儂迮悝炾乒犯伙及扔奈申件弘午反ˋ

Sep 25, 20211 like1,511 views

Sho Tanaka

儂迮悝炾乒犯伙及扔奈申件弘卞勾中化

赻撩畿賡
靡ゴㄩtsho
�棻丟犯奴失/�第丟犯奴失卞化犯奈正煦昴氪 /犯奈正巨件斥瓦失午仄
化伊戊丟件玉巨件斥件脹及嶱逃 /灍蚾/綅蚚毛童絞﹝
犯奈正卞憝歹日內它尼皮失皿伉弗奈扑亦件及嶱逃井日扔奈田奈扔奶
玉﹜奶件白仿�慷升盟𣷣仁𣶹岈﹝
掛逃桶反垀扽允月瞎�午反憝�S卅仁﹜
��及�賤午捼屯凶仇午毛逃桶中凶仄引允﹝

掛逃桶及正奈必永玄
- 巨件斥瓦失笢�氪卅升反豖И井手仄木引六氏﹝
場陑氪笢�氪奻�氪
犯奈正扔奶巨件氾奴旦玄 ? ? →
申斥生旦交奈扒奈 ? ? ?
巨件斥瓦失 ? → ?

@mshr-ito今氏及囀⺍毛統蕉卞仄化中引允﹝
@mshr-ito今氏
丐曰互午丹仍介中引允﹝
統蕉ㄩMLOps及衙猁午辻迮悝�乒犯伙及扔奈申件弘扑旦氾丞

扔奈申件弘午反ˋ
- 辻迮悝�乒犯伙及恅犢分午釬傖仄凶乒犯伙毛魂蚚允月凶戶卞
§枑鼎§允月仇午
- 瞰ㄩAPI匹瞳蚚匹五月袨颷卞允月﹜失皿伉弗奈扑亦件午仄化枑鼎匹五月袨颷卞允月﹜ etc.
- �躇卅ヽ癹反皺尹引六氏匹仄凶互�凶午仇欠
2017 Oreilly支 2018 Google i/o匹及桶�反復庲
- 奻�方曰嘉中及反引分復庲匹五內仍湔眭卅源中凶日諒尹化手日尹月午翑井曰引允﹝

辻迮悝�由奶皿仿奶件及湮引井卅�傖
MLOps 伊矛伙0: 忒�皿伕本旦
統蕉ㄩMLOps: 儂迮悝炾卞云仃月輥適腔犯伉田伉奈午赻雄趙及由奶皿仿奶件

乒犯伙及扔奈申件弘卞斛猁卅猁匼
- 岈ゴ卞悝�仄凶辻迮悝�乒犯伙
- 悝�𦤦心辻迮悝�乒犯伙毛瞳蚚允月凶戶及⻌薯犯奈正及杻釾講及ゴ�I燴
- 俋窒井日辻迮悝�乒犯伙毛瞳蚚允月凶戶及奶件正白尼奈旦
⻌薯犯奈正
�I燴
↗
Model
I/F

乒犯伙及扔奈申件弘互剠中�磁
扔奈申件弘及辻夔互卅仁﹜
API毛灍蚾仄卅中午中仃卅中
�磁反﹜乒犯伙釬傖眕俋卞戊奈犯奴件弘互斛猁
統蕉ㄩFastAPI毛妏勻化辻迮悝�乒犯伙毛api趙仄化心凶

扔奈申件弘毛枑鼎仄化中月白伊奈丞伐奈弁瞰
辻迮悝�仿奶皮仿伉匹扔奈申件弘毛枑鼎仄化中月瞰
- TensorFlow
- PyTorch etc.
�〝卅仿奶皮仿伉及乒犯伙毛�𡛟仄化中月扔奈申件弘白伊奈丞伐奈弁瞰
- BentoML
- NVIDIA Triton Inference Server etc.
統蕉ㄩMLOps及衙猁午辻迮悝�乒犯伙及扔奈申件弘扑旦氾丞

PyTorch毛妏勻凶瞰
統蕉ㄩpytorch/serve: Model Serving on PyTorch

��反鼠宒Doc
統蕉ㄩserve/README.md at master ﹞ pytorch/serve ﹞ GitHub

云歹曰
扔奈申件弘反﹜API卅升戊奈犯奴件弘仄化灍蚾仄卅仁化手辻迮悝�乒犯伙毛
扑旦氾丞卞��g卞瞎心煋戈仇午互匹五引允﹝
今引介引卅辻迮悝�乒犯伙及仿奶皮仿伉匹扔奈申件弘及辻夔毛枑鼎仄化中月及匹﹜
民奈丞匹瞳蚚仄化中月仿奶皮仿伉卅升郔羥卅手及毛腢亼仄化灍蚾允月仇午互褫夔匹允﹝

統蕉
https://qiita.com/mshr-ito/items/f456f60fc5c557a19cf5
https://ubuntu.com/blog/guide-to-ml-model-serving
https://www.oreilly.com/library/view/serving-machine-learning/9781492024095/
https://twitter.com/TensorFlow/status/994356340716920832
https://qiita.com/mshr-ito/items/f456f60fc5c557a19cf5
https://cloud.google.com/architecture/mlops-continuous-delivery-and-automation-pipelines-in-machine-learning?hl=ja
https://github.com/pytorch/serve/

Several recent papers have explored self-supervised learning methods for vision transformers (ViT). Key approaches include: 1. Masked prediction tasks that predict masked patches of the input image. 2. Contrastive learning using techniques like MoCo to learn representations by contrasting augmented views of the same image. 3. Self-distillation methods like DINO that distill a teacher ViT into a student ViT using different views of the same image. 4. Hybrid approaches that combine masked prediction with self-distillation, such as iBOT.

SSII2019OS: 旮𦵴悝�卞井井月媆嶲毛傻仁仄化心引六氏井ˋ ‵煦汃悝�及�戶‵SSII

▽DL�掂頗▼High-Resolution Image Synthesis with Latent Diffusion ModelsDeep Learning JP

You Only Look One-level Feature及賤掊午�六井仃凶昜极𨈘堤及方手支引�Yusuke Uchida

乒犯伙詢厒趙啃恁Yusuke Uchida

賒砉本件扑件弘扑件禾斥它丞 (SSII 2019) 及き賒本永扑亦件☆旮𦵴悝�及詢厒趙 ? 詢厒民永皿﹜煦汃悝�﹜幏講乒犯伙 ??★及琌栳揃蹋匹允﹝ 旮𦵴悝�乒犯伙毛詢厒趙允月狟�6意�及忒楊及賤掊匹允﹝ - �心煋心及煦賤 (Factorization) - 皉尌曰 (Pruning) - 失奈平氾弁民乓抻坰 (Neural Architecture Search; NAS) - 婌ヽ皺賸﹜�腔�呾弘仿白?(Early Termination, Dynamic Computation Graph) - 淛隱 (Distillation) - 講赽趙 (Quantization)

Swin Transformer (ICCV'21 Best Paper) 毛俇韏卞燴賤允月揃蹋Yusuke Uchida

MLOps 卞價勿仁 AI/ML 灍綅蚚郔ゴ� ‵賒砉﹜�賒犯奈正卞云仃月 MLOps 岈瞰及仍畿賡‵ㄗ茬砉ロ�丟犯奴失悝頗2021爛隄撫湮頗き賒本永扑亦...NTT DATA Technology & Innovation

昜极𨈘堤白伊奈丞伐奈弁紼紼嗨梗喧梗釵喧勳棗紳匹辦巠卅羲逃Tatsuya Suzuki

MixMatch: A Holistic Approach to Semi- Supervised Learningharmonylab

▽丟正扔奈矛奶▼杅宒玉伉皮件諒呇丐曰悝炾cvpaper. challenge

▽DL�掂頗▼晟惤眕俋匹及Transformer及引午戶 (ViT, Perceiver, Frozen Pretrained Transformer etc)Deep Learning JP

This document summarizes recent research on applying self-attention mechanisms from Transformers to domains other than language, such as computer vision. It discusses models that use self-attention for images, including ViT, DeiT, and T2T, which apply Transformers to divided image patches. It also covers more general attention modules like the Perceiver that aims to be domain-agnostic. Finally, it discusses work on transferring pretrained language Transformers to other modalities through frozen weights, showing they can function as universal computation engines.

BERT+XLNet+RoBERTa�銜刓ぃ

BERT 毛笢陑卞賤掊仄凶揃蹋匹允ㄝBERT 卞掀屯月午ㄛXLNet 午 RoBERTa 及囀⺍反��卞袚勻化卅中匹允ㄝ丐午ㄛ赻釬及�反奻井日狟匹允互ㄛ竘勻�勻化五凶�反狟井日奻卞卅勻化中月及匹蛁砩仄化仁分今中ㄝ手仄嶲綃中脹丐勻凶日党淏允月及匹ㄛ晟勻化仁分今中ㄝㄗ杻卞ㄛRoBERTa 及荎惤毛掂心嶲綃尹互切斤勻午窕中匹允ㄝ晟中偈允中引六氏ㄝㄘ BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding XLNet: Generalized Autoregressive Pretraining for Language Understanding RoBERTa: A Robustly Optimized BERT Pretraining Approach

紼郭款梭棗滄匹悝少紼郭倏梯莽仇午反元戶Kenichi Sonoda

SSII2022 [SS1] 瓦亙奈仿伙3D桶政及郔陔�砃? 瓦亙奈仿伙生永玄匹卅氏匹手桶六月ˋˋ ??SSII

6/9 (躂) 14:15‵14:45 丟奶件頗� 琌�ㄩ樓枘湮ル庌ㄗ絁宒頗扦Preferred Networksㄘ? 衙猁ㄩ萸�/丟永扑亙/示弁本伙卅升及𣶹懂腔卅3D桶政卞測歹勻化瓦亙奈仿伙生永玄卞方月�憝杅匹ʊ棒啋犯奈正毛桶仄化仄引丹忒楊互輪爛腎�仄﹜�〝卅正旦弁匹醴�引仄中傖彆毛伃仆湮五卅蛁醴毛摩戶化中月﹝仇及桶政反﹜�呾戊旦玄互詢中測歹曰卞⺥�俶互詢仁今引介引卅手及毛扑件皿伙井勾戊件由弁玄卞桶政匹五月仇午互眭日木化中月﹝掛琌栳匹反﹜公及價渙﹜𡛟蚚﹜郔珂傷及袨暿卞勾中化﹜測桶腔卅旃噶毛𣷣仁畿賡允月﹝

輪爛及Hierarchical Vision TransformerYusuke Uchida

[DL�掂頗]※Spatial Attention Point Network for Deep-learning-based Robust Autono...Deep Learning JP

儂迮悝炾匹ざ井卅中凶戶及戊奈玉扢數Takahiro Kubo

[DL�掂頗]data2vec: A General Framework for Self-supervised Learning in Speech,...Deep Learning JP

Transformer 丟正扔奈矛奶cvpaper. challenge

Spark MLlib匹反元戶月旦弗奈仿皮伙卅辻迮悝�NTT DATA OSS Professional Services

▽DL�掂頗▼SimCSE: Simple Contrastive Learning of Sentence Embeddings (EMNLP 2021)Deep Learning JP

市伙穴件白奴伙正⻌藷Yasunori Nihei

⻌嬡 Kubeflow ‵Kubernetes匹辻迮悝�毛反元戶月凶戶卞‵ ㄗNTT Tech Conference #4 琌栳揃蹋ㄘNTT DATA Technology & Innovation

[DL�掂頗]Understanding Black-box Predictions via Influence Functions Deep Learning JP

[DL�掂頗]Model soups: averaging weights of multiple fine-tuned models improves ...Deep Learning JP

囮啖井日悝少儂迮悝炾𡛟蚚Hiroyuki Masuda

扦囀辭�頗匹及逃桶揃蹋匹允﹝ ☆囮㷍岈瞰毛籵元化﹜辻迮悝�及𨈘�匹眚尹月屯五禾奶件玄毛悝少★毛戊件本皿玄卞釬傖仄引仄凶﹝AI?辻迮悝�毛𨈘�允月𣷣仁及源〝卞魂蚚仄化中凶分仃月午倷中匹允﹝ 丐午互五毛狟�卞𤩸五引仄凶﹝方仃木壬仇切日手仍統桽仁分今中﹝ https://qiita.com/bezilla/items/1e1abac767e10d0817d1

FastAPI毛妏勻化辻迮悝�乒氾?伙毛api趙仄化心凶Sho Tanaka

紼郭倏梯莽及衙猁午場悝氪互𠸎毛勾仃凶幻丹互謎中仇午Sho Tanaka

More Related Content

What's hot (20)

昜极𨈘堤白伊奈丞伐奈弁紼紼嗨梗喧梗釵喧勳棗紳匹辦巠卅羲逃Tatsuya Suzuki

MixMatch: A Holistic Approach to Semi- Supervised Learningharmonylab

▽丟正扔奈矛奶▼杅宒玉伉皮件諒呇丐曰悝炾cvpaper. challenge

▽DL�掂頗▼晟惤眕俋匹及Transformer及引午戶 (ViT, Perceiver, Frozen Pretrained Transformer etc)Deep Learning JP

BERT+XLNet+RoBERTa�銜刓ぃ

紼郭款梭棗滄匹悝少紼郭倏梯莽仇午反元戶Kenichi Sonoda

SSII2022 [SS1] 瓦亙奈仿伙3D桶政及郔陔�砃? 瓦亙奈仿伙生永玄匹卅氏匹手桶六月ˋˋ ??SSII

輪爛及Hierarchical Vision TransformerYusuke Uchida

[DL�掂頗]※Spatial Attention Point Network for Deep-learning-based Robust Autono...Deep Learning JP

儂迮悝炾匹ざ井卅中凶戶及戊奈玉扢數Takahiro Kubo

[DL�掂頗]data2vec: A General Framework for Self-supervised Learning in Speech,...Deep Learning JP

Transformer 丟正扔奈矛奶cvpaper. challenge

Spark MLlib匹反元戶月旦弗奈仿皮伙卅辻迮悝�NTT DATA OSS Professional Services

▽DL�掂頗▼SimCSE: Simple Contrastive Learning of Sentence Embeddings (EMNLP 2021)Deep Learning JP

市伙穴件白奴伙正⻌藷Yasunori Nihei

⻌嬡 Kubeflow ‵Kubernetes匹辻迮悝�毛反元戶月凶戶卞‵ ㄗNTT Tech Conference #4 琌栳揃蹋ㄘNTT DATA Technology & Innovation

[DL�掂頗]Understanding Black-box Predictions via Influence Functions Deep Learning JP

[DL�掂頗]Model soups: averaging weights of multiple fine-tuned models improves ...Deep Learning JP

囮啖井日悝少儂迮悝炾𡛟蚚Hiroyuki Masuda

昜极𨈘堤白伊奈丞伐奈弁紼紼嗨梗喧梗釵喧勳棗紳匹辦巠卅羲逃Tatsuya Suzuki

MixMatch: A Holistic Approach to Semi- Supervised Learningharmonylab

▽丟正扔奈矛奶▼杅宒玉伉皮件諒呇丐曰悝炾cvpaper. challenge

▽DL�掂頗▼晟惤眕俋匹及Transformer及引午戶 (ViT, Perceiver, Frozen Pretrained Transformer etc)Deep Learning JP

BERT+XLNet+RoBERTa�銜刓ぃ

紼郭款梭棗滄匹悝少紼郭倏梯莽仇午反元戶Kenichi Sonoda

SSII2022 [SS1] 瓦亙奈仿伙3D桶政及郔陔�砃? 瓦亙奈仿伙生永玄匹卅氏匹手桶六月ˋˋ ??SSII

輪爛及Hierarchical Vision TransformerYusuke Uchida

[DL�掂頗]※Spatial Attention Point Network for Deep-learning-based Robust Autono...Deep Learning JP

儂迮悝炾匹ざ井卅中凶戶及戊奈玉扢數Takahiro Kubo

[DL�掂頗]data2vec: A General Framework for Self-supervised Learning in Speech,...Deep Learning JP

Transformer 丟正扔奈矛奶cvpaper. challenge

Spark MLlib匹反元戶月旦弗奈仿皮伙卅辻迮悝�NTT DATA OSS Professional Services

▽DL�掂頗▼SimCSE: Simple Contrastive Learning of Sentence Embeddings (EMNLP 2021)Deep Learning JP

市伙穴件白奴伙正⻌藷Yasunori Nihei

⻌嬡 Kubeflow ‵Kubernetes匹辻迮悝�毛反元戶月凶戶卞‵ ㄗNTT Tech Conference #4 琌栳揃蹋ㄘNTT DATA Technology & Innovation

[DL�掂頗]Understanding Black-box Predictions via Influence Functions Deep Learning JP

[DL�掂頗]Model soups: averaging weights of multiple fine-tuned models improves ...Deep Learning JP

囮啖井日悝少儂迮悝炾𡛟蚚Hiroyuki Masuda

Similar to 儂迮悝炾乒犯伙及扔奈申件弘午反ˋ (20)

FastAPI毛妏勻化辻迮悝�乒氾?伙毛api趙仄化心凶Sho Tanaka

紼郭倏梯莽及衙猁午場悝氪互𠸎毛勾仃凶幻丹互謎中仇午Sho Tanaka

儂迮悝炾𡛟蚚及凶戶及末白玄它尼失巨件斥瓦失伉件弘由正奈件HironoriTAKEUCHI1

Agile and Modeling in embedded systems safety and securityKenji Hiranabe

儂迮悝炾及薯毛竘五堤允凶戶及甡湔俶奪燴Takahiro Kubo

巨件正奈皿仿奶朮午儂迮悝炾撮扲maruyama097

Machine Learning Service毛妏勻化心方丹Yoshitaka Seo

辻迮悝�𡛟蚚扑旦氾丞及失奈平氾弁民乓?犯扒奶由正奈件(2020-07 玉仿白玄田奈斥亦件))HironoriTAKEUCHI1

陔仄中末白玄它尼失巨件斥瓦失伉件弘及凶戶及由正奈件仿件必奈斥卞砃仃化Hironori Washizaki

價插栳炾博喳20141021膽𣷣碩珧

儂迮悝炾𡛟蚚失奈平氾弁民乓?犯扒奶件由正奈件衙紝Hironori Washizaki

丟伙矢奶及迵陓乒犯伉件弘匹假�?假陑及凶戶卞灍犛仄化中月仇午Yuhi Kawakami

價插栳炾博喳20151006膽𣷣碩珧

Workflow pipeline patternyohei okawa

2021-03-05.osc online-spring(connect-cms)絁宒頗扦左奈皿件末奈旦?伐奈弁扑亦永皿

辻迮悝�及穴奶弁伕扔奈甲?旦化?及綅蚚及灍歠卞勾中化 #mlopsHiroaki Kudo

儂迮悝炾犯扒奶件由正奈件及斛猁俶午儂迮悝炾仿奶白扔奶弁伙Hironori Washizaki

釦梗莽莽勳棗紳4:☆珂輛申永弘犯奈正𡛟蚚毛盓尹月儂迮悝炾卞А戶日木月陔撮扲★ㄞ掀㛡蔚ⅸPreferred Networks

PFI左奈皿件本立瓜奈2012☆嗣�趙允月ロ�毛盓尹月撮胍★ 2012爛9堎21゜ㄗ踢ㄘ灍囥 ?衙猁?Hadoop互湮�耀犯奈正�I燴卞𣷣仁蚚中日木木月珨源﹜公及癹賜手�尹宎戶化五凶﹝珨源﹜犯奈正卞サ戈恚螝卅秪彆憝�S支�A砃毛逃�仄儕僅及謎中軑䛐毛灍政允月辻迮悝�撮胍反俶夔砃奻午羥蚚��及�湮毛適仃化中月﹝掛琌栳匹反申永弘犯奈正午公及珂筳失皿伉弗奈扑亦件卞勾中化﹜嶲毛�什辻迮悝�撮胍及紝萸井日郔陔�砃卞勾中化扴屯月﹝杻卞﹜PFI互白巧奈市旦仄化中月伉失伙正奶丞俶午玄伊奈扔申伉氾奴卞勾中化�仄仁扴屯﹜Jubatus午Bazil午中丹庨ⅲ毛畿賡允月﹝

20190807_Aidemy Azure AI 仇?畿賡Deep Learning Labㄗ犯奴奈皿仿奈瓦件弘?仿示ㄘ

儂迮悝炾馱悝午儂迮悝炾𡛟蚚扑旦氾丞及羲逃高釦鳥硃娶喧釦楚本立瓜奈(2021/3/30)Nobukazu Yoshioka

FastAPI毛妏勻化辻迮悝�乒氾?伙毛api趙仄化心凶Sho Tanaka

紼郭倏梯莽及衙猁午場悝氪互𠸎毛勾仃凶幻丹互謎中仇午Sho Tanaka

儂迮悝炾𡛟蚚及凶戶及末白玄它尼失巨件斥瓦失伉件弘由正奈件HironoriTAKEUCHI1

Agile and Modeling in embedded systems safety and securityKenji Hiranabe

儂迮悝炾及薯毛竘五堤允凶戶及甡湔俶奪燴Takahiro Kubo

巨件正奈皿仿奶朮午儂迮悝炾撮扲maruyama097

Machine Learning Service毛妏勻化心方丹Yoshitaka Seo

辻迮悝�𡛟蚚扑旦氾丞及失奈平氾弁民乓?犯扒奶由正奈件(2020-07 玉仿白玄田奈斥亦件))HironoriTAKEUCHI1

陔仄中末白玄它尼失巨件斥瓦失伉件弘及凶戶及由正奈件仿件必奈斥卞砃仃化Hironori Washizaki

價插栳炾博喳20141021膽𣷣碩珧

儂迮悝炾𡛟蚚失奈平氾弁民乓?犯扒奶件由正奈件衙紝Hironori Washizaki

丟伙矢奶及迵陓乒犯伉件弘匹假�?假陑及凶戶卞灍犛仄化中月仇午Yuhi Kawakami

價插栳炾博喳20151006膽𣷣碩珧

Workflow pipeline patternyohei okawa

2021-03-05.osc online-spring(connect-cms)絁宒頗扦左奈皿件末奈旦?伐奈弁扑亦永皿

辻迮悝�及穴奶弁伕扔奈甲?旦化?及綅蚚及灍歠卞勾中化 #mlopsHiroaki Kudo

儂迮悝炾犯扒奶件由正奈件及斛猁俶午儂迮悝炾仿奶白扔奶弁伙Hironori Washizaki

釦梗莽莽勳棗紳4:☆珂輛申永弘犯奈正𡛟蚚毛盓尹月儂迮悝炾卞А戶日木月陔撮扲★ㄞ掀㛡蔚ⅸPreferred Networks

20190807_Aidemy Azure AI 仇?畿賡Deep Learning Labㄗ犯奴奈皿仿奈瓦件弘?仿示ㄘ

儂迮悝炾馱悝午儂迮悝炾𡛟蚚扑旦氾丞及羲逃高釦鳥硃娶喧釦楚本立瓜奈(2021/3/30)Nobukazu Yoshioka