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
Introduction to LSTM
Search
KenjiH
July 31, 2017
Technology
0
130
Introduction to LSTM
KenjiH
July 31, 2017
Tweet
Share
More Decks by KenjiH
See All by KenjiH
Quantum Computation and Quantum Information (Chapter 5.4)
kenjih
0
76
Quantum Computation and Quantum Information (Chapter 5.2)
kenjih
0
48
Quantum Computation and Quantum Information (Chapter 3.1)
kenjih
0
91
Quantum Computation and Quantum Information (Chapter 1.1 - 1.3)
kenjih
0
98
Deep Learning with Python Chap.1
kenjih
0
58
Deep Learning with Python Chap.2
kenjih
0
62
Computer Vision Applications of MRF , CRF and Other Grid Models
kenjih
0
260
論文紹介:Probabilistic Matrix Factorization
kenjih
0
770
AAAI・WSDM 2017論文読み会
kenjih
0
1.4k
Other Decks in Technology
See All in Technology
開発と脆弱性と脆弱性診断についての話
su3158
1
1.1k
あとはAIに任せて人間は自由に生きる
kentaro
3
1.1k
Figma + Storybook + PlaywrightのMCPを使ったフロントエンド開発
yug1224
9
2.6k
見てわかるテスト駆動開発
recruitengineers
PRO
5
360
認知戦の理解と、市民としての対抗策
hogehuga
0
360
コスト削減の基本の「キ」~ コスト消費3大リソースへの対策 ~
smt7174
2
140
新卒(ほぼ)専業Kagglerという選択肢
nocchi1
1
2.3k
夢の印税生活 / Life on Royalties
tmtms
0
280
つくって納得、つかって実感! 大規模言語モデルことはじめ
recruitengineers
PRO
24
6.1k
フルカイテン株式会社 エンジニア向け採用資料
fullkaiten
0
8.6k
あなたの知らない OneDrive
murachiakira
0
240
実践データベース設計 ①データベース設計概論
recruitengineers
PRO
3
250
Featured
See All Featured
Imperfection Machines: The Place of Print at Facebook
scottboms
268
13k
Statistics for Hackers
jakevdp
799
220k
No one is an island. Learnings from fostering a developers community.
thoeni
21
3.4k
Faster Mobile Websites
deanohume
309
31k
Rebuilding a faster, lazier Slack
samanthasiow
83
9.1k
Raft: Consensus for Rubyists
vanstee
140
7.1k
How to Think Like a Performance Engineer
csswizardry
25
1.8k
Code Review Best Practice
trishagee
70
19k
Balancing Empowerment & Direction
lara
2
590
10 Git Anti Patterns You Should be Aware of
lemiorhan
PRO
656
61k
A Tale of Four Properties
chriscoyier
160
23k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
23
1.4k
Transcript
-45.ೖ Kenji Hayashida
-45.ͱ MPOHTIPSUUFSNNFNPSZ SFDVSSFOUOFVSBMOFUXPSLNPEFMͷҰछ γʔέϯεσʔλͷੳʹར༻ ςΩετʢࣗಈ༁ʣ ΦʔσΟΦʢԻೝࣝʣ ϏσΦʢΠϝʔδΩϟϓγϣϯʣ
/FVSBM/FUXPSLT <>ΑΓҾ༻
3//T ྫɿөըͷը૾ʹϥϕϧΛ͚͍ͭͨ աڈϑϨʔϜͷใΛར༻͢Δ ֤ը૾Λݟͨͱ͖ɺϥϕϧΛग़ྗͭͭ͠ɺࣝΛߋ৽͢ΔϞσϧΛߟ͑Δ <>ΑΓҾ༻
-45.T 3//ͷ ࣝͷߋ৽͕ແடংʹߦΘΕΔʢใ͕͙͢ʹมΘΔࣦΘΕΔʣ ظهԱΛ࣋ͭͷ͕ࠔ -45. ظϝϞϦ ٫ػߏʢظهԱ͔ΒԿΛอ࣋͢ΔࣺͯΔ͔அʣ อଘػߏʢ৽͍͠σʔλ͔Β͑ΒΕͨใΛظهԱʹՃʣ ϫʔΩϯάϝϞϦ ϑΥʔΧεʢ͍·ඞཁͳใͷΈΛར༻ʣ
-45.T ࣌ࠁͷͱ͖ೖྗɹ͓Αͼ લ࣌ࠁͷظϝϞϦɹɹɹɹɺϫʔΩϯάϝϞϦ͕༩͑ΒΕΔ xt ltmt 1 wmt 1 t
-45.T ࣌ࠁͷͱ͖ೖྗɹ͓Αͼ લ࣌ࠁͷظϝϞϦɹɹɹɹɺϫʔΩϯάϝϞϦ͕༩͑ΒΕΔ xt ltmt 1 wmt 1 t ظϝϞϦͷͲͷ෦Λอ࣋͢Δ͔
อ࣋ɺഁغ
-45.T ࣌ࠁͷͱ͖ೖྗɹ͓Αͼ લ࣌ࠁͷظϝϞϦɹɹɹɹɺϫʔΩϯάϝϞϦ͕༩͑ΒΕΔ xt ltmt 1 wmt 1 t ظϝϞϦʹ͢ใ
-45.T ࣌ࠁͷͱ͖ೖྗɹ͓Αͼ લ࣌ࠁͷظϝϞϦɹɹɹɹɺϫʔΩϯάϝϞϦ͕༩͑ΒΕΔ xt ltmt 1 wmt 1 t ظϝϞϦʹ͢ใͷ͏ͪɺ
Ͳͷ෦͕ॏཁ͔ʁ
-45.T ࣌ࠁͷͱ͖ೖྗɹ͓Αͼ લ࣌ࠁͷظϝϞϦɹɹɹɹɺϫʔΩϯάϝϞϦ͕༩͑ΒΕΔ xt ltmt 1 wmt 1 t ظϝϞϦΛߋ৽
-45.T ࣌ࠁͷͱ͖ೖྗɹ͓Αͼ લ࣌ࠁͷظϝϞϦɹɹɹɹɺϫʔΩϯάϝϞϦ͕༩͑ΒΕΔ xt ltmt 1 wmt 1 t ظϝϞϦͷ͏ͪɺ͍·ॏཁͳͷʁ
-45.T ࣌ࠁͷͱ͖ೖྗɹ͓Αͼ લ࣌ࠁͷظϝϞϦɹɹɹɹɺϫʔΩϯάϝϞϦ͕༩͑ΒΕΔ xt ltmt 1 wmt 1 t ϫʔΩϯάϝϞϦΛநग़
-45.T <>ΑΓҾ༻
આ໌Ͱ༻͍ͨݺশ Ұൠతͳݺশ ظϝϞϦMUNU ηϧঢ়ଶʢDFMMTUBUFʣDU ϫʔΩϯάϝϞϦXNU ӅΕʢIJEEFOTUBUFʣIU ٫ػߏSFNFNCFSU ٫ήʔτʢGPSHFUHBUFʣGU อଘϕΫτϧTBWFU ೖྗήʔτʢJOQVUHBUFʣJU
ϑΥʔΧεϕΫτϧGPDVTU ग़ྗήʔτʢPVUQVUHBUFʣPU ͔Γ໊͍͢લͰઆ໌͖͕ͯͨ͠ɺ֤ཁૉผ໊ͰݺΕΔ͜ͱ͕ଟ͍ -45.EJBHSBNT
-45.EJBHSBNT <>ΑΓҾ༻
GU٫ήʔτ ϝϞϦʔηϧ$UͷͲͷ෦Λอ࣋͢Δ͔ อ࣋ɺഁغ -45.EJBHSBNT <>ΑΓҾ༻
JUɹೖྗήʔτ ৽͍͠ೖྗ͔Β͑ΒΕͨใͷ͏ͪɺͲ͕͜ॏཁ͔ -45.EJBHSBNT <>ΑΓҾ༻
ηϧঢ়ଶΛߋ৽ -45.EJBHSBNT <>ΑΓҾ༻
ϝϞϦηϧ͔Βग़ྗήʔτΛ௨ͯ͡ɺ ӅΕঢ়ଶΛऔಘ -45.EJBHSBNT <>ΑΓҾ༻
-45.WBSJBOUT ͷ͖ͧ݀ʢQFFQIPMFʣ ֤ήʔτ͔Βηϧͷঢ়ଶΛࢀরͰ͖Δ -45.ʹ͞·͟·ͳѥछ͕ଘࡏ͢Δ <>ΑΓҾ༻
-45.WBSJBOUT ٫ήʔτͱೖྗήʔτͷಉҰԽ -45.ʹ͞·͟·ͳѥछ͕ଘࡏ͢Δ <>ΑΓҾ༻
-45.WBSJBOUT (BUFE3FDVSSFOU6OJU (36 ɹηϧͱӅΕঢ়ଶΛ༥߹ ɹ٫ήʔτͱೖྗήʔτΛ༥߹ -45.ʹ͞·͟·ͳѥछ͕ଘࡏ͢Δ <>ΑΓҾ༻
-45.ͷจհ ͔͜͜ΒҎԼͷจͷʹͳΓ·͢ l4FRVFODFUP4FRVFODF7JEFPUP5FYUz 4VCIBTIJOJ7FOVHPQBMBO .BSDVT 3PISCBDI +F⒎%POBIVF 3BZNPOE.PPOFZ 5SFWPS%BSSFMM ,BUF4BFOLP
*OUSPEVDUJPO ϏσΦΩϟϓγϣϯͷ͍͠ͱ͜Ζ ՄมͷೖྗͱՄมͷग़ྗ ΦϒδΣΫτɺγʔϯɺΞΫγϣϯɺ ଐੑͷଟ༷ੑ ݦஶͳίϯςΩετͷదͳநग़ ϏσΦΫϦοϓͱઆ໌จ͔ΒͳΔσʔληοτΛ༻͍ͯɺϏσΦͷΩϟϓγϣ ϯΛੜ͢ΔϞσϧʢ475ʣΛֶश FOEUPFOEͰֶशՄೳͳϞσϧ 4FRVFODFUP4FRVFODFͷֶशͰ࣮ͷ͋Δ-45.Λ͏
<>ΑΓҾ༻
3FMBUFE8PSL ॳظͷϏσΦΩϟϓγϣϯͷख๏ εςʔδͷύΠϓϥΠϯ ҙຯͷ͋ΔίϯςΩετͷநग़ʢTVCKFDU WFSC PCKFDUʣ ςϯϓϨʔτΛͬͯจষΛੜ ީิͱͳΔPCKFDU BDUJPO TDFOFΛࣝผ͢ΔͨΊʹݸผͷࣝผث͕ඞཁ
ݴޠදݱೳྗ͕ेͰͳ͍ʢୀ۶ͳจষʹͳΓ͕ͪʣ ࠷৽ͷϏσΦΩϟϓγϣϯͷख๏ $//Λͬͯಛநग़ͯ͠ɺͦͷಛϕΫτϧΛ୯ޠͷγʔέϯεʹσ ίʔυ͢Δ ظґଘؔΛֶश͢Δͷʹదͨ͠-45.͕ΘΕΔ͜ͱ͕ଟ͍
3FMBUFE8PSL ࠷৽ͷϏσΦΩϟϓγϣϯͷख๏4VCIBTIJOJΒʢʣ NFBOQPPMJOHΛͬͯɺϏσΦͷใΛҰͭͷಛϕΫτϧͰදݱ ϑϨʔϜͷॱংؔΛແࢹ͍ͯ͠ΔͨΊɺ࣌ؒใΛ׆༻Ͱ͖ͳ͍ <>ΑΓҾ༻
3FMBUFE8PSL ࠷৽ͷϏσΦΩϟϓγϣϯͷख๏-J:BPΒʢʣ ϏσΦΛ 8 ) 5 ͷ̏࣍ݩͰදݱ͠ɺ͔ͦ͜ΒಛྔΛநग़ நग़ͨ͠ಛྔΛ%$//ʹೖྗ͠ɺBDUJPOΛࣝผ BUUFOUJPOΛͬͯϑϨʔϜͷಛྔͷॏΈΛௐ ࣌ؒ࣠ͷಛநग़
BUUFOUJPOʹΑͬͯύϑΥʔϚϯε্ʹޭ <>ΑΓҾ༻
"QQSPBDI ೖྗϑϨʔϜγʔέϯε ग़ྗ୯ޠγʔέϯε ҎԼͷ͖݅ͭ֬Λਪఆ͢ΔϞσϧΛߟ͑Δ ೖग़ྗͷγʔέϯε͕Մม͚ͩͲͲ͏ͬͯରॲ͢Δʁ ϑϨʔϜγʔέϯεΛͭͣͭΤϯίʔυ͠ɺϏσΦΛҰͭͷજࡏมදݱͰද͢ જࡏมϕΫτϧΛ୯ޠͷγʔέϯεʹσίʔυ͢Δ ͭલͷ୯ޠ ϏσΦͷજࡏมදݱͷ࠷৽ঢ়ଶ
"QQSPBDI ݸͷӅΕϢχοτΛ࣋ͭͭͷ-45.ΛελοΫ͢Δ ৭ͷ-45.ϨΠϠʔը૾ͷΤϯίʔυΛ୲ ৭ͷ-45.ϨΠϠʔ୯ޠͷσίʔυΛ୲ QBEθϩϕΫτϧɺ#04CFHJOJOHPGTFOUFODFλάɺ&04FOEPGTFOUFODF λά ҎԼͷରΛ࠷େԽ͢ΔΑ͏ʹύϥϝʔλΛֶश ͨͩ͠ σίʔμ-45.ͷग़ྗ <>ΑΓҾ༻
7JEFPBOE5FYU3FQSFTFOUBUJPO 3(#GSBNFT $//Ͱಛநग़ *NBHF/FUEBUBTFUͷը૾ͰֶशࡁΈͷ$// 0QUJDBM'MPX BDUJWJUZͷࣝผʹ༗ޮͰ͋ΔͱݴΘΕ͍ͯΔ qPXJNBHFTΛ$//ʹೖྗ͠ಛྔΛੜ 0QUJDBM'MPXΛBDUJWJUZࣝผ͢ΔͨΊͷσʔληοτʢ6$'WJEFPEBUBTFUʣ ͰֶशࡁΈͷ$//Λͬͨ ্هͭͷಛΛͬͯ͑ΒΕͨ݁ՌΛ݁߹ͨ͠ͷΛ࠷ऴతͳείΞͱ͢Δ
ϏσΦͷಛྔ ςΩετͷಛྔ POFIPUFODPEJOHʢPG/DPEJOHʣΛઢܗมͯ͠ɺ࣍ݩʹͨ͠ͷ
&YQFSJNFOU <>ΑΓҾ༻
"QQFOEJY0QUJDBM'MPX ࿈ଓ͢ΔϑϨʔϜ͔ΒମͷҠಈใΛநग़͢Δख๏ <>ΑΓҾ༻ <>ΑΓҾ༻
3FGFSFODFT <>&YQMPSJOH-45.T <>6OEFSTUBOEJOH-45./FUXPSLT <>4FRVFODFUP4FRVFODFr7JEFPUP5FYU <>5SBOTMBUJOH7JEFPTUP/BUVSBM-BOHVBHF6TJOH%FFQ3FDVSSFOU/FVSBM/FUXPSLT <>%FTDSJCJOH7JEFPTCZ&YQMPJUJOH5FNQPSBM4USVDUVSF <>0QUJDBM'MPX