ʻO kahi hoʻolālā no ke aʻo hohonu he hui pū ʻana o nā pilina, nā hale waihona puke a me nā mea paahana e wehewehe a hoʻomaʻamaʻa i nā kumu hoʻohālike Machine Learning me ka wikiwiki a pololei.
Ma muli o ka hoʻohana ʻana o ka ʻike hohonu i ka nui o ka ʻikepili i kūkulu ʻole ʻia, ʻaʻole kikokikona, pono ʻoe i kahi hoʻolālā e hoʻomalu i ka pilina ma waena o nā "papa" a hana wikiwiki i ka hoʻomohala ʻana ma o ke aʻo ʻana mai ka ʻikepili hoʻokomo a me ka hoʻoholo kūʻokoʻa.
Inā makemake ʻoe e aʻo e pili ana i ke aʻo hohonu ma 2021, e noʻonoʻo e hoʻohana i kekahi o nā frameworks i hōʻike ʻia ma lalo nei. E hoʻomanaʻo e koho i kahi e kōkua iā ʻoe e hoʻokō i kāu mau pahuhopu a me kāu ʻike.
1. Nānā
Ke kamaʻilio e pili ana i ke aʻo hohonu, Nānā ʻO ka pinepine ka papa hana mua i ʻōlelo ʻia. ʻO ka mea kaulana loa, ʻaʻole hoʻohana wale ʻia kēia framework e Google - ʻo ka hui ke kuleana o kāna hana ʻana - akā na nā hui ʻē aʻe e like me Dropbox, eBay, Airbnb, Nvidia, a me nā mea ʻē aʻe he nui.
Hiki ke hoʻohana ʻia ʻo TensorFlow e hoʻomohala i nā API kiʻekiʻe a haʻahaʻa, e ʻae iā ʻoe e holo i nā noi ma kahi o nā ʻano mea like ʻole. ʻOiai ʻo Python kāna ʻōlelo mua, hiki ke kiʻi ʻia a hoʻokele ʻia ka interface o Tensoflow me ka hoʻohana ʻana i nā ʻōlelo papahana ʻē aʻe e like me C++, Java, Julia, a me JavaScript.
No ka wehe ʻana, ʻae ʻo TensorFlow iā ʻoe e hana i kekahi mau hoʻohui me nā API ʻē aʻe a loaʻa ke kākoʻo wikiwiki a me nā mea hou mai ke kaiāulu. ʻO kona hilinaʻi ʻana i nā "graphs static" no ka helu ʻana e hiki ai iā ʻoe ke hana koke i ka helu ʻana a i ʻole mālama i nā hana no ke komo ʻana i kahi manawa ʻē aʻe. ʻO kēia mau kumu, i hoʻohui ʻia i ka hiki iā ʻoe ke "nānā" i ka hoʻomohala ʻana o kāu pūnaewele neural ma o TensorBoard, e hoʻolilo iā TensorFlow i ka papa hana kaulana loa no ke aʻo hohonu.
Key Features
- Kauhuka-kahi kumu
- kohoʻia
- Hoʻopau wikiwiki
2. ʻO PyTorch
ʻO PyTorch kahi hana i kūkulu ʻia e Facebook e kākoʻo i ka hana o kāna mau lawelawe. Mai ka lilo ʻana i open-source, ua hoʻohana ʻia kēia framework e nā hui ʻē aʻe ma Facebook, e like me Salesforce a me Udacity.
Ke hana nei kēia anga i nā kiʻikuhi i hoʻonui hou ʻia, e ʻae iā ʻoe e hoʻololi i ka hoʻolālā ʻana o kāu waihona i kāu hana ʻana. Me PyTorch ʻoi aku ka maʻalahi o ka hoʻomohala ʻana a hoʻomaʻamaʻa i kahi pūnaewele neural, ʻoiai me ka ʻole o ka ʻike i ke aʻo hohonu.
No ka wehe ʻana a hoʻokumu ʻia ma Python, hiki iā ʻoe ke hana maʻalahi a wikiwiki hoʻi i ka PyTorch. He ʻano hana maʻalahi nō hoʻi ia e aʻo ai, hoʻohana, a me ka debug. Inā he mau nīnau kāu, hiki iā ʻoe ke hilinaʻi i ke kākoʻo nui a me nā mea hou mai nā kaiāulu ʻelua - ke kaiāulu Python a me ke kaiāulu PyTorch.
Key Features
- Like ke aʻo
- Kākoʻo GPU a me CPU
- Nui nā API e hoʻonui i nā hale waihona puke
3. ʻO Apache MX Pūnaewele
Ma muli o kona scalability kiʻekiʻe, hana kiʻekiʻe, hoʻoponopono wikiwiki, a me ke kākoʻo GPU kiʻekiʻe, ua hana ʻia kēia ʻano e Apache no ka hoʻohana ʻana i nā papahana ʻoihana nui.
Aia ka MXNet i ka pilina Gluon e hiki ai i nā mea hoʻomohala o nā pae akamai āpau e hoʻomaka me ke aʻo hohonu ma ke ao, ma nā polokalamu lihi, a ma nā polokalamu kelepona. I loko o nā laina liʻiliʻi o Gluon code, hiki iā ʻoe ke kūkulu i ka regression linear, convolutional networks a me nā LSTM recurrent for ʻike mea, ʻike ʻōlelo, ʻōlelo paipai, a me ka pilikino.
Hiki ke hoʻohana ʻia ʻo MXNet ma nā polokalamu like ʻole a kākoʻo ʻia e kekahi nā ʻōlelo hoʻonohonoho e like me Java, R, JavaScript, Scala a me Go. ʻOiai he haʻahaʻa ka helu o nā mea hoʻohana a me nā lālā o kona kaiāulu, ua kākau maikaʻi ʻia ʻo MXNet i nā palapala a me ka mana nui no ka ulu ʻana, ʻoiai i kēia manawa ua koho ʻo Amazon i kēia framework ma ke ʻano he mea hana mua no ka Machine Learning ma AWS.
Key Features
- 8 paa olelo
- Hāʻawi ʻia ka hoʻomaʻamaʻa ʻana, kākoʻo i nā ʻōnaehana multi-CPU a me multi-GPU
- Hybrid front-end, hiki ke hoʻololi i waena o nā ʻano imperative a me nā ʻano hōʻailona
4. ʻO Microsoft Cognitive Toolkit
Inā ʻoe e noʻonoʻo ana i ka hoʻomohala ʻana i nā noi a i ʻole nā lawelawe e holo ana ma Azure (Microsoft cloud services), ʻo Microsoft Cognitive Toolkit ka hoʻolālā e koho ai no kāu mau papahana aʻo hohonu. He kumu wehe kēia, a kākoʻo ʻia e nā ʻōlelo papahana e like me Python, C++, C#, Java, a me nā mea ʻē aʻe. Hoʻolālā ʻia kēia ʻano hana e "noʻonoʻo e like me ka lolo o ke kanaka", no laila hiki iā ia ke hana i ka nui o nā ʻikepili i hoʻonohonoho ʻole ʻia, ʻoiai e hāʻawi ana i ka hoʻomaʻamaʻa wikiwiki a me ka hoʻolālā intuitive.
Ma ke koho ʻana i kēia hoʻolālā - ka mea like ma hope o Skype, Xbox, a me Cortana - e loaʻa iā ʻoe ka hana maikaʻi mai kāu mau noi, scalability a me ka hoʻohui maʻalahi me Azure. Eia naʻe, ke hoʻohālikelike ʻia me TensorFlow a i ʻole PyTorch, ua hoʻemi ʻia ka helu o nā lālā o kona kaiāulu a me ke kākoʻo.
Hāʻawi kēia wikiō i kahi hoʻolauna piha a me nā hiʻohiʻona noi.
Key Features
- Holoi i nā palapala
- Kākoʻo mai ka hui Microsoft
- Nānā kiʻi pololei
5. Keras
E like me PyTorch, ʻo Keras kahi waihona waihona Python no nā papahana ʻikepili. Hana ʻia ka API paʻa ma kahi pae kiʻekiʻe a ʻae i ka hoʻohui ʻana me nā API haʻahaʻa haʻahaʻa e like me TensorFlow, Theano, a me Microsoft Cognitive Toolkit.
ʻO kekahi mau mea maikaʻi o ka hoʻohana ʻana i ka paʻakikī, ʻo ia ka maʻalahi o ke aʻo ʻana - ʻo ia ka ʻōnaehana i manaʻo ʻia no ka poʻe hoʻomaka i ke aʻo hohonu; kona wikiwiki o ka waiho ʻana; loaʻa ke kākoʻo nui mai ke kaiāulu python a mai nā kaiāulu o nā ʻano hana ʻē aʻe i hoʻohui ʻia.
Aia nā Keras i nā hoʻokō like ʻole o ka nā poloka kūkulu o nā pūnaewele neural e like me nā papa, nā hana pahuhopu, nā hana hoʻāla, a me nā mea hoʻoponopono makemakika. Hoʻokipa ʻia kāna code ma GitHub a aia nā ʻaha kūkā a me kahi ala kākoʻo Slack. Ma waho aʻe o ke kākoʻo no ka maʻamau nā hanana laulā, Hāʻawi ʻo Keras i ke kākoʻo no Convolutional Neural Networks a me Recurrent Neural Networks.
ʻAe ʻo Keras nā kumu hoʻohālike hohonu e hana ʻia ma nā smartphones ma IOS a me Android, ma ka Java Virtual Machine, a i ʻole ma ka pūnaewele. ʻAe ia i ka hoʻohana ʻana i ka hoʻomaʻamaʻa māhele ʻia o nā kumu hoʻohālike hohonu ma nā pūʻulu o Graphics Processing Units (GPU) a me Tensor Processing Units (TPU).
Key Features
- Nā hiʻohiʻona i hoʻomaʻamaʻa mua ʻia
- Kākoʻo backend lehulehu
- Ke kākoʻo kaiaulu nui a me ka mea hoʻohana
6. Apple Core ML
Ua hoʻomohala ʻia ʻo Core ML e Apple e kākoʻo i kāna kaiaolaola - IOS, Mac OS, a me iPad OS. Hana ʻia kāna API ma kahi haʻahaʻa, e hoʻohana maikaʻi ana i nā kumuwaiwai o CPU a me GPU, e hiki ai i nā hiʻohiʻona a me nā noi i hana ʻia e hoʻomau i ka holo ʻana me ka ʻole o kahi pilina pūnaewele, e hōʻemi ana i ka "wāwae hoʻomanaʻo" a me ka hoʻohana mana o ka hāmeʻa.
ʻO ke ala e hoʻokō ai ʻo Core ML i kēia ʻaʻole pololei ma ka hana ʻana i kahi waihona aʻo mīkini ʻē aʻe i hoʻopaʻa ʻia no ka holo ʻana ma nā iphones/ipads. Akā, ʻoi aku ka like o Core ML me kahi mea hoʻopili e lawe i nā kikoʻī hoʻohālike a me nā ʻāpana hoʻomaʻamaʻa i hōʻike ʻia me nā polokalamu aʻo mīkini ʻē aʻe a hoʻololi iā ia i faila i lilo i kumu no kahi polokalamu iOS. Hana ʻia kēia hoʻololi ʻana i kahi kumu hoʻohālike Core ML i ka wā o ka hoʻomohala ʻana i ka app, ʻaʻole i ka manawa maoli i ka hoʻohana ʻia ʻana o ka app, a ua maʻalahi ʻia e ka coremltools python library.
Hāʻawi ʻo Core ML i ka hana wikiwiki me ka hoʻohui maʻalahi o aʻo aʻo nā hiʻohiʻona i nā noi. Kākoʻo ia i ke aʻo hohonu me ka ʻoi aku o 30 mau ʻano papa a me nā kumu lāʻau hoʻoholo, kākoʻo i nā mīkini vector, a me nā ʻano laina regression, i kūkulu ʻia ma luna o nā ʻenehana haʻahaʻa e like me Metal a me Accelerate.
Key Features
- Maʻalahi e hoʻohui i nā polokalamu
- Hoʻohana maikaʻi loa i nā kumuwaiwai kūloko, ʻaʻole pono ke komo pūnaewele
- Palekana: ʻaʻole pono e haʻalele ka ʻikepili i ka hāmeʻa
7. ONNX
ʻO ka papa hana hope loa ma kā mākou papa inoa ʻo ONNX. Ua puka mai kēia hoʻolālā mai kahi hui like ʻana ma waena o Microsoft a me Facebook, me ka pahuhopu e hoʻomaʻamaʻa i ke kaʻina hana o ka hoʻoili ʻana a me ke kūkulu ʻana i nā kumu hoʻohālike ma waena o nā ʻano hana like ʻole, nā mea hana, nā manawa holo a me nā mea hoʻohui.
Hoʻomaopopo ʻo ONNX i kahi ʻano faila maʻamau i hiki ke holo ma nā paepae he nui, ʻoiai e hoʻohana ana i nā pono o nā API haʻahaʻa haʻahaʻa e like me nā mea mai Microsoft Cognitive Toolkit, MXNet, Caffe a me (hoʻohana ʻana i nā mea hoʻololi) Tensorflow a me Core ML. ʻO ke kumu ma hope o ONNX ʻo ia ke aʻo ʻana i kahi kumu hoʻohālike ma kahi ʻōpala a hoʻokō me ka hoʻohana ʻana i nā inferences a me nā wānana ʻē aʻe.
ʻO ka LF AI Foundation, he sub-organization o ka Linux Foundation, he hui i hoʻolaʻa ʻia e kūkulu i kahi kaiaola e kākoʻo. Wehewehe hana hou i ka naʻauao hana (AI), aʻo mīkini (ML), a me ke aʻo hohonu (DL). Ua hoʻohui ʻia ʻo ONNX ma ke ʻano he papahana puka puka ma 14 Nowemapa 2019. ʻO kēia neʻe ʻana o ONNX ma lalo o ka malu o ka LF AI Foundation i ʻike ʻia he mea nui i ka hoʻokumu ʻana iā ONNX ma ke ʻano he mea kūʻai aku-neutral open-format standard.
ʻO ka ONNX Model Zoo kahi hōʻiliʻili o nā hiʻohiʻona i hoʻomaʻamaʻa mua ʻia ma Deep Learning i loaʻa ma ka ʻano ONNX. No kēlā me kēia hiʻohiʻona aia Nā puke puke Jupyter no ka hoʻomaʻamaʻa kumu hoʻohālike a me ka hana ʻana i ka inference me ke kumu hoʻohālike i aʻo ʻia. Ua kākau ʻia nā puke puke ma Python a loaʻa nā loulou i ka ʻikepili hoʻomaʻamaʻa a me nā kuhikuhi ʻana i ka palapala ʻepekema kumu e wehewehe ana i ke ʻano hoʻohālike.
Key Features
- Pākuʻi hana
- Hoʻoponopono Paʻa Paʻa
Panina
ʻO kēia kahi hōʻuluʻulu o nā frameworks maikaʻi loa no haʻawina hohonu. Nui nā papa hana no kēia kumu, manuahi a uku ʻia paha. No ke koho ʻana i ka mea maikaʻi loa no kāu pāhana, e ʻike mua i kahi kahua āu e hoʻomohala ai i kāu noi.
ʻO nā papa hana maʻamau e like me TensorFlow a me Keras nā koho maikaʻi loa e hoʻomaka. Akā inā pono ʻoe e hoʻohana i ka OS a i ʻole nā pono kikoʻī kikoʻī, a laila ʻo Core ML a me Microsoft Cognitive Toolkit paha nā koho maikaʻi loa.
Aia kekahi mau papa hana e pili ana i nā polokalamu Android, nā mīkini ʻē aʻe, a me nā kumu kikoʻī i ʻōlelo ʻole ʻia ma kēia papa inoa. Inā makemake ka hui hope iā ʻoe, manaʻo mākou e ʻimi i kā lākou ʻike ma Google a i ʻole nā pūnaewele aʻo mīkini ʻē aʻe.
Waiho i ka Reply