturi cu cele mai clare linkuri c\u0103tre resurse:<\/p>\n<p><\/p>\n<ul>\n<li>O serie de arhitecturi bazate pe 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>. Pentru detectarea obiectelor \u00een imagini, prin mecanismul Region Proposal Network (RPN), sunt selectate regiuni delimitate (bounding boxes). Ini\u021bial, \u00een loc de RPN a fost utilizat un mecanism mai lent, Selective Search. Apoi, regiunile delimitate sunt furnizate ca input unei re\u021bele neuronale obi\u0219nuite pentru clasificare. \u00cen arhitectura R-CNN exist\u0103 cicluri explicite 'for' pentru a itera pe regiuni delimitate, cu un total de p\u00e2n\u0103 la 2000 de treceri prin re\u021beaua intern\u0103 AlexNet. Datorit\u0103 ciclurilor explicite 'for', viteza de procesare a imaginilor este \u00eencetinit\u0103. Num\u0103rul de cicluri explicite, treceri prin re\u021beaua intern\u0103, se reduce cu fiecare nou\u0103 versiune a arhitecturii \u0219i se fac zeci de alte modific\u0103ri pentru a spori viteza \u0219i a schimba sarcina de detectare a obiectelor \u00een segmentarea obiectelor \u00een Mask R-CNN.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/L0tzmv--CGY\">YOLO<\/a><\/noindex> (<strong>Y<\/strong>ou <strong>O<\/strong>nly <strong>L<\/strong>ook <strong>O<\/strong>nce) \u2013 prima re\u021bea neuronal\u0103 care recunoa\u0219te obiectele \u00een timp real pe dispozitive mobile. Caracteristica distinctiv\u0103: diferen\u021bierea obiectelor cu o singur\u0103 trecere (este suficient s\u0103 se priveasc\u0103 o singur\u0103 dat\u0103). A\u0219adar, \u00een arhitectura YOLO nu exist\u0103 cicluri \u201efor\u201d evidente, ceea ce face ca re\u021beaua s\u0103 func\u021bioneze rapid. De exemplu, o analogie: \u00een NumPy, la opera\u021bii cu matrice, de asemenea, nu exist\u0103 cicluri \u201efor\u201d evidente, care sunt realizate \u00een NumPy la niveluri mai joase ale arhitecturii prin limbajul de programare C. YOLO folose\u0219te o re\u021bea de feronier\u0103 predefinite. Pentru a evita identificarea multipl\u0103 a aceluia\u0219i obiect, se utilizeaz\u0103 un coeficient de suprapunere a feronierelor (IoU, <strong>I<\/strong>ntersection <strong>o<\/strong>ver <strong>U<\/strong>nion). Aceast\u0103 arhitectur\u0103 func\u021bioneaz\u0103 pe un interval larg \u0219i are o <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\">robustez\u0103<\/a><\/noindex>: modelul poate fi antrenat pe fotografii, dar \u00een acela\u0219i timp func\u021bioneaz\u0103 bine pe tablouri desenate.<\/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>D<\/strong>etector) \u2013 sunt utilizate cele mai reu\u0219ite \u201etrucuri\u201d ale arhitecturii YOLO (de exemplu, non-maximum suppression) \u0219i se adaug\u0103 altele noi, pentru ca re\u021beaua neuronal\u0103 s\u0103 func\u021bioneze mai rapid \u0219i mai precis. Caracteristica distinctiv\u0103: diferen\u021bierea obiectelor cu o singur\u0103 trecere printr-o re\u021bea predeterminat\u0103 de feronier\u0103 (default box) pe o piramid\u0103 de imagini. Piramida de imagini este codificat\u0103 \u00een tensori convolu\u021bionali \u00een timpul opera\u021biunilor consecutive de convolu\u021bie \u0219i pooling (\u00een timpul opera\u021biei max-pooling, dimensiunea spa\u021bial\u0103 scade). Astfel, se definesc at\u00e2t obiecte mari, c\u00e2t \u0219i mici \u00eentr-o singur\u0103 trecere a re\u021belei.<\/li>\n<li>MobileSSD (<strong>Mobile<\/strong>NetV2 + <strong>SSD<\/strong>) \u2013 o combina\u021bie din dou\u0103 arhitecturi de re\u021bele neuronale. Prima re\u021bea <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/352804\/\">MobileNetV2<\/a><\/noindex> func\u021bioneaz\u0103 rapid \u0219i cre\u0219te precizia recunoa\u0219terii. MobileNetV2 este utilizat \u00een locul VGG-16, care a fost utilizat ini\u021bial \u00een <noindex><a rel=\"nofollow\" href=\"https:\/\/arxiv.org\/abs\/1512.02325\">articolului original<\/a><\/noindex>. A doua re\u021bea SSD determin\u0103 loca\u021bia obiectelor \u00een imagine.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/ge_RT5wvHvY\">SqueezeNet<\/a><\/noindex> \u2013 o re\u021bea neuronal\u0103 foarte mic\u0103, dar precis\u0103. Ea nu rezolv\u0103 singur\u0103 problema detect\u0103rii obiectelor. Totu\u0219i, poate fi utilizat\u0103 \u00een combina\u021bia diferitelor arhitecturi. \u0218i utilizat\u0103 pe dispozitive mobile. Caracteristica distinctiv\u0103 este c\u0103 datele sunt comprimate mai \u00eent\u00e2i la patru filtre convolu\u021bionale 1\u00d71, iar apoi sunt extinse la patru 1\u00d71 \u0219i patru 3\u00d73 filtre convolu\u021bionale. O astfel de itera\u021bie de comprimare-extindere a datelor se nume\u0219te \u201eFire Module\u201d.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/b6jhopSMit8\">DeepLab<\/a><\/noindex> (Segmentarea semantic\u0103 a imaginilor cu re\u021bele neuronale convolu\u021bionale profunde) \u2013 segmentarea obiectelor dintr-o imagine. Caracteristica distinctiv\u0103 a arhitecturii este convolu\u021bia dilatat\u0103, care p\u0103streaz\u0103 rezolu\u021bia spa\u021bial\u0103. Urmeaz\u0103 o etap\u0103 de procesare ulterioar\u0103 a rezultatelor folosind un model grafic de probabilitate, ceea ce permite eliminarea zgomotelor minore \u00een segmentare \u0219i \u00eembun\u0103t\u0103\u021birea calit\u0103\u021bii imaginii segmentate. Sub denumirea temut\u0103 de \u201emodel grafic probabilistic\u201d se ascunde un simplu filtru Gaussian, care este aproximat pe cinci puncte.<\/li>\n<li>Am \u00eencercat s\u0103 \u00een\u021beleg structura <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>ec\u021bie), dar n-am \u00een\u021beles prea multe.<\/li>\n<li>De asemenea, am urm\u0103rit cum func\u021bioneaz\u0103 tehnologia \u201eaten\u021bie\u201d: <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>. Caracteristica distinctiv\u0103 a arhitecturii \u201eaten\u021bie\u201d este capacitatea de a eviden\u021bia automat regiunile de interes ridicat din imagine (RoI, <strong>R<\/strong>regiuni <strong>o<\/strong>f <strong>I<\/strong>de interes) utiliz\u00e2nd o re\u021bea neuronal\u0103 numit\u0103 Attention Unit. Regiunile de interes ridicat seam\u0103n\u0103 cu regiunile delimitate (bounding boxes), dar spre deosebire de acestea, nu sunt fixe \u00een imagine \u0219i pot avea margini neclare. Apoi, din regiunile de interes ridicat sunt extrase caracteristici (features), care sunt \u201ealimentate\u201d re\u021belelor neuronale recurente cu arhitecturi <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/5lUUrREboSk\">LSDM, GRU sau Vanilla RNN<\/a><\/noindex>. Re\u021belele neuronale recurente sunt capabile s\u0103 analizeze rela\u021biile dintre caracteristici \u00een secven\u021b\u0103. Aceste re\u021bele erau ini\u021bial folosite pentru traducerea textului \u00een alte limbi, iar acum \u0219i pentru traducerea <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/e-WB4lfg30M\">imaginii \u00een text<\/a><\/noindex> \u0219i <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/rAbhypxs1qQ\">textului \u00een imagine<\/a><\/noindex>.<\/li>\n<\/ul>\n<p><\/p>\n<p>Pe m\u0103sur\u0103 ce am studiat aceste arhitecturi <strong>am realizat c\u0103 nu \u00een\u021beleg nimic.<\/strong>. \u0218i nu este vorba c\u0103 re\u021beaua mea neuronal\u0103 are probleme cu mecanismul de aten\u021bie. Crearea tuturor acestor arhitecturi pare a fi un imens hackathon, unde autorii concureaz\u0103 \u00een hack-uri. Hack (hack) \u2013 o solu\u021bie rapid\u0103 pentru o problem\u0103 software dificil\u0103. Cu alte cuvinte, \u00eentre toate aceste arhitecturi nu exist\u0103 o leg\u0103tur\u0103 logic\u0103 vizibil\u0103 \u0219i \u00een\u021beleas\u0103. Tot ce le une\u0219te este un set de hack-uri de succes pe care le \u00eemprumut\u0103 unul de la altul, plus o opera\u021bie comun\u0103 de convolu\u021bie cu feedback <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/Ilg3gGewQ5U\">(propagarea invers\u0103 a erorii). Nu exist\u0103<\/a><\/noindex> g\u00e2ndire sistemic\u0103 <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/272473\/\">! Nu este clar ce trebuie schimbat \u0219i cum s\u0103 se optimizeze realiz\u0103rile existente.<\/a><\/noindex>! Nu este clar ce s\u0103 schimb\u0103m \u0219i cum s\u0103 optimiz\u0103m realiz\u0103rile existente.<\/p>\n<p><\/p>\n<p>Ca rezultat al lipsei unei rela\u021bii logice \u00eentre hack-uri, acestea sunt extrem de greu de re\u021binut \u0219i aplicat \u00een practic\u0103. Acestea sunt cuno\u0219tin\u021be fragmentate. \u00cen cel mai bun caz, c\u00e2teva momente interesante \u0219i nea\u0219teptate sunt re\u021binute, dar majoritatea a ceea ce este \u00een\u021beles \u0219i incomprehensibil dispare din memorie dup\u0103 c\u00e2teva zile. Va fi bine dac\u0103 \u00eentr-o s\u0103pt\u0103m\u00e2n\u0103 \u00ee\u0219i va aduce aminte m\u0103car denumirea arhitecturii. \u0218i totu\u0219i, citirea articolelor \u0219i vizionarea videoclipurilor de prezentare au durat c\u00e2teva ore \u0219i chiar zile de timp de lucru!<\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Re\u021bele neurale. \u00cencotro se \u00eendreapt\u0103 toate acestea\" src=\"\/wp-content\/uploads\/2020\/01\/86f0f24e3be1d0f1a210f3132897d981.png\" style=\"display:block;margin: 0 auto;\" \/><\/p>\n<p><\/p>\n<p>Figura 2 \u2013 <noindex><a rel=\"nofollow\" href=\"https:\/\/www.asimovinstitute.org\/neural-network-zoo\/\">Zoologicul re\u021belelor neuronale<\/a><\/noindex><\/p>\n<p><\/p>\n<p>Majoritatea autorilor de articole \u0219tiin\u021bifice, \u00een opinia mea personal\u0103, fac tot posibilul pentru ca aceste cuno\u0219tin\u021be fragmentate s\u0103 nu fie \u00een\u021belese de cititor. Dar formalele de gerunziu \u00een propozi\u021bii de zece r\u00e2nduri cu formule, luate \u00abde pe plafon\u00bb \u2013 este un subiect pentru un articol separat (problema <noindex><a rel=\"nofollow\" href=\"https:\/\/en.wikipedia.org\/wiki\/Publish_or_perish\">publish or perish<\/a><\/noindex>).<\/p>\n<p><\/p>\n<p>Din acest motiv, a ap\u0103rut necesitatea de a sistematiza informa\u021biile despre re\u021bele neuronale \u0219i, astfel, de a cre\u0219te calitatea \u00een\u021belegerii \u0219i memor\u0103rii. Prin urmare, tema principal\u0103 a analizei tehnologiilor \u0219i arhitecturilor re\u021belelor neuronale artificiale a devenit urm\u0103toarea sarcin\u0103: <strong>a afla \u00eencotro se \u00eendreapt\u0103 toate acestea<\/strong>, \u0219i nu construirea unei re\u021bele neuronale specifice \u00een mod separat.<\/p>\n<p><\/p>\n<p>\u00cencotro se \u00eendreapt\u0103 toate acestea. Principalele rezultate:<\/p>\n<p><\/p>\n<ul>\n<li>Num\u0103rul startup-urilor \u00een domeniul \u00eenv\u0103\u021b\u0103rii automate \u00een ultimii doi ani <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/recognitor\/blog\/455676\/\">a sc\u0103zut drastic<\/a><\/noindex>. Cauza posibil\u0103: \u00abre\u021belele neuronale nu mai sunt ceva nou\u00bb.<\/li>\n<li>Fiecare va putea crea o re\u021bea neuronal\u0103 func\u021bional\u0103 pentru a rezolva o sarcin\u0103 simpl\u0103. Pentru aceasta, va lua un model gata realizat din \u00abzoologicul modelelor\u00bb (model zoo) \u0219i va antrena ultimul strat al re\u021belei neuronale (<noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/yofjFQddwHE\">transfer learning<\/a><\/noindex>) pe datele gata disponibile din <noindex><a rel=\"nofollow\" href=\"https:\/\/toolbox.google.com\/datasetsearch\">Google Dataset Search<\/a><\/noindex> sau din <noindex><a rel=\"nofollow\" href=\"https:\/\/www.kaggle.com\/datasets\">25 de mii de dataset-uri Kaggle<\/a><\/noindex> \u00een cloud-ul gratuit <noindex><a rel=\"nofollow\" href=\"https:\/\/www.dataschool.io\/cloud-services-for-jupyter-notebook\/\">Jupyter Notebook<\/a><\/noindex>.<\/li>\n<li>Mari produc\u0103tori de re\u021bele neuronale au \u00eenceput s\u0103 creeze <strong>\u00abzoologice de modele\u00bb<\/strong> (model zoo). Cu ajutorul lor, se poate crea rapid o aplica\u021bie comercial\u0103: <noindex><a rel=\"nofollow\" href=\"https:\/\/tfhub.dev\/\">TF Hub<\/a><\/noindex> pentru TensorFlow, <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/open-mmlab\/mmdetection\">MMDetection<\/a><\/noindex> pentru PyTorch, <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/facebookresearch\/Detectron\">Detectron<\/a><\/noindex> pentru Caffe2, <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/wkentaro\/chainer-modelzoo\">chainer-modelzoo<\/a><\/noindex> pentru Chainer \u0219i <noindex><a rel=\"nofollow\" href=\"https:\/\/modelzoo.co\/\">altele<\/a><\/noindex>.<\/li>\n<li>Re\u021bele neuronale func\u021bion\u00e2nd \u00een <strong>timp real<\/strong> (real-time) pe dispozitive mobile. De la 10 la 50 de cadre pe secund\u0103.<\/li>\n<li>Aplicarea re\u021belelor neuronale \u00een telefoane (TF Lite), \u00een browsere (TF.js) \u0219i \u00een <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/19ZNz2N79u4\">obiecte de uz casnic<\/a><\/noindex> (IoT, <strong>I<\/strong>Internet <strong>o<\/strong>f <strong>T<\/strong>things). \u00cen special \u00een telefoane, care deja suport\u0103 re\u021bele neuronale la nivel de \u201ehardware\u201d (neuroacceleratoare).<\/li>\n<li>Fiecare dispozitiv, articole de \u00eembr\u0103c\u0103minte \u0219i, posibil, chiar \u0219i alimentele vor avea <strong>adres\u0103 IP-v6<\/strong> \u0219i se vor comunica \u00eentre ele\u201d \u2013 <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/GG7H8Xa4m8I?t=85\">Sebastian Thrun<\/a><\/noindex>.<\/li>\n<li>Cre\u0219terea num\u0103rului de publica\u021bii \u00een domeniul \u00eenv\u0103\u021b\u0103rii automate a \u00eenceput <noindex><a rel=\"nofollow\" href=\"http:\/\/data-mining.philippe-fournier-viger.com\/too-many-machine-learning-papers\">s\u0103 dep\u0103\u0219easc\u0103 legea lui Moore<\/a><\/noindex> (dubl\u00e2ndu-se la fiecare doi ani) din 2015. Este evident c\u0103 sunt necesare re\u021bele neuronale pentru analiza articolelor.<\/li>\n<li>Tehnologiile urm\u0103toare c\u00e2\u0219tig\u0103 popularitate:\n<ul>\n<li><strong>PyTorch<\/strong> \u2013 popularitatea cre\u0219te rapid \u0219i, se pare, o dep\u0103\u0219e\u0219te pe TensorFlow.<\/li>\n<li>Selectarea automat\u0103 a hiperparametrilor <strong>AutoML<\/strong> \u2013 popularitatea cre\u0219te treptat.<\/li>\n<li>Sc\u0103derea gradual\u0103 a preciziei \u0219i cre\u0219terea vitezei de calcul: <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/rln_kZbYaWc\">logica fuzzy<\/a><\/noindex>, algoritmi <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/MIPkK5ZAsms\">de boosting<\/a><\/noindex>, calcule inexacte (aproximative), cuantificare (atunci c\u00e2nd greut\u0103\u021bile re\u021belei neuronale sunt convertite \u00een numere \u00eentregi \u0219i cuantificate), neuroacceleratoare.<\/li>\n<li>Traducere <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/e-WB4lfg30M\">imaginii \u00een text<\/a><\/noindex> \u0219i <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/rAbhypxs1qQ\">textului \u00een imagine<\/a><\/noindex>.<\/li>\n<li>Crearea <noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/OrHLacCDZVQ\">de obiecte 3D din video<\/a><\/noindex>, acum deja \u00een timp real.<\/li>\n<li>Principalul \u00een DL este c\u0103 sunt multe date, dar colectarea \u0219i etichetarea lor nu este u\u0219oar\u0103. De aceea, se dezvolt\u0103 automatizarea etichet\u0103rii (<noindex><a rel=\"nofollow\" href=\"https:\/\/youtu.be\/NcKTn4C91Yc\">annotarea automatizat\u0103<\/a><\/noindex>) pentru re\u021bele neuronale folosind re\u021bele neuronale.<\/li>\n<\/ul>\n<\/li>\n<li>Cu re\u021bele neuronale, informatica a devenit brusc <strong>o \u0219tiin\u021b\u0103 experimental\u0103<\/strong> \u0219i a ap\u0103rut <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/480348\">criza reproducibilit\u0103\u021bii<\/a><\/noindex>.<\/li>\n<li>Banii din IT \u0219i popularitatea re\u021belelor neuronale au ap\u0103rut simultan, c\u00e2nd calculul a devenit o valoare de pia\u021b\u0103. Economia din aur-valut\u0103 devine <strong>aur-valut\u0103-computational\u0103.<\/strong>Consulta\u021bi articolul meu despre <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\">eco-fizic\u0103<\/a><\/noindex> \u0219i motivul apari\u021biei banilor \u00een IT.<\/li>\n<\/ul>\n<p><\/p>\n<p>Treptat, apare o nou\u0103 <noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/481844\">metodologie de programare ML\/DL<\/a><\/noindex> (\u00cenv\u0103\u021bare Automat\u0103 &amp; \u00cenv\u0103\u021bare Profund\u0103), care se bazeaz\u0103 pe reprezentarea unui program ca un ansamblu de modele neuronale antrenate.<\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Re\u021bele neurale. \u00cencotro se \u00eendreapt\u0103 toate acestea\" src=\"\/wp-content\/uploads\/2020\/01\/17df76a613191522145abeba177c2685.png\" style=\"display:block;margin: 0 auto;\" \/><\/p>\n<p><\/p>\n<p>Figura 3 \u2013 ML\/DL ca nou\u0103 metodologie de programare<\/p>\n<p><\/p>\n<p>Cu toate acestea, nu a ap\u0103rut <strong>\u201eteoria re\u021belelor neuronale\u201d<\/strong>, \u00een cadrul c\u0103reia se poate g\u00e2ndi \u0219i lucra sistematic. Ceea ce se nume\u0219te acum \u201eteorie\u201d este de fapt algoritmi experimentali, euristici.<\/p>\n<p><\/p>\n<p>Linkuri c\u0103tre resursele mele \u0219i nu numai:<\/p>\n<p><\/p>\n<ul>\n<li>Newsletter privind \u0219tiin\u021ba datelor. \u00cen principal, despre procesarea imaginilor. Cine dore\u0219te s\u0103 primeasc\u0103, s\u0103 trimit\u0103 e-mail (foobar167gmailcom). Trimitem linkuri c\u0103tre articole \u0219i videoclipuri pe m\u0103sur\u0103 ce acumul\u0103m material.<\/li>\n<li>List\u0103 general\u0103 <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/articles\/blob\/master\/Machine_Learning\/courses_on_machine_learning.md\">de cursuri \u0219i articole<\/a><\/noindex>, pe care le-am parcurs \u0219i pe care mi-a\u0219 dori s\u0103 le parcurg.<\/li>\n<li><noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/foobar167\/articles\/blob\/master\/Ubuntu\/13_Keras_and_TensorFlow_how-tos.md#exercises\">Cursuri \u0219i videoclipuri pentru \u00eencep\u0103tori<\/a><\/noindex>, de la care merit\u0103 s\u0103 \u00eencepem s\u0103 \u00eenv\u0103\u021b\u0103m despre re\u021bele neuronale. Plus un ghid <noindex><a rel=\"nofollow\" href=\"https:\/\/foobar167.github.io\/page\/vvedeniye-v-mashinnoye-obucheniye-i-iskusstvennyye-neyronnyye-seti.html\">\u201eIntroducere \u00een \u00eenv\u0103\u021barea automat\u0103 \u0219i re\u021bele neuronale artificiale\u201d<\/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\">Instrumente utile<\/a><\/noindex>, unde fiecare va g\u0103si ceva interesant pentru sine.<\/li>\n<li>S-au dovedit a fi extrem de utile <strong>canalele video care analizeaz\u0103 articole \u0219tiin\u021bifice<\/strong> despre Data Science. C\u0103uta\u021bi-le, abona\u021bi-v\u0103 la ele \u0219i transmite\u021bi link-uri colegilor vo\u0219tri \u0219i mie de asemenea. Exemple:\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\">Blogul lui Chengwei Zhang<\/a><\/noindex> cu <noindex><a rel=\"nofollow\" href=\"https:\/\/github.com\/Tony607\">Tony607<\/a><\/noindex> instruc\u021biuni pas cu pas \u0219i cod surs\u0103 deschis.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p><\/p>\n<p>V\u0103 mul\u021bumesc pentru aten\u021bie!<\/p>\n<p>Sursa: <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 5.0.2.1 - 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\/ro\/blog\/news\/nejroseti-kuda-eto-vse-dvizhetsya\" \/>\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\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\/ro\/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\udd47Re\u021bele neuronale. \u00cen ce direc\u021bie se \u00eendreapt\u0103 totul | ProHoster","description":"Articolul const\u0103 \u00een dou\u0103 p\u0103r\u021bi: O descriere scurt\u0103 a unor arhitecturi de re\u021bea pentru detectarea obiectelor \u00een imagini \u0219i segmentarea imaginilor cu cele mai clare link-uri pentru mine c\u0103tre resurse.","canonical_url":"https:\/\/prohoster.info\/ro\/blog\/news\/nejroseti-kuda-eto-vse-dvizhetsya","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\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\/ro\/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\/ro\/wp-json\/wp\/v2\/posts\/54780","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=54780"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/54780\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media?parent=54780"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/categories?post=54780"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/tags?post=54780"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}