Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
“머신러닝 엔지니어가 다녀온 F8” 김준성 - F8 2019 Meetup:Seoul
Search
Sponsored
·
Ship Features Fearlessly
Turn features on and off without deploys. Used by thousands of Ruby developers.
→
Junseong
May 15, 2019
Technology
240
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
“머신러닝 엔지니어가 다녀온 F8” 김준성 - F8 2019 Meetup:Seoul
Junseong
May 15, 2019
More Decks by Junseong
See All by Junseong
House Share Platfrom with Senior Citizen 👵🏻🏠 - Junction Asia 2023
codertimo
0
110
Haru - Diary App with Stable Diffusion (JUNCTION 2022)
codertimo
0
1.9k
Imaginar.AI - JUNCTION ASIA 2022 Hackathon
codertimo
0
3.4k
Python Memory Management
codertimo
0
710
[파이콘 2019] 100억건의 카카오톡 데이터로 똑똑한 일상대화 인공지능 만들기
codertimo
7
4.2k
All you need is Google's AI Tech: I/O Extended Seoul 2019
codertimo
1
500
[네이버 테크톡] 사람들과 자연스러운 대화를 나누는 인공지능 '핑퐁' 만들기
codertimo
0
510
Other Decks in Technology
See All in Technology
生成AI×AWS CDK×AWS FISで"振り返れる"ミニGameDayをつくろう
yoshimi0227
2
530
CDKで書くECSのベストプラクティス、 改めて考え直す2026 #cdkconf2026
makies
3
940
凡エンジニアがこの先生きのこるためには。〜TypeScript完全に理解したい〜
alchemy1115
2
420
Type-safe IaC for Dart
coborinai
0
180
Aurora MySQL 8.4リリース! Rubyistが備えること / what-rubyist-should-prepare-for-aurora-mysql-8-4
fkmy
0
780
どこまでAIに任せるか 〜確率論と決定論の境界決定〜
shukob
0
460
Amplify Gen2でbackend.tsにCDKを定義する/しない事によるCDKの挙動の違いとユースケース
smt7174
1
490
生成 AI 時代にいま一度「問い合わせ」について考えてみる
kazzpapa3
1
130
2年前に削除したPHPクラスが、 ある日突然決済をエラーにした
ykagano
1
790
AI x 開発生産性を取り巻く予算戦略と投資対効果
i35_267
7
2.9k
発表と総括 / Presentations and Summary
ks91
PRO
0
190
「休む」重要さ
smt7174
6
1.6k
Featured
See All Featured
How to Build an AI Search Optimization Roadmap - Criteria and Steps to Take #SEOIRL
aleyda
1
2.1k
Fireside Chat
paigeccino
42
4k
The browser strikes back
jonoalderson
0
1.4k
Faster Mobile Websites
deanohume
310
32k
How STYLIGHT went responsive
nonsquared
100
6.2k
SEOcharity - Dark patterns in SEO and UX: How to avoid them and build a more ethical web
sarafernandez
0
220
How to train your dragon (web standard)
notwaldorf
97
6.7k
How to Get Subject Matter Experts Bought In and Actively Contributing to SEO & PR Initiatives.
livdayseo
0
160
Chasing Engaging Ingredients in Design
codingconduct
0
240
How Fast Is Fast Enough? [PerfNow 2025]
tammyeverts
3
660
How to Create Impact in a Changing Tech Landscape [PerfNow 2023]
tammyeverts
55
3.4k
Distributed Sagas: A Protocol for Coordinating Microservices
caitiem20
333
23k
Transcript
ݠन۞ ূפযо ֗ৡ Facebook F8 ӣળࢿ Developer Circles Seoul F8
Viewing Meetup ScatterLab Machine Learning Engineer
F8 & ML F8 Overview Introduction # Timeline Experience Total
Review 1. 2. 3. 4. 5.
Source: Optional source placeholder. Delete if not used. ӣળࢿ (Junseong
Kim) fb.com/codertimo ScatterLab Machine Learning Engineer ੌ࢚ച ੋҕמ ೝಯ Machine Learning Team (সӝמਃਗ) Naver Clova AI Research Intern Atlas Guide NLP/Machine Learning Engineer General Domain Dialog System/Chatbot (Chitchat) General Sentence Embedding (BERT, ElMo) And Diverse Natural Language Processing Tasks… Neural Machine Translation codertimo/BERT-pytorch 2.4k+ stars Research Domain Experience 2017, 2014 Intel ISEF Computer Science Finalist 2016, 2014 Intel KSEF Computer Science Grand Award Working Experience
F8 Overview
2019 Facebook F8 Overview ױࣽ ࢎۈр connectionਸ ֈয ழޭפ౭ܳ ҳࢿೞחؘ
ನழझ Facebook UI/UX with Community | Dev Circle
2019 Facebook F8 Overview ױࣽ ࢎۈр connectionਸ ֈয ழޭפ౭ܳ ҳࢿೞחؘ
ನழझ ࢎۈр ழޭפா࣌ਸ ೞחؘী ब ؼ ࣻ ب۾ Facebook UI/UX with Community | Dev Circle Messenger, WhatsApp, Portal
2019 Facebook F8 Overview ױࣽ ࢎۈр connectionਸ ֈয ழޭפ౭ܳ ҳࢿೞחؘ
ನழझ ࢎۈٜ ۄߡदܳ ୭ೠ ࠁഐೡ ࣻ ب۾ ࢎۈр ழޭפா࣌ਸ ೞחؘী ब ؼ ࣻ ب۾ Facebook UI/UX with Community | Dev Circle Messenger, WhatsApp, Portal On-Device Model, Encryption
2019 Facebook F8 Overview ױࣽ ࢎۈр connectionਸ ֈয ழޭפ౭ܳ ҳࢿೞחؘ
ನழझ ࢎۈٜ ۄߡदܳ ୭ೠ ࠁഐೡ ࣻ ب۾ ࢎۈр ழޭפா࣌ਸ ೞחؘী ब ؼ ࣻ ب۾ ݠन۞ ݽ؛ਸ ా೧ ࠁ ڙڙೠ Facebook UI/UX with Community | Dev Circle Messenger, WhatsApp, Portal On-Device Model, Encryption NLP, Vision, Audio
2019 Facebook F8 Overview ױࣽ ࢎۈр connectionਸ ֈয ழޭפ౭ܳ ҳࢿೞחؘ
ನழझ ࢎۈٜ ۄߡदܳ ୭ೠ ࠁഐೡ ࣻ ب۾ ࢎۈр ழޭפா࣌ਸ ೞחؘী ब ؼ ࣻ ب۾ ݠन۞ ݽ؛ਸ ా೧ ࠁ ڙڙೠ ࢎۈٜীѱ ࢜۽ ҃ਸ Facebook UI/UX with Community | Dev Circle Messenger, WhatsApp, Portal On-Device Model, Encryption NLP, Vision, Audio Oculus
2019 Facebook F8 Overview ױࣽ ࢎۈр connectionਸ ֈয ழޭפ౭ܳ ҳࢿೞחؘ
ನழझ ࢎۈٜ ۄߡदܳ ୭ೠ ࠁഐೡ ࣻ ب۾ ࢎۈр ழޭפா࣌ਸ ೞחؘী ब ؼ ࣻ ب۾ ݠन۞ ݽ؛ਸ ా೧ ࠁ ڙڙೠ ࢎۈٜীѱ ࢜۽ ҃ਸ Facebook UI/UX with Community | Dev Circle Messenger, WhatsApp, Portal On-Device Model, Encryption NLP, Vision, Audio Oculus ৮ ࢜܂ѱ Facebook
F8 & ML
Facebook F8ীࢲ ࠁৈળ ಕझ࠘ AI Research to Production : Pytorch
v1.0 stable release Social Protection: աࢂ ѱदޛ, ؆Ӗٜ ఠ݂ Ҵয ӝ҅ߣ NMT(Neural Machine Translation) Video, Image Recognition & Captioning: ݫఋؘఠ ୶ Semi-Supervised Learning : ࣗӏݽ ۨ࠶ ؘఠ + ӏݽ णؘఠ ࢿ & ച ੋध ….
Facebook F8ীࢲ ࠁৈળ ಕझ࠘ AI Research to Production : Pytorch
v1.0 stable release Social Protection: աࢂ ѱदޛ, ؆Ӗٜ ఠ݂ Ҵয ӝ҅ߣ NMT(Neural Machine Translation) Video, Image Recognition & Captioning: ݫఋؘఠ ୶ Semi-Supervised Learning : ࣗӏݽ ۨ࠶ ؘఠ + ӏݽ णؘఠ ࢿ & ച ੋध …. Ӓীب ࡳਵۄҊ ೞݶ
Facebook F8ীࢲ ࠁৈળ ಕझ࠘ AI Research to Production : Pytorch
v1.0 stable release Ҵয ӝ҅ߣ NMT(Neural Machine Translation) Semi-Supervised Learning : ࣗӏݽ ۨ࠶ ؘఠ + ӏݽ णؘఠ Ӓীب ࡳਵۄҊ ೞݶ
Semi-Supervision Learning: ળب ण ࣚ൜ షֈ ୭Ҋ ҕѺࣻ۽ ܻ ও
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Semi-Supervision Learning: ળب ण ࣚ൜ షֈ ୭Ҋ ҕѺࣻ۽ ܻ ও
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding ࣚ൜ షֈ ୭Ҋ ҕѺࣻ۽ ܻ ও ݠन۞ই ۠Ѣ ݏ୶ݶࢲ যખ ҕࠗ೧ࠊۄ! ݃!
Semi-Supervision Learning: ળب ण ࣚ൜ షֈ ୭Ҋ ҕѺࣻ۽ ܻ ও
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding ࣚ൜ షֈ ୭Ҋ ҕѺࣻ۽ ܻ ও ݠन۞ই ۠Ѣ ݏ୶ݶࢲ যખ ҕࠗ೧ࠊۄ! ݃! ֎ ޤ ۠ѩפ
Semi-Supervision Learning: ળب ण BERT: Pre-training of Deep Bidirectional Transformers
for Language Understanding ֎ Ӓۗ؊פ ࢎۈਸ ҂যਃ
Semi-Supervision Learning: ળب ण BERT: Pre-training of Deep Bidirectional Transformers
for Language Understanding https://github.com/codertiimo/BERT-pytorch ӵঌഘࠁ
Semi-Supervision Learning: ળب ण BERT: Pre-training of Deep Bidirectional Transformers
for Language Understanding য় Ӓ۞ݶ ۨ࠶݂ Ѣ হযب ৮ ૨ ੜغա? Ѥ ഄݺঠ!! ա য়٣য়ب ੜغա?????
Semi-Supervision Learning: ળب ण BERT: Pre-training of Deep Bidirectional Transformers
for Language Understanding ֎ Ӓۗणפ. ࠗ ֫ ࢿמਸ যਃ ݆ࣻ ࢎۈ ࣻরѤ ؘఠܳ ੌੌ ۨ࠶݂ೞח ߑधਸ ୭ࣗച ೣ
Ҵয ӝ҅ߣ NMT(Neural Machine Translation) ࣚ൜ షֈ ୭Ҋ ҕѺࣻ۽ ܻ
ও Sonny has became the best striker in Tottenham ટોનીહેમમાં સોની +ે, -.ાઈકર બ3યા છ ે ಕझ࠘ ࣁ҅ 100+ ࢚ যܳ ਗೞח ࢲ࠺झ Ҵয ߣ ಕझ࠘ য়ے োҳ җઁ
Ҵয ӝ҅ߣ NMT(Neural Machine Translation) ೞ݅ ߣ NMTח زੌೠ
Ҵয ޙ ਵ۽ ਃೣ ۠ ؘఠܳ ݅٘ח Ѫ Ѣ ࠛоמী оӰҊ ցޖ ݆ दрҗ ࠺ਊ ਃೣ
Ҵয ӝ҅ߣ NMT(Neural Machine Translation) ೞ݅ ߣ NMTח زੌೠ
Ҵয ޙ ਵ۽ ਃೣ ۠ ؘఠܳ ݅٘ח Ѫ Ѣ ࠛоמী оӰҊ ցޖ ݆ दрҗ ࠺ਊ ਃೣ Semi-Supervision Learningਵ۽ ߓ য ೧ܳ Ҵয۽ ഛೞח ݽ؛ ѐߊ п য Representation Space ܳ Matchingೞח ߑߨۿਸ ઁदೣ + On-device ীࢲ प೯оמೞب۾
Ҵয ӝ҅ߣ NMT(Neural Machine Translation) https://code.fb.com/ai-research/laser-multilingual-sentence-embeddings/
Research to Production : Pytorch v1.0 stable release ѐߊীࢲח ୭ъ
UXܳ یೞ݅ Production Scaleীࢲח ୭ড؍ Pytorch Pythonࢿמҗ Dynamic Graph Computation ࣘب ೞ ޙઁ
Research to Production : Pytorch v1.0 stable release ONNX, Caffe2,
Jit compile Production ޙઁ ೧Ѿ + MOBILE-ML
Experience
ࢎप ݈ب উغח ѐ ী Google I/O যࢲ ҕ۽ ࠺೯ӝ
ҳੋ ש ೧ழహ ଵо + ש ഐభীࢲ ࣼ߅ ࣃۆоח ࠺೯ӝীࢲ F8оח Ҋ١Ү ߓ ୡো ഥ ݃ஜژ ҳӖীࢲ Ӕޖೞन ࢶߓ৬ ୡ ߣѐ & ైয ಕझ࠘ীࢲ ࠌ؍ ٜ࠙ਸ ӏݽ۽ য়ۄੋ ࢶ ਮഄש ֎ਕী ઑੋ೧ పठۄ য়షੌۤب प
ழޭפ౭
ࣁ࣌
ഥࢎ ഘࠁ(?) ੌ࢚ച ੋҕמ ೝಯ
߸ైয
Total Review
݈ Ҋ ਬೞҊ જ ࢎۈٜਸ ࢎӈҊ ڪয֥ ࣻ Ҋ নೠ
҃ਸ ೡ ࣻ Ҋ ৈ۞ ঠӝܳ ա־ݴ ৈ۞ ࢎਸ ନਵݴ ಕझ࠘ Ҷૉܳ Ѷೞݴ ೞܖೞܖо ೯ࠂೞҊ ח ࣁ࣌ ցޖ ݆Ҋ ݡਸѪ ݃पѪ օ۰णפ ࢎیפ❤ хࢎפ Facebook F8 +Dev Circle ղ֙ী ࢎ࠺۽ۄب ԙ द য়Ҋरযਃ റഥ উೞपѩפ ƃƃ
Thank you