Zviri Mukati[Viga][Ratidza]
Artificial Intelligence (AI) pakutanga yaifungidzirwa sechiroto chekure, tekinoroji yeramangwana, asi hazvisisirizvo.
Icho chaimbove chinyorwa chekutsvaga chave kuputika munyika chaiyo. AI yave kuwanikwa munzvimbo dzakasiyana siyana, kusanganisira kubasa kwako, chikoro, kubhengi, zvipatara, uye kunyange foni yako.
Ndiwo maziso emotokari dzinozvityaira, manzwi aSiri neaAlexa, pfungwa dziri kumashure kwekufanotaura kwemamiriro ekunze, maoko ari kumashure kwekuvhiyiwa kwakabatsirwa nerobhoti, nezvimwe.
chakagadzirwa njere (AI) iri kuita chinhu chakajairika chehupenyu hwazvino. Mumakore akati wandei apfuura, AI yakabuda semutambi mukuru mune dzakasiyana siyana dzeIT tekinoroji.
Pakupedzisira, iyo neural network inoshandiswa neAI kudzidza zvinhu zvitsva.
Saka nhasi tichadzidza nezve Neural Networks, kuti inoshanda sei, marudzi avo, maapplication, uye zvimwe zvakawanda.
Chii chinonzi Neural Network?
In machine learning, neural network isoftware-programmed network yeartificial neurons. Inoedza kutevedzera uropi hwomunhu nokuva nezvikamu zvakawanda zve“neuron,” zvakafanana netsinga dziri muuropi hwedu.
Iyo yekutanga layer ye neuroni inobvuma mafoto, vhidhiyo, ruzha, zvinyorwa, uye zvimwe zvinopinda. Iyi data inoyerera kuburikidza nematanho ese, paine imwe layer inobuda ichiyerera ichienda kune inotevera. Izvi zvakakosha kumabasa akaomesesa, akadai sekugadziriswa kwemutauro wechisikigo pakudzidza kwemichina.
Nekudaro, mune zvimwe zviitiko, kuvavarira kudzvanya system kudzikisa saizi yemuenzaniso uchichengetedza huchokwadi uye kushanda zvakanaka kunodiwa. Kuchekerera neural network inzira yekumanikidza inosanganisira kubvisa uremu kubva kune yakadzidzwa modhi. Funga nezve artificial intelligence neural network yakadzidziswa kusiyanisa vanhu nemhuka.
Mufananidzo wacho uchakamurwa kuita zvikamu zvinopenya uye zvakasviba neyokutanga layer ye neuroni. Iyi data ichapfuudzwa muchikamu chinotevera, icho chinoona kuti mipendero iripi.
Iyo inotevera layer inoedza kuona mafomu akagadzirwa nemupendero. Zvinoenderana nedata rayakadzidziswa, iyo data ichapfuura neakawanda akaturikidzana nenzira yakafanana kuona kana mufananidzo wawakaratidza uri wemunhu kana wemhuka.
Kana data yapihwa muneural network, inotanga kuigadzirisa. Mushure mezvo, iyo data inogadziriswa kuburikidza nemazinga ayo kuti iwane mhedzisiro yaunoda. A neural network muchina unodzidza kubva kune yakarongeka yekuisa uye inoratidza mhedzisiro. Kune marudzi matatu ekudzidza anogona kuitika mumaneural network:
- Kudzidza Kunotariswa - Zvinopinza uye zvinobuda zvinopihwa kune algorithms uchishandisa zvakanyorwa data. Mushure mekudzidziswa nzira yekuongorora data, vanofanotaura zvinotarisirwa.
- Kudzidza Kusina Kutariswa - An ANN inodzidza pasina rubatsiro rwemunhu. Iko hakuna data rakanyorwa, uye zvinobuda zvinosarudzwa nemapateni anowanikwa mune yakabuda data.
- Kusimbisa Kudzidza ndipo apo network inodzidza kubva kune mhinduro yainogamuchira.
Neural network inoshanda sei?
Artificial neurons inoshandiswa mumaneural network, ayo ari sophisticated systems. Iyo artificial neurons, inozivikanwawo seperceptrons, inoumbwa nezvinhu zvinotevera:
- chiyamuro
- uremu
- kutsvetera
- Activation Basa
- goho
Iwo matinji emaneuroni anoumba neural network. Neural network ine zvikamu zvitatu:
- Input layer
- Yakavanzika layer
- Output layer
Data iri muchimiro chenhamba yenhamba inotumirwa kune yekuisa layer. Network's hidden layers ndiwo anoita macalculation akawanda. Iyo yekubuda layer, yekupedzisira asi isiri diki, inofanotaura mhedzisiro. Maneuroni anotonga mumwe kune mumwe mune neural network. Neurons anoshandiswa kugadzira imwe neimwe layer. Iyo data inoendeswa kune yakavanzika layer mushure meiyo yekuisa layer yaiwana.
Huremu hunoshandiswa kune imwe neimwe yekuisa. Mukati mezvikamu zvakavanzwa zveneural network, huremu kukosha kunoturikira data inouya. Huremu hunoshanda nekuwanza data rekuisa nehuremu huremu muchikamu chekuisa.
Inobva yatanga kukosha kwekutanga kwakavanzwa. Iyo data yekupinza inoshandurwa uye inopfuudzwa kune imwe nhanho kuburikidza neakavanzika akaturikidzana. Iyo yekubuda layer ine basa rekugadzira yekupedzisira mhedzisiro. Izvo zvinopinza uye huremu zvinowedzerwa, uye mhedzisiro inounzwa kune yakavanzika layer neurons sehuwandu. Neuron imwe neimwe inopiwa kurerekera. Kuti uverenge huwandu, neuron yega yega inowedzera mapits ainogamuchira.
Mushure meizvozvo, kukosha kunopfuura kuburikidza ne activation function. Mhedzisiro ye activation function inosarudza kana neuron yaitwa kana kwete. Kana neuron ichishanda, inotumira ruzivo kune mamwe matinji. Iyo data inogadzirwa munetiweki kusvika neuron yasvika pakubuda layer ichishandisa nzira iyi. Forward propagation ndiyo imwe izwi reizvi.
Iyo tekinoroji yekudyisa data mune yekupinda node uye kuwana inobuda kuburikidza neinobuda node inozivikanwa sekudyisa-mberi kuparadzira. Kana iyo data yekupinza yakagamuchirwa neyakavanzika layer, feed-forward propagation inoitika. Iyo inogadziriswa zvinoenderana neiyo activation basa uye yozopfuura kune yakabuda.
Mhedzisiro yacho inofungidzirwa neneuron mune inobuda layer ine mukana wepamusoro. Backpropagation inoitika kana zvabuda zvisirizvo. Huremu hunotangwa kune yega yega yekuisa uchigadzira neural network. Backpropagation ndiyo maitiro ekugadzirisa huremu hwega yega yekuisa kuti uderedze zvikanganiso uye nekupa yakanyatso kuburitsa.
Mhando dzeNeural Network
1. Perceptron
Iyo Minsky-Pepa perceptron modhi ndeimwe yeakareruka uye yekare neuron modhi. Ndiyo diki diki yuniti yeneural network inoita mamwe maverengero kuitira kuti iwane hunhu kana hungwaru hwebhizinesi mune inouya data. Zvinotora huremu hwekupinza uye inoshandisa iyo activation basa kuti uwane yekupedzisira mhedzisiro. TLU (threshold logic unit) ndiro rimwe zita rekuti perceptron.
Perceptron ibhinari classifier iyo inotariswa yekudzidza sisitimu inokamura data kuita mapoka maviri. Logic Masuo zvakadai se AND, OR, and NAND zvinogona kuitwa nema perceptrons.
2. Feed-Forward Neural Network
Iyo inonyanya kukosha vhezheni yeneural network, umo data rekuisa rinoyerera chete munzira imwe, inopfuura neyakagadzirwa neural node uye ichibuda kuburikidza neinobuda node. Input and output layers dziripo munzvimbo dzakavanzika dzinogona kunge dziripo kana kuti dzisipo. Ivo vanogona kuratidzwa senge imwe-yakapetwa kana yakawanda-layered feed-mberi neural network yakavakirwa pane izvi.
Huwandu hwema layers anoshandiswa hunotariswa nekuoma kwebasa racho. Inongopararira mberi munzira imwe chete uye haiparadziri ichidzokera shure. Pano, uremu hunoramba huripo. Zvinopinza zvinowanzwa nezviremu zvekudyisa activation function. A classification activation function kana nhanho activation basa rinoshandiswa kuita izvi.
3. Multi-layer perceptron
Nhanganyaya kune yakaoma neural nets, umo data rekuisa rinofambiswa kuburikidza neakawanda akaturikidzana eaartificial neuron. Iyo yakanyatsobatanidzwa neural network, sezvo node yese yakabatana kune ese neuroni mune inotevera layer. Multiple akavanzika akaturikidzana, kureva, angangoita matatu kana anopfuura akaturikidzana, aripo mune yekupinza uye yekubuda layer.
Iine bidirectional propagation, zvinoreva kuti inogona kuparadzira zvese kumberi nekumashure. Zvinopinza zvinowedzerwa nehuremu uye zvinotumirwa kune activation basa, uko zvinoshandurwa kuburikidza nebackpropagation kuderedza kurasikirwa.
Huremu hunhu hwakadzidzwa nemuchina kubva kuNeural Networks, kuzvitaura zviri nyore. Zvichienderana nekusiyana pakati pezvinotarisirwa zvinobuda uye zvekudzidziswa, ivo vanozvigadzirisa. Softmax inoshandiswa seyekubuda layer activation basa mushure mekusaita mutsara activation mabasa.
4. Convolutional Neural Network
Mukupesana neyechinyakare-maviri-dimensional array, convolution neural network ine matatu-dimensional kumisikidzwa yemaneuroni. Yekutanga layer inozivikanwa seconvolutional layer. Neuron yega yega mune convolutional layer inongogadzirisa ruzivo kubva pachikamu chidiki chenzvimbo yekuona. Sesefa, maficha ekuisa anotorwa mubatch mode.
Iyo network inonzwisisa mapikicha muzvikamu uye inogona kuita zviito izvi kakawanda kuti ipedze iyo yese kugadzirisa mufananidzo.
Mufananidzo wacho unoshandurwa kubva kuRGB kana HSI kuita greyscale panguva yekugadzirisa. Kumwe kusiyanisa kukosha kwepixel kuchabatsira mukuona mipendero, uye mifananidzo inogona kurongwa mumapoka akati wandei. Unidirectional propagation inoitika kana CNN iine imwe kana akawanda convolutional layer inoteverwa nekubatanidza, uye bidirectional propagation inoitika kana kubuda kweiyo convolution layer inotumirwa kune yakazara yakabatana neural network yekuisa mufananidzo.
Kubvisa zvimwe zvinhu zvemufananidzo, mafirita anoshandiswa. MuMLP, izvo zvinoiswa zvinoyerwa uye zvinopihwa mukuita activation basa. RELU inoshandiswa mukutenderera, nepo MLP inoshandisa isina mutsara activation basa inoteverwa nesoftmax. Mumufananidzo uye kucherechedzwa kwevhidhiyo, semantic parsing, uye kuona paraphrase, convolutional neural network inoburitsa yakanakisa mhedzisiro.
5. Radial Bias Network
Vector yekupinza inoteverwa nerumwe rweRBF neurons uye yekubuda layer ine node imwe yechikamu chimwe nechimwe muRadial Basis Function Network. Iko kupinza kunoiswa mumapoka nekuienzanisa nedhata data kubva pakudzidziswa seti, uko neuron yega yega inochengetedza prototype. Uyu ndiwo mumwe wemienzaniso yekudzidziswa seti.
Neuron yega yega inoverengera nhambwe yeEuclidean pakati pekupinza uye prototype yayo apo vheta nyowani yekuisa [iyo n-dimensional vector yauri kuedza kuisa mumapoka] ichifanira kuiswa muchikamu. Kana isu tiine makirasi maviri, Kirasi A neKirasi B, iwo matsva ekuisa muchikamu akada kufanana nekirasi A prototypes pane ekirasi B prototypes.
Nekuda kweizvozvo, inogona kunyorwa kana kuiswa mukirasi A.
6. Recurrent Neural Network
Recurrent Neural Networks akagadzirirwa kuchengetedza zvakabuda zvechitubu uye wozozvidyisa kudzorera mune yekuisa kubatsira kufanotaura mhedzisiro ye layer. A feed-forward neural network inowanzova iyo yekutanga layer, inoteverwa neinodzokororwa neural network layer, apo basa rendangariro rinorangarira chikamu cheruzivo rwachaive nacho mudanho renguva yapfuura.
Iyi scenario inoshandisa kumberi kuparadzira. Inochengetedza data inozodiwa mune ramangwana. Muchiitiko chekuti kufanotaura kwakashata, chiyero chekudzidza chinoshandiswa kugadzirisa zvidiki. Nekuda kweizvozvo, sezvo iyo backpropagation inofambira mberi, ichawedzera kuve yakarurama.
Applications
Neural network inoshandiswa kubata matambudziko edata mumhando dzakasiyana siyana; mimwe mienzaniso inoratidzwa pasi apa.
- Kuzivikanwa Kwechiso - Kuzivikanwa Kwemeso Solutions inoshanda seyakanyatso ongorora masisitimu. Recognition system inobatanidza mafoto edhijitari nezviso zvevanhu. Iwo anoshandiswa mumahofisi ekusarudza kupinda. Nekudaro, masisitimu anosimbisa chiso chemunhu uye ochienzanisa nerondedzero yemaID akachengetwa mudura rayo.
- Stock Prediction - Investment inoratidzirwa kune njodzi dzemusika. Zvakangooma kufanoona zvichaitika mune ramangwana pamusika wemasheya. Pamberi peneural network, iyo inogara ichichinja bullish uye bearish zvikamu zvaive zvisingafungidzike. Asi, chii chakachinja zvese? Chokwadi, tiri kutaura nezveneural network… A Multilayer Perceptron MLP (rudzi rwefeedforward artificial intelligence system) inoshandiswa kugadzira kufanotaura kwemasheya kwakabudirira munguva chaiyo.
- Social Media -Kusinei nekuti corny inganzwika sei, social media yakachinja nzira yekuvapo. Maitiro evashandisi vesocial media anodzidzwa vachishandisa Artificial Neural Networks. Kuongororwa kwemakwikwi, data rinopihwa zuva nezuva kuburikidza nekudyidzana kwechokwadi rinounganidzwa uye kuongororwa. Zviito zvevashandisi vesocial media zvinodzokororwa neneural network. Maitiro emunhu anogona kubatana nemaitiro ekushandisa kwevanhu kana data yaongororwa kuburikidza nesocial media network. Dhata kubva pasocial media application inocherwa uchishandisa Multilayer Perceptron ANN.
- Hutano - Vanhu vari munyika yanhasi vari kushandisa mabhenefiti ehunyanzvi muindasitiri yehutano. Mubhizimusi rehutano, Convolutional Neural Networks inoshandiswa kuona X-ray, CT scans, uye Ultrasound. Iyo data yekufungidzira yekurapa yakagamuchirwa kubva kune yambotaurwa bvunzo inoongororwa uye inoongororwa uchishandisa neural network modhi, sezvo CNN inoshandiswa mukugadzirisa mifananidzo. Mukuvandudzwa kwemasisitimu ekuzivikanwa kwezwi, iyo recurrent neural network (RNN) inoshandiswa zvakare.
- Chirevo cheMamiriro ekunze - Kusati kwaitwa hungwaru hwekugadzira, fungidziro dzedhipatimendi remamiriro ekunze hadzina kumbobvira dzakanyatsojeka. Kufembera kwemamiriro ekunze kunoitwa zvakanyanya kufanotaura mamiriro ekunze achaitika mune ramangwana. Kufembera kwemamiriro ekunze kuri kushandiswa kutarisira mukana wenjodzi dzinongoitika dzoga munguva yazvino. Kufembera kwemamiriro ekunze kunoitwa pachishandiswa multilayer perceptron (MLP), convolutional neural networks (CNN), uye recurrent neural network (RNN).
- Dziviriro - Logistics, ongororo yekurwisa, uye nzvimbo yechinhu zvese zvinoshandisa neural network. Ivo zvakare vanoshandirwa mumhepo uye mugungwa patrols, pamwe nekutonga anozvimiririra drones. Artificial intelligence iri kupa indasitiri yezvidziviriro kukwidziridzwa kwainoda kukwirisa tekinoroji yayo. Pakuona kuvepo kwemigodhi yepasi pemvura, Convolutional Neural Networks (CNN) inoshandiswa.
Advantages
- Kunyangwe kana ma neurons mashoma muneural network asiri kushanda nemazvo, neural network inoramba ichiburitsa zvinobuda.
- Neural network ine kugona kudzidza munguva-chaiyo uye kujairana nekuchinja kwavo marongero.
- Neural network inogona kudzidza kuita akasiyana mabasa. Kupa mhinduro chaiyo zvichienderana nedata rakapihwa.
- Neural network ine simba uye kugona kubata akati wandei mabasa panguva imwe chete.
payakaipira
- Neural network inoshandiswa kugadzirisa matambudziko. Iyo haiburitse tsananguro yekuseri kwe "nei uye sei" yakaita mutongo wayakaita nekuda kwekuoma kwema network. Nekuda kweizvozvo, kuvimba kwetiweki kunogona kuderedzwa.
- Zvikamu zveneural network zvinoenderana pane chimwe. Kureva kuti, neural network inoda (kana inotsamira zvakanyanya pa) makomputa ane yakakwana computing simba.
- A neural network process haina mutemo chaiwo (kana mutemo wechigunwe). Muyedzo-uye-yekukanganisa maitiro, yakaringana network chimiro inotangwa nekuyedza iyo yakakwana network. Ndiwo maitiro anoda kuwanda kwekugadzirisa.
mhedziso
Munda we neural networks iri kuwedzera nokukurumidza. Izvo zvakakosha kuti udzidze uye unzwisise pfungwa dziri muchikamu ichi kuti ugone kubata nazvo.
Mhando dzakawanda dzeneural network dzakafukidzwa muchinyorwa chino. Iwe unogona kushandisa neural network kugadzirisa matambudziko edata mune mamwe minda kana iwe ukadzidza zvakawanda nezve ichi chirango.
Leave a Reply