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
개발자의 인공지능 뽀개기
Search
Teddy
August 07, 2018
Programming
0
160
개발자의 인공지능 뽀개기
Teddy
August 07, 2018
Tweet
Share
More Decks by Teddy
See All by Teddy
스타트업 개발자의 이야기 (국민대)
teddykims
1
100
Git 기초
teddykims
0
83
클라이언트 권한 부여, 인증 시스템, OAuth-JWT
teddykims
0
110
2018.06, 싸이그래머 '심리상담과 IT'
teddykims
0
120
WebSocket, Socket.io
teddykims
1
57
2018.05, 소물, '소프트웨어와 미래 기술'
teddykims
0
82
소프트웨어와 미래 기술 (Software, Future Tech)
teddykims
0
190
리눅스 메모리 보호기법, 보안시스템의 이해
teddykims
0
240
무선 네트워크 해킹과 보안 (Network Hacking)
teddykims
0
220
Other Decks in Programming
See All in Programming
コマンドとリード間の連携に対する脅威分析フレームワーク
pandayumi
1
460
Claude Codeと2つの巻き戻し戦略 / Two Rewind Strategies with Claude Code
fruitriin
0
140
izumin5210のプロポーザルのネタ探し #tskaigi_msup
izumin5210
1
140
「ブロックテーマでは再現できない」は本当か?
inc2734
0
1k
AIエージェント、”どう作るか”で差は出るか? / AI Agents: Does the "How" Make a Difference?
rkaga
4
2k
Best-Practices-for-Cortex-Analyst-and-AI-Agent
ryotaroikeda
1
110
CSC307 Lecture 10
javiergs
PRO
1
660
HTTPプロトコル正しく理解していますか? 〜かわいい猫と共に学ぼう。ฅ^•ω•^ฅ ニャ〜
hekuchan
2
690
16年目のピクシブ百科事典を支える最新の技術基盤 / The Modern Tech Stack Powering Pixiv Encyclopedia in its 16th Year
ahuglajbclajep
5
1k
なぜSQLはAIぽく見えるのか/why does SQL look AI like
florets1
0
480
疑似コードによるプロンプト記述、どのくらい正確に実行される?
kokuyouwind
0
390
Apache Iceberg V3 and migration to V3
tomtanaka
0
170
Featured
See All Featured
Why You Should Never Use an ORM
jnunemaker
PRO
61
9.7k
Building Adaptive Systems
keathley
44
2.9k
Unlocking the hidden potential of vector embeddings in international SEO
frankvandijk
0
170
Sam Torres - BigQuery for SEOs
techseoconnect
PRO
0
190
Ruling the World: When Life Gets Gamed
codingconduct
0
150
Building the Perfect Custom Keyboard
takai
2
690
Taking LLMs out of the black box: A practical guide to human-in-the-loop distillation
inesmontani
PRO
3
2k
Navigating the Design Leadership Dip - Product Design Week Design Leaders+ Conference 2024
apolaine
0
190
Responsive Adventures: Dirty Tricks From The Dark Corners of Front-End
smashingmag
254
22k
Between Models and Reality
mayunak
1
190
Impact Scores and Hybrid Strategies: The future of link building
tamaranovitovic
0
200
[SF Ruby Conf 2025] Rails X
palkan
1
760
Transcript
ѐߊܳ ਤೠ ੋҕמ ѐӝ
ա݅ ࢚ࢎܳ ݅աח о ए ߑߨ बܻ࢚ ݫन ۽झ [TROST]
()ോ݃ஹಌפ ()ോ݃ஹಌפ ӣక
ې
None
None
None
None
None
None
߸ച द
None
None
None
None
None
None
None
None
None
ੋҕמ ੋрਸ ӡ ࣻ Ҋ ૐݺػ ߣ૩ ࢎѤ
AI (Artificial Intelligence, ੋҕמ)
ӝ҅۽ࠗఠ ٜ݅য מ ੋр ࢎҊ۱ਸ צ ੋрۢ ࢤпೞח ஹೊఠ
ੑ۱߉ чਵ۽ ࠗఠ Ѿҗ чਸ ղח ೣࣻܳ ٜ݅য ղח Ѫ ъੋҕמ <-> ডੋҕמ (അ ੋҕמ)
ੋҕמ ޤ? ੋҕੋ Ѫ ޤ? מ ޤ? ӝ҅ب ࢤпೡ ࣻ
ਸө? ܻח যڌѱ ࢤпೞחѦө?
“ੋҕמ ߊݺۆ زରীࢲ ߄௰ܳ ڍযմ ٍ Ӓ ܻী ߊਸ ׳ӝ
ਤ೧ Ҋबೞח Ѫ.” “ੋҕמ ই पഅغ ঋ ޖо”
ੋҕמ ࠄې ݾ ‘बܻ’ী ೠ पੋ Ӕ য מ ޖੋ
ߋഃղח Ѫ ୭ୡ ݾ ؘఠ৬ धٜ ݽৈ ܲ ࠙ঠীࢲ ࢎ Ѿҗ ஏ ١ ঘ࣌ਸ ਤ೧ ѐߊغӝ दೞݴ ੋҕמ җ ߊ द
194x ~ 195x – ੋҕמ ࢤ “গפঈ ࢤ” > “ౚ݂
పझ” > “IBM ୭ୡ AI ஶಌ۠झ” > “AI ࢤ” 195x ~ 196x – ൞ݎ ੋҕמ “ੋр ࢤпೞח ӝ҅” > “ੋҕמ ѐߊ য ’LISP’ ࢤ” > “ৈ۞ ੋҕמ ۿ ١” 196x ~ 197x – ੋҕמ ঐӝ “Ҵ/Ҵ ࠗ ਗ ױ” > “ੋҕ न҃ݎ ೠ҅ ૐݺ” 197x ~ 199x – ޙо दझమ “ࢿ ആঘ ജ ױ ӏ ӝ߈ ޙо दझమ” > “স҅ ۞࠳” > “ੌࠄҗ Ҵ ষդ ਗ” 199x ~ 200x – ੋҕמ അ৬ ې “ۡ ֎ਕ, ਬ ঌҊ્ܻ ١” > “ؘఠ әૐ” > “ੋҕמ ӝٜ҅ ࢤ”
IBM ٩ ࠶ܖ IBM '٩ ࠶ܖ’ ۽ ੋр झିೖ оܻ
झ۽৬ ѹܞࢲ थܻ IBM Watson োয ܻܳ ਤ೧ࢲ ٜ݅য ஹೊఠ. ઁಌ٣ ௰ૉࣳীࢲ ିೖٜਸ ׂ۞ߡ۷. ۨ٘ಫ ۨ٘ಫ Ҵ ܻನפই݀(UCLA) ઁܻ ࠳ےষ Үࣻ৬ োҳ ѐߊೠ ۽Ӓ۔ਵ۽ ߧદ ࠁܳ ࠙ࢳ೧ 10~12दр ٍ ߧદо ੌযզ दрҗ ࣗܳ بೞח ۽Ӓ۔. ۽झঙۨझ ҃ (LAPD)җ दগౣ ҃ ١ ੌࠗ Ҵ ҃җ Ҵ ٜ҃ ۨ٘ಫਸ بੑೠ റ ߧદਯ 20% о Ҋ ೠ. ۄ झப ֎ӝݮ۠(CMU)ীࢲ ѐߊ೮Ҋ, Ҵ ೖஎߡӒ ҃ 2016֙ 10ਘࠗఠ ਊೞҊ ח ۽Ӓ۔ਵ۽, খਵ۽ ߊࢤೡ ߧદ दрҗ ࣗܳ ஏ೧ ೧ ࠁܳ ٜ҃ ֢࠘, झ݃ಪী ղࠗ ాनݎਸ ా೧ ׳ೠ.
Apple Siri Apple iPhone ী ػ োয ܻ AI.
೧ઉ ݅, ࢿੋध ࠗ࠙ पदрਵ۽ ߊ׳ೞח Ѫਵ۽ ୶ػ. ҳӖ ঌҊ ਬۣ ߄ق ିೖҗ Ѿೞৈ थܻ೮ਵݴ, 2016֙ 3ਘ ࣁج 9ױҗ Ҵীࢲ 4थ 1ಁ۽ थܻܳ Ѣف. ࢿ ࠻झ࠺ ࢿо ҕѐೠ Ҋࢿמ ੋҕמ ࠺ࢲ গܻா࣌ ࢎਸ ନয ޛܳ زਵ۽ ੋधೡ ࣻ ח ࠁػ ӝמਸ оҊ . बब AIܳ ߑೞҊ ݅ ࢎप ӝઓী ١۾غয ח ޙٜী ೧ ١۾غয ח ߸ਸ ೞח рױೠ ࣻળ ߑधۄ ੋҕמۄ ೞӟ ިೞ. ޙٜਸ ࠙ࢳೞח Ѫب ױযܳ оҊ ࠙ࢳೞח ࣻળ рױೠ ߑध.
ੋҕמ ܻо ਗೞח ژח ޖ ౠ ೯زਸ, ӝ҅о झझ۽ ஏ,
౸ױೞৈ ח Ѫ
णػ ղਊਸ ӝ߈ਵ۽ ېܳ ஏೠ. ਤ ஏ җਸ ਤ೧ ӝ҅ܳ
णदఃח ੌ۲ স ਃ ӝ҅ ण ষդ ন ؘఠܳ झझ۽ णೞҊ ܻೞৈ ޙઁী ೠ ೧ਸ ইղח ӝࣿ.
ӝ҅ח যڌѱ णਸ?
None
None
աޖ: ӝա оо ݾ۽ ػ ֙ࢤ धޛ ୶ ೠ҅
୶࢚Ѣա ҳੋ पо ח ࢚ ই ӝୡੋ ѐ֛ਸ ݠ݁ࣘী ܻѱ ೞӝ ਤ೧ࢲ, ࢚ਸ աఋղח ࣻ ݆ ࢠਸ ߉ইٜݶࢲ ݠ݁ࣘীࢲ ୷ػ ഋక۽ ݀೧ աоب۾ ֱী ݐӝח Ѫ աޖ ࣻ ݆ ࢠٜਸ ࠁݴ ೨ब ౠٜ ୷ػ Ѿҗޛ۽ ֱ ࣘীࢲ ই ୶࢚ੋ ഋక۽ ٜ݅য
ਬইӝ द ‘աޖ’ ۄח ࢚ ѐ֛ਸ ഊ؍ җҗ ݒ ൚ࢎ
ࣻ ݆ ࢠ ؕযܻܳ णೠ. => ࣻ ݆ ؘఠܳ णೠ.
दп, ୢп, റпੋ хп ࠁ/ౠٜਸ ֱী ೠ. => ӝ҅о ೧ೡ ࣻ ח ഋక۽ ࣻച (ܻ) ೞৈ ೠ.
None
ੋҕמ ? AI ? ӝ҅ण ? ݠन۞ ? ٩۞ ?
न҃ݎ ? ML ?
ੋҕמ ӝ҅ ण ٩۞ ੋҕמ(AI) > ݠन ۞(Machine Learning) >
٩۞ (Deep Learning)
- ୡӝ ٩۞ ‘ੋҕ न҃ݎ’ ۄח ઁ۽ 1942֙ ࡸܻо
दغ. - ੋр ֱ न҃ݎਸ ࠄٮ ݅ٚ ੋҕ न҃ݎ ݽ؛۽ ӝઓ ݠन۞ ೠ҅ (ؘఠী ನೣػ নೠ ߸ࣻܳ ঈ ೞ ޅೣ) ܳ ࠁ৮ೠ ߑߨਵ۽ ځয়ܰӝ द೮. - ੋр ֱ(۠) ҳઑܳ ਬࢎೞѱ ٜ݅যࠁח Ѽ݅ द ജ҃җ ؘఠ৬ ೞ٘ਝয ೠ҅۽ જ ಣਸ ޅ ೞणפ. ੋҕמ ӝ҅ ण ٩۞
ੋҕמ ӝ҅ ण ٩۞ - 2012֙, ҳӖҗ झగಌ٘ Andrew NG
Үࣻח 1݅ 6,000ѐ ஹೊఠ۽ ড 10র ѐ ࢚ न҃ݎਵ۽ ܞ ‘बகन҃ݎ(Deep Neural Network)’ਸ ҳഅ೮. ܳ ా೧ ਬౚ࠳ীࢲ 1,000݅ ѐܳ ࡳই ࠙ࢳೠ ٍ, ஹೊఠо ࢎۈҗ Ҋন ࢎਸ ࠙ܨೞب۾ ೞחؘ ࢿҕ೮ . ஹೊఠо ࢚ী աৡ Ҋন ഋక৬ ࢤӣ࢜ܳ ੋध ೞҊ ౸ױೞח җਸ झझ۽ णೞѱ ೠ Ѫ.
ੋҕמ ӝ҅ ण ٩۞ - ؘఠо ߑ೧Ҋ GPU ١ਵ۽
ࢿמҗ ജ҃ ࠁ৮غݴ 2010-12֙, बக ब҃ݎ, ٩۞ ࢤ ೮ण פ. - ঔܻ(CPU), ݫੋ (۳) - Ӓېܻ(GPU), ࠁઑ (߽۳) - CPU ח ࣽࢲ۽ ؘఠܳ ܻೞח ߑधী ౠചػ ҳ ઑܳ оҊ . GPU ח ࠺Ү рױೠ, ؏ ࠗझ۞ ੌਸ ബਯਵ۽ ܻೡ ࣻ . ܻ೧ঠ ೡ ݺ۸ য৬ ؘఠ ࢿѺী ٮۄ ٸ۽ח CPU, ٸ۽ח GPUо ࡅܳ ࣻо .
ੋҕמ Ҿӓੋ ݾҊ, ٩۞ਸ Ӓী ܰӝ ਤೠ ࣻ ݆ ࣻױ
ೞաੌ ࡺ!
Q. ٩۞ о ऽாੋоਃ? ޖઑѤ ੋо ਃ? Q. ח ޖट
ঌҊ્ܻਸ ࢎਊೞҊ रযਃ. যٸ ਃ? Q. ঌҊח ৮ ࢜۽ AI ੋѤоਃ?
Q. ઁ ࣼઁ न ೧ ࣻ աਃ? Q. ઁ թ,
ৈҳо غয ࣻ աਃ? Q. ঌҊ ऱ ੜ ೞաਃ?
ҕࠗ ೞҊ रযਃ! http://hunkim.github.io/ml/
ӝࣿ ਲ਼җ ې
None
None
None
None
None
None
None
None
None
None
None
None
ഒࢲ ೡ ࣻ ח Ѫ 0
ېח ѾҴ ౠ ӝࣿ, সҗ ݽٚ Ѫ ਲ਼ীࢲ द
4ର, Nର স ഄݺ ࠺ૉפझ ҕә ߑध ߸ച স ܻ࠙
(ؘఠ җ VS ੌ߈ੋ) ੌ࢚ ࢤഝ ۖಬച
झ౭࠵ ഐఊ “৮߷ೠ ੋҕמ ӝࣿ ѐߊ, ੋܨ ݷݎਸ ࠛ۞ৢࣻب..” ࠽
ѱ “ੋр ੌܻܳ ࡐਸ ۽ࠈীѱ ࣗٙࣁܳ ѥח ଼ਸ..”
None