Hurongwa hwekudzidza kwakadzama hunosanganisira musanganiswa wenzvimbo, maraibhurari uye maturusi ekutsanangura nekudzidzisa mamodheru eChidzidzo cheMichina nekukurumidza uye nemazvo.
Nekuti kudzidza kwakadzama kunoshandisa huwandu hukuru hwe data isina kurongeka, isiri-yezvinyorwa, iwe unoda chimiro chinodzora kudyidzana pakati pe "matanho" uye inoita kusimudzira kwemuenzaniso nekukurumidza nekudzidza kubva kune data rekuisa uye kuita sarudzo dzekuzvimiririra.
Kana iwe uchifarira kudzidza nezve kudzidza kwakadzama muna 2021, funga kushandisa imwe yematanho anoratidzwa pazasi. Rangarira kusarudza imwe ichakubatsira kuzadzisa zvinangwa uye chiono chako.
1. TensorFlow
Kana tichitaura nezvekudzidza kwakadzama, TensorFlow kazhinji ndiyo nheyo yekutanga kutaurwa. Inonyanya kufarirwa, chimiro ichi hachingoshandiswi neGoogle chete - iyo kambani inotarisira kugadzirwa kwayo - asiwo nemamwe makambani akaita seDropbox, eBay, Airbnb, Nvidia, uye mamwe akawanda.
TensorFlow inogona kushandiswa kugadzira yakakwira uye yakaderera level APIs, ichikubvumidza iwe kumhanya maapplication pane chero mhando yechigadzirwa. Kunyangwe Python uriwo mutauro wayo wekutanga, Tensoflow's interface inogona kuwanikwa nekudzorwa uchishandisa mimwe mitauro yekuronga seC ++, Java, Julia, uye JavaScript.
Kuve yakavhurika-sosi, TensorFlow inokutendera iwe kuti uite akati wandei kusanganisa nemamwe maAPI uye uwane rutsigiro nekukurumidza uye zvigadziriso kubva munharaunda. Kuvimba kwayo ne "static graph" yekombuta inobvumidza iwe kuti uite maverengero ekukurumidza kana kuchengetedza mashandiro ekuwana pane imwe nguva. Izvi zvikonzero, zvakawedzerwa pamukana wekuti iwe unogona "kutarisa" kuvandudzwa kweneural network yako kuburikidza neTensorBoard, ita TensorFlow iyo inonyanya kufarirwa chimiro chekudzidza kwakadzama.
Key Features
- Vhura-sosi
- Kubvuma sanduko
- Fast debugging
2. PyTorch
PyTorch chimiro chakagadzirwa neFacebook kutsigira kushanda kwemasevhisi ayo. Kubva zvave kuvhurika-sosi, chimiro ichi chave kushandiswa nemakambani asiri Facebook, akadai Salesforce uye Udacity.
Iyi dhizaini inoshanda zvine simba yakagadziridzwa magirafu, ichikutendera iwe kuti uchinje kune dhizaini yedataset yako paunenge uchiigadzirisa. NePyTorch zviri nyore kugadzira uye kudzidzisa neural network, kunyangwe pasina ruzivo mukudzidza kwakadzama.
Kuve yakavhurika-sosi uye yakavakirwa paPython, unogona kuita nyore uye nekukurumidza kusanganisa kuPyTorch. Iyo zvakare iri nyore chimiro chekudzidza, kushandisa, uye kugadzirisa. Kana uine mibvunzo, unogona kuvimba nerutsigiro rwakakura uye zvigadziriso kubva kunharaunda mbiri - Python nharaunda uye PyTorch nharaunda.
Key Features
- Nyore kudzidza
- Inotsigira GPU uye CPU
- Yakapfuma seti yemaAPIs ekuwedzera maraibhurari
3. Apache MX Net
Nekuda kwekukwirira kwayo, kuita kwepamusoro, kugadzirisa nekukurumidza, uye kutsigirwa kweGPU, iyi gadziriro yakagadzirwa neApache kuti ishandiswe mumapurojekiti makuru emaindasitiri.
MXNet inosanganisira iyo Gluon interface inobvumira vagadziri veese hunyanzvi mazinga kuti tanga nekudzidza kwakadzama pamusoro pegore, pamidziyo yemupendero, uye panharembozha. Mumitsetse mishoma yeGluon kodhi, unogona kuvaka mutsara kudzoreredza, convolutional network uye inodzokororwa maLSTM e. kuona kwechinhu, kucherechedzwa kwekutaura, kurudziro, uye hunhu.
MXNet inogona kushandiswa pamidziyo yakasiyana siyana uye inotsigirwa nevanoverengeka kuronga mitauro seJava, R, JavaScript, Scala uye Go. Kunyangwe nhamba yevashandisi nenhengo munharaunda yayo yakadzikira, MXNet ine magwaro akanyorwa zvakanaka uye mukana mukuru wekukura, kunyanya iko zvino iyo Amazon yasarudza hurongwa uhu sechishandiso chekutanga cheMuchina Kudzidza paAWS.
Key Features
- 8 mitauro inosungirwa
- Distributed Training, inotsigira akawanda-CPU uye akawanda-GPU masisitimu
- Hybrid kumberi-kumagumo, inobvumira kushandura pakati pekukosha uye ekufananidzira modhi
4. Microsoft Cognitive Toolkit
Kana iwe uri kufunga kugadzira maapplication kana masevhisi anomhanya paAzure (Microsoft Cloud services), iyo Microsoft Cognitive Toolkit ndiyo hurongwa hwekusarudza mapurojekiti ako ekudzidza akadzama. Iyi yakavhurika-sosi, uye inotsigirwa nemitauro yekuronga sePython, C ++, C #, Java, pakati pevamwe. Iyi dhizaini yakagadzirirwa "kufunga sehuropi hwemunhu", saka inogona kugadzirisa huwandu hwakawanda hwe data isina kurongeka, ichipa kukurumidza kudzidziswa uye intuitive dhizaini.
Nekusarudza iyi dhizaini - iyo yakafanana kuseri kweSkype, Xbox, uye Cortana - iwe unowana kuita kwakanaka kubva kune ako maapplication, scalability uye nyore kusanganisa neAzure. Nekudaro, kana ichienzaniswa neTensorFlow kana PyTorch, huwandu hwenhengo munharaunda yayo uye rutsigiro rwakaderedzwa.
Vhidhiyo inotevera inopa sumo yakazara uye mienzaniso yekushandisa:
Key Features
- Clear zvinyorwa
- Tsigiro kubva kuMicrosoft timu
- Yakananga girafu kuona
5. Keras
Kufanana nePyTorch, Keras iraibhurari yePython-yakavakirwa pamapurojekiti ane data. Iyo keras API inoshanda padanho repamusoro uye inobvumira kubatanidzwa neakaderera level APIs seTensorFlow, Theano, uye Microsoft Cognitive Toolkit.
Mamwe mabhenefiti ekushandisa kera kureruka kwayo kudzidza - kuve iyo inokurudzirwa chimiro chevanotanga mukudzidza kwakadzama; kumhanya kwayo kwekutumira; kuva nerutsigiro rwakakura kubva kunharaunda yepython uye kubva kunharaunda dzemamwe marongero ayo akabatanidzwa nawo.
Keras ine akasiyana mashandisirwo eiyo kuvaka zvidhinha zveneural network akadai sematanho, mabasa echinangwa, activation mabasa, uye masvomhu optimizers. Kodhi yayo inotambirwa paGitHub uye kune maforamu uye chiteshi chekutsigira cheSlack. Kuwedzera kutsigira mureza neural networks, Keras inopa rutsigiro rweConvolutional Neural Networks uye Recurrent Neural Networks.
Keras inobvumira mienzaniso yekudzidza yakadzama kugadzirwa pama smartphones pane ese iOS ne Android, paJava Virtual Machine, kana pawebhu. Inobvumirawo kushandiswa kwekudzidziswa kwakagoverwa kwemhando dzekudzidza dzakadzika pamasumbu eGraphics Processing Units (GPU) uye Tensor Processing Units (TPU).
Key Features
- Pre-kudzidziswa mienzaniso
- Multiple backend rutsigiro
- Mushandisi-ane hushamwari uye yakakura nharaunda rutsigiro
6. Apple Core ML
Core ML yakagadziridzwa neApple kutsigira ecosystem yayo - IOS, Mac OS, uye iPad OS. API yayo inoshanda pamwero wakaderera, ichishandisa zvakanaka zviwanikwa zveCPU uye zveGPU, izvo zvinobvumira modhi uye maapplication akagadzirwa kuti arambe achimhanya kunyangwe pasina internet yekubatanidza, izvo zvinoderedza "memory tsoka" uye mashandisiro emagetsi emudziyo.
Maitiro eCore ML anoita izvi haasi chaiwo nekugadzira imwe raibhurari yekudzidza yemuchina iyo yakagadziridzwa kuti ishande pamaiphone/ipads. Panzvimbo iyoyo, Core ML yakafanana nekombuta inotora maratidziro emhando uye akadzidziswa paramita anotsanangurwa neimwe muchina kudzidza software uye oishandura kuita faira rinova sosi yeApp app. Kushandurwa kweCore ML modhi kunoitika panguva yekuvandudzwa kweapp, kwete munguva chaiyo sezvo app iri kushandiswa, uye inofambiswa neraibhurari yecoremltools python.
Core ML inopa kukurumidza kuita nekubatanidzwa kuri nyore kwe machine learning modhi mumashandisirwo. Inotsigira kudzidza kwakadzama nemhando dzinopfuura makumi matatu dzematekinoroji pamwe nemiti yesarudzo, michina yekutsigira mavheti, uye mitsetse yekudzoreredza nzira, zvese zvakavakwa pamusoro peiyo yakaderera-level tekinoroji seSimbi uye Kumhanyisa.
Key Features
- Zviri nyore kubatanidza mumaapps
- Kushandiswa kwakakwana kwezviwanikwa zvemuno, zvisingade kuwana internet
- Kuvanzika: iyo data haifanirwe kusiya mudziyo
7. ONNX
Yekupedzisira framework pane yedu rondedzero ndeyeONNX. Iyi sisitimu yakabuda mukubatana pakati peMicrosoft neFacebook, nechinangwa chekurerutsa maitiro ekutamisa nekuvaka mamodheru pakati pematanho akasiyana, maturusi, nguva dzekumhanya uye ma compiler.
ONNX inotsanangura yakajairika faira mhando inogona kumhanya pamapuratifomu akawanda, uku uchishandisa mabhenefiti eakaderera-level APIs seaya anobva kuMicrosoft Cognitive Toolkit, MXNet, Caffe uye (kushandisa converters) Tensorflow uye Core ML. Musimboti uri kuseri kweONNX kudzidzisa modhi pane stack uye kuishandisa uchishandisa mamwe mafungidziro uye fungidziro.
Iyo LF AI Foundation, diki-sangano reLinux Foundation, isangano rakazvipira kuvaka ecosystem kutsigira. pachena-mabviro hunyanzvi muhungwaru hwekugadzira (AI), kudzidza muchina (ML), uye kudzidza kwakadzama (DL). Yakawedzera ONNX sepurojekiti yevakapedza kudzidza musi wa14 Mbudzi 2019. Kufamba uku kweONNX pasi peamburera yeLF AI Foundation kwakaonekwa sechiitiko chakakosha mukutanga ONNX semutengesi-asina kwaakarerekera akavhurika-fomati mwero.
Iyo ONNX Model Zoo muunganidzwa wemamodheru akafanodzidziswa muDeep Learning inowanikwa muONNX fomati. Kune imwe neimwe modhi iripo Jupyter zvinyorwa yekudzidzira modhi uye kuita inference neyakadzidziswa modhi. Mabhuku ekunyorera akanyorwa muPython uye ane zvinongedzo kune iyo kudzidzisa dataset uye zvirevo zvegwaro resainzi rekutanga rinotsanangura mavakirwo emuenzaniso.
Key Features
- Framework interoperability
- Hardware Optimization
mhedziso
Iyi ipfupiso yezvakanakisa zvimiro zve kudzidza zvakadzika. Kune akati wandei maitiro echinangwa ichi, emahara kana kubhadharwa. Kuti usarudze zvakanakira purojekiti yako, tanga waziva kuti ndeipi puratifomu yauchange uchigadzira yako application.
General masisitimu seTensorFlow neKeras ndiwo akanakisa sarudzo dzekutanga. Asi kana iwe uchida kushandisa OS kana mudziyo-chaiwo mabhenefiti, ipapo Core ML uye Microsoft Cognitive Toolkit inogona kunge iri yakanakisa sarudzo.
Kune mamwe marongero akanangana nemidziyo yeAroid, mimwe michina, uye zvinangwa zvisina kutaurwa mune iyi runyorwa. Kana boka rekupedzisira richikufarira, tinokurudzira kuti utsvage ruzivo rwavo paGoogle kana mamwe masaiti ekudzidza muchina.
Leave a Reply