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
人工知能と機械学習 / AI and ML
Search
Yukino Baba
PRO
September 26, 2019
Education
1
520
人工知能と機械学習 / AI and ML
Yukino Baba
PRO
September 26, 2019
Tweet
Share
More Decks by Yukino Baba
See All by Yukino Baba
大規模言語モデルのバイアス
yukinobaba
PRO
4
710
人間とAIの協働(駒場祭2023)
yukinobaba
PRO
7
1.4k
人工知能と機械学習 / Artificial Intelligence and Machine Learning
yukinobaba
PRO
4
1.2k
大規模言語モデル時代のHuman-in-the-Loop機械学習
yukinobaba
PRO
20
6k
壁のためのAIと卵のためのAI
yukinobaba
PRO
7
6.9k
人間と人工知能の協働
yukinobaba
PRO
0
7.7k
人工知能のしくみ / How AI learns
yukinobaba
PRO
1
320
Human-in-the-Loop 機械学習 / Human-in-the-Loop Machine Learning
yukinobaba
PRO
16
13k
CrowDEA: Multi-view Idea Prioritization with Crowds
yukinobaba
PRO
1
280
Other Decks in Education
See All in Education
ルクソールとツタンカーメン
masakamayama
1
890
ヘイトスピーチがある世界のコミュニケーション
ktanishima
0
150
情報処理工学問題集 /infoeng_practices
kfujita
0
120
20240810_ワンオペ社内勉強会のノウハウ
ponponmikankan
2
890
ACT FAST 20240830
japanstrokeassociation
0
330
Human Perception and Cognition - Lecture 4 - Human-Computer Interaction (1023841ANR)
signer
PRO
0
710
Introduction - Lecture 1 - Human-Computer Interaction (1023841ANR)
signer
PRO
0
1.7k
Kaggle 班ができるまで
abap34
1
190
勉強したらどうなるの?
mineo_matsuya
10
6.5k
1030
cbtlibrary
0
300
Evaluation Methods - Lecture 6 - Human-Computer Interaction (1023841ANR)
signer
PRO
0
700
認知情報科学科_キャリアデザイン_大学院の紹介
yuyakurodou
0
130
Featured
See All Featured
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
27
850
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
280
13k
Embracing the Ebb and Flow
colly
84
4.5k
What's new in Ruby 2.0
geeforr
343
31k
Scaling GitHub
holman
458
140k
A Modern Web Designer's Workflow
chriscoyier
693
190k
How STYLIGHT went responsive
nonsquared
95
5.2k
Distributed Sagas: A Protocol for Coordinating Microservices
caitiem20
329
21k
実際に使うSQLの書き方 徹底解説 / pgcon21j-tutorial
soudai
169
50k
Why You Should Never Use an ORM
jnunemaker
PRO
54
9.1k
How To Stay Up To Date on Web Technology
chriscoyier
788
250k
The Power of CSS Pseudo Elements
geoffreycrofte
73
5.3k
Transcript
ஜେֶใՊֶྨഅઇ೫ CBCB!DTUTVLVCBBDKQ ஜେֶڞ௨Պʢใʣ ʮσʔλαΠΤϯεʯ ਓೳͱػցֶश
/39 ຊߨٛͰֶͿ͜ͱ w ਓೳ͕Ͳ͏͍͏໘Ͱ׆༻͞Ε͍ͯΔͷ͔ w ͍·ͷਓೳʹԿ͕Ͱ͖Δͷ͔ w ػցֶशͰͲͷΑ͏ʹσʔλ͔ΒͷֶशΛߦ͏ͷ͔ ਓೳʹԿ͕Ͱ͖Δͷ͔ɾػցֶशͲ͏͍͏͘͠Έ͔ 2
ਓೳ
/39 ਓೳͱ w ਓೳ ʮతͳػցɺಛʹɺతͳίϯϐϡʔλϓϩάϥϜΛ࡞ΔՊֶͱٕज़ʯ w ैདྷͷίϯϐϡʔλϓϩάϥϜతͰͳ͔ͬͨ w ਓ͕ؒೳΛͬͯͰ͖Δͷͱಉ͜͡ͱΛ ίϯϐϡʔλϓϩάϥϜͰͰ͖ΔΑ͏ʹ͍ͨ͠
ਓೳʮతͳίϯϐϡʔλϓϩάϥϜʯ 4 IUUQTXXXBJHBLLBJPSKQXIBUTBJ"*GBRIUNM (*)
/39 ਓೳͱ w తͰͳ͍ίϯϐϡʔλϓϩάϥϜ ༩͑ΒΕͨϧʔϧͲ͓Γͷಈ࡞͔͠Ͱ͖ͳ͍ ίϯϐϡʔλʹѻ͑ΔܗࣜͰਓ͕ؒϧʔϧΛ࡞Δඞཁ͕͋Δ ෳࡶͳͰͯ͢ͷঢ়گΛཏͨ͠ϧʔϧΛ࡞Δ͜ͱࠔ w తͳίϯϐϡʔλϓϩάϥϜʢਓೳʣ ঢ়گʹԠͯ͡ͲͷΑ͏ʹಈ࡞͢Εྑ͍͔ΛྟػԠมʹஅ͢Δ
ܦݧڭࡐ͔ΒϧʔϧΛֶश͢Δ গͷϧʔϧ͔Βਪͯ͠அ͢Δ ਓೳঢ়گʹԠͯ͡ྟػԠมʹಈ࡞͢Δ 5
/39 w తͰͳ͍ίϯϐϡʔλϓϩάϥϜϧʔϧʹैͬͯೝࣝ͢Δ w ༷ʑͳචͷखॻ͖จࣈΛཏͨ͠ϧʔϧΛ࡞Δͷࠔ ਓೳͱ ྫɿखॻ͖ࣈจࣈΛೝࣝ͢ΔίϯϐϡʔλϓϩάϥϜ 6 खॻ͖ࣈจࣈͷग़యɿIUUQZBOOMFDVODPNFYECNOJTU
1 1 ʮࠨӈࡾͷҰͷྖҬ͕ന৭ͳΒʯ
/39 ਓೳڭࡐʢը૾ͱࣈͷϖΞʣΛͬͯೝࣝϧʔϧΛֶश͢Δ ਓೳͱ ྫɿखॻ͖ࣈจࣈΛೝࣝ͢ΔίϯϐϡʔλϓϩάϥϜ 7 1 1 9 9 7
7 ϧʔϧΛֶश 9 ϧʔϧΛར༻ ڭࡐ ʜ खॻ͖ࣈจࣈͷग़యɿIUUQZBOOMFDVODPNFYECNOJTU
/39 ৗͷதͷਓೳ w εϚʔτεϐʔΧʔɿ ਓؒͷԻࢦࣔʹै͍༷ʑͳಈ࡞Λߦ͏ w ϩϘοτআػɿ োΛආ͚ͳ͕Β෦શମΛআ͢Δ w إೝূɿ
Χϝϥલͷਓ͕ొ͞ΕͨਓͱಉҰਓ͔Ͳ͏͔Λఆ w ίϯςϯπɾਪનɿ Ӿཡཤྺߪങཤྺʹͱ͖ͮϢʔβ͕ΉΞΠςϜΛఏࣔ զʑͷৗͷ͞·͟·ͳ໘Ͱਓೳ͕׆༻͞Ε͍ͯΔ 8
/39 ਓؒΛ͑Δਓೳ w *#.8BUTPOɿ ΞϝϦΧͷΫΠζ൪ʮ+FPQBSEZʯʹઓɺৗ࿈ग़ԋऀΛഁΓ༏উ w ౦ϩϘ͘Μɿ େֶೖࢼࢼͷֶɾੈք࢙Ͱภࠩ͑Λୡ w %FFQ#MVFɿ
νΣεͷੈքνϟϯϐΦϯʹউར w "MQIB(Pɿ ғޟͷੈքτοϓع࢜ʹউར ΫΠζήʔϜͰਓؒΛ͑Δਓೳ͕ొ 9
/39 ݱࡏͷਓೳʹͰ͖Δ͜ͱɿಛԽܕWT൚༻ܕ w ਓೳʮಛԽܕʯͱʮ൚༻ܕʯͷೋछྨʹେผ͞ΕΔ w ಛԽܕਓೳɿ ͋Δಛఆͷঢ়گʹ͓͍ͯతʹ;Δ·͏ਓೳ w ൚༻ܕਓೳɿ
ਓؒͱಉ༷ʹ༷ʑͳঢ়گɾʹ͓͍ͯతͳ;Δ·͍͕Ͱ͖Δਓೳ w ݱࡏͷਓೳಛԽܕਓೳͰ͋Γ ൚༻ܕਓೳະ࣮ͩݱ͞Ε͍ͯͳ͍ ݱࡏͷਓೳಛఆͷ͜ͱ͚ͩͰ͖ΔಛԽܕਓೳ 10
/39 ݱࡏͷਓೳʹͰ͖Δ͜ͱɿը૾ॲཧ w ը૾ೝࣝɿ ࣸਅʹ͍ࣸͬͯΔͷΛೝࣝ͢Δ w ମݕग़ɿ ࣸਅͷͲ͜ʹԿ͕͍ࣸͬͯΔͷ͔Λೝࣝ͢Δ w ը૾ੜɿ
ຊͷࣸਅͷΑ͏ͳը૾Λਓతʹ࡞Γग़͢ ಛԽܕਓೳͰ༷ʑͳ͜ͱ͕࣮ݱͰ͖Δ 11 ग़యɿIUUQDTOTUBOGPSEFEVTMJEFTDTO@@MFDUVSFQEG ग़యɿIUUQTBSYJWPSHBCT ຊͷࣸਅ ਓతͳࣸਅ (*) (*)
/39 ݱࡏͷਓೳʹͰ͖Δ͜ͱɿݴޠॲཧ w จॻྨɿ จॻΛಛఆͷΧςΰϦʹྨ͢Δ w ػց༁ɿ ͋ΔݴޠͰॻ͔ΕͨจষΛผͷݴޠʹ༁͢Δ w ใݕࡧɿ
ಛఆͷใ͕ܝࡌ͞Ε͍ͯΔΣϒϖʔδΛݟ͚ͭΔ w ࣭Ԡɿ จষͰ༩͑ΒΕ࣭ͨʹର͢Δ͑Λฦ͢ ಛԽܕਓೳͰ༷ʑͳ͜ͱ͕࣮ݱͰ͖Δ 12
/39 ݱࡏͷਓೳʹͰ͖Δ͜ͱɿԻॲཧ w Իೝࣝɿ ͞Ε͍ͯΔ༰ΛจষͰॻ͖ى͜͢ w ऀಉఆɿ ొ͞Εͨਓͱಉ͡ਓ͕͍ͯ͠Δ͔ఆ͢Δ w Ի߹ɿ
ਓ͕͍ؒͯ͠ΔΑ͏ͳΛਓతʹ࡞Γग़͢ ಛԽܕਓೳͰ༷ʑͳ͜ͱ͕࣮ݱͰ͖Δ 13
/39 ݱࡏͷਓೳʹͰ͖Δ͜ͱɿϩϘοτ w ڥೝࣝɿ ηϯαใ͔ΒपғʹԿ͕͋Δ͔Λೝࣝ͢Δ w ܦ࿏ɾߦಈܭըɿ Ҡಈɾಈ࡞ࢦࣔʹै͏ͨΊʹͲͷΑ͏ͳܦ࿏Λ௨Δ͔ɺ Ͳͷ෦ΛͲͷΑ͏ʹಈ͔͔͢Λܭը͢Δ ಛԽܕਓೳͰ༷ʑͳ͜ͱ͕࣮ݱͰ͖Δ
14
/39 ਓೳͷ׆༻ྫ ྫᶃεϚʔτεϐʔΧʔ ಛԽܕਓೳͷ߹ͤͰ͞Βʹߴͳ͜ͱΛ࣮ݱͰ͖Δ 15 4UFQىಈΩʔϫʔυͷೝࣝ 4UFQԻʹΑΔࢦࣔΛจষͱͯ͠ೝࣝ 4UFQࢦࣔจ͔ΒతΛೝࣝ 4UFQࢦࣔΛ࣮ߦ
/39 ਓೳͷ׆༻ྫ ྫᶄࣗಈӡస ಛԽܕਓೳͷ߹ͤͰ͞Βʹߴͳ͜ͱΛ࣮ݱͰ͖Δ 16 ը૾ͷग़యɿIUUQTTUPSBHFHPPHMFBQJTDPNTEDQSPEWTBGFUZSFQPSUXBZNPTBGFUZSFQPSUQEG 4UFQଞͷं྆าߦऀͷݕ 4UFQଞͷं྆าߦऀͷߦಈ༧ଌ 4UFQత·Ͱͷܦ࿏Λܭը 4UFQܦ࿏ʹै͏ͨΊͷಈ࡞Λܭը
/39 ਓೳͷ·ͱΊ w ਓೳతͳίϯϐϡʔλϓϩάϥϜͰ ϧʔϧΛֶशɾਪͯ͠ྟػԠมʹಈ࡞Ͱ͖Δ w ͍·ͷਓೳಛԽܕͰ͋Δಛఆͷ͔͠ղ͚ͳ͍ w ༷ʑͳͰಛԽܕਓೳͷݚڀ͕ਐΊΒΕ ͦΕΒͷ߹ͤͰΑΓߴͳਓೳ͕࣮ݱ͞Εͭͭ͋Δ
͍·ͷਓೳಛԽܕ͕༷ͩʑͳ͜ͱ͕Ͱ͖Δ 17
ػցֶश
/39 ػցֶशͱ w ػցֶश ίϯϐϡʔλϓϩάϥϜʹσʔλ͔ΒϧʔϧΛֶशͤ͞Δٕज़ w େྔͷೖग़ྗྫͷσʔλΛڭࡐͱͯ͠༩͑ ೖྗͱग़ྗΛରԠ͚ͮΔϧʔϧΛֶशͤ͞Δ ೖྗͱग़ྗΛରԠ͚ͮΔϧʔϧΛֶश͢Δ
19 ຊߨٛͰऔΓ্͛Δػցֶशɺݫີʹʮڭࢣ͖ػցֶशʯͱݺΕΔ 7 ೖྗ ग़ྗ ը૾ ࣈ
/39 ػցֶशͱ ೖྗͱग़ྗΛରԠ͚ͮΔϧʔϧΛֶश͢Δ 20 The quick brown fox jumps over
the lazy dog ૉૣ͍৭ͷޅ ͷΖ·ͳݘΛඈͼ ӽ͑Δ “Hello, world!” ೖྗ ग़ྗ ݈߁ϦεΫɿߴ ग़యɿIUUQDTOTUBOGPSEFEVTMJEFTDTO@@MFDUVSFQEG ग़యɿIUUQTNBHFOUBUFOTPSqPXPSHOTZOUIGBTUHFO จষ จষ ը૾ ମͱҐஔ Ի จষ ױऀใ ݈߁ϦεΫ (*) (**)
/39 ػցֶशͱ w εϚʔτϑΥϯɾηϯαɾΠϯλʔωοτɾΣϒαʔϏεͷීٴͰ σʔλ͕૿େͨ͠ͷ͕ػցֶशͷൃలͷ͖͔͚ͬ w ιʔγϟϧϝσΟΞͷߘ ΣϒαʔϏε্Ͱͷਓʑͷߦಈϩάػցֶशͷڭࡐͱͯ͠༻͍ΒΕΔ w ࠂ৴ͳͲͰΣϒαʔϏεӡӦاۀརӹΛ্͛ɺ
ແྉͰΣϒαʔϏεΛఏڙ͠ͳ͕Βڭࡐ༻ͷσʔλΛऩू͍ͯ͠Δ ͋ΒΏΔͷ͕σʔλԽ͞ΕػցֶशͰΘΕΔ 21
/39 ػցֶशͷ͘͠Έɿ܇࿅σʔλ w ྫɿʮजᙾ͕ྑੑ͔ѱੑ͔ʯఆ͢ΔͨΊͷϧʔϧΛֶशͤ͞Δ w ڭࡐͱͯ͠༻͍ΔσʔλΛ܇࿅σʔλͱݺͿ w ֶशͨ͠ϧʔϧΛ༻͍ͯ ܇࿅σʔλʹͳ͍जᙾ͕ྑੑ͔ѱੑ͔Λਖ਼͘͠ఆͰ͖ΔΑ͏ʹ͍ͨ͠ ೖग़ྗྫΛ܇࿅σʔλͱͯ͠༻͍ͯϧʔϧΛֶश͢Δ
22 जᙾը૾ͷग़యɿW. N. Street et al. , Nuclear feature extraction for breast tumor diagnosis. Proc. of the International Symposium on Electronic Imaging, 1993. ྑੑ ѱੑ ྑੑ ѱੑ ܇࿅σʔλ ೖྗʢजᙾʣΛ ग़ྗʢྑੑPSѱੑʣʹ ରԠ͚ͮΔ ϧʔϧ ֶश
/39 ػցֶशͷ͘͠Έɿ܇࿅σʔλ w ίϯϐϡʔλͰܭࢉͰ͖ΔΑ͏ʹ܇࿅σʔλͰදݱ͞Ε͍ͯΔ w ྫɿʮେ͖͞ʯͱʮͰ͜΅͜ʯͰजᙾΛදݱ͢Δ ʢԿΒ͔ͷखஈͰԽ͞Ε͍ͯΔͱ͢Δʣ ܇࿅σʔλͰදݱ͞Εͳ͍ͱ͍͚ͳ͍ 23 ܇࿅σʔλͷྫ
ೖྗ ग़ྗ େ͖͞ Ͱ͜΅͜ जᙾ 0.252 0.014 ྑੑ जᙾ 0.327 0.340 ྑੑ जᙾ 1.000 0.371 ѱੑ जᙾ 0.223 0.369 ѱੑ ʜ ʜ ʜ ʜ
/39 ػցֶशͷ͘͠Έɿ܇࿅σʔλ ܇࿅σʔλͰදݱ͞Εͳ͍ͱ͍͚ͳ͍ 24 211 220 223 … 213 217
217 220 … 210 214 217 216 … 205 … … … … … 216 217 214 … 181 The quick brown fox jumps over the lazy dog aardvark … brown … dog … zzzat 0 … 1 … 1 … 0
/39 ػցֶशͷ͘͠Έɿ܇࿅σʔλ ܇࿅σʔλͰදݱ͞Εͳ͍ͱ͍͚ͳ͍ 25 ܇࿅σʔλͷྫ ೖྗ ग़ྗ େ͖͞ Ͱ͜΅͜ जᙾ
0.252 0.014 ྑੑ जᙾ 0.327 0.340 ྑੑ जᙾ 1.000 0.371 ѱੑ जᙾ 0.223 0.369 ѱੑ ʜ ʜ ʜ ʜ ਤࣔ େ͖͞ Ͱ͜΅͜ ྑੑͷजᙾ ѱੑͷजᙾ जᙾ जᙾ
/39 ػցֶशͷ͘͠Έɿϧʔϧ w ίϯϐϡʔλ͕ܭࢉͰ͖ΔΑ͏ʹϧʔϧࣜͰදݱ͞ΕΔ w ୯७ͳϧʔϧͷܗࣜɿ ϧʔϧࣜͰදݱ͞Εͳ͍ͱ͍͚ͳ͍ 26 େ͖͞
Ͱ͜΅͜ ͳΒྑੑɺͦΕҎ֎ͳΒѱੑ z = w1 × + w2 × + b; z < 0
/39 ػցֶशͷ͘͠Έɿϧʔϧ ϧʔϧࣜͰදݱ͞Εͳ͍ͱ͍͚ͳ͍ 27 େ͖͞ Ͱ͜΅͜ ྑੑͷजᙾ ѱੑͷजᙾ [ʾͷྖҬ [ͷྖҬ
ͷ߹ɿ େ͖͞ Ͱ͜΅͜ w1 = 1, w2 = − 1, b = 0 z = − େ͖͞ Ͱ͜΅͜ z = w1 × + w2 × + b;
/39 ػցֶशͷ͘͠Έɿϧʔϧ X X Cͷ͕มΘΔͱϧʔϧมΘΔ 28 େ͖͞ Ͱ͜΅͜ ྑੑͷजᙾ ѱੑͷजᙾ
[ʾͷྖҬ [ͷྖҬ ͷ߹ɿ େ͖͞ Ͱ͜΅͜ w1 = 1, w2 = 1, b = − 1 z = + −1 େ͖͞ Ͱ͜΅͜ z = w1 × + w2 × + b;
/39 ػցֶशͷ͘͠Έɿϧʔϧ ܇࿅σʔλʹͯ·Δྑ͍ϧʔϧΛֶश͍ͨ͠ 29 େ͖͞ Ͱ͜΅͜ େ͖͞ Ͱ͜΅͜ w1
= 1, w2 = − 1, b = 0 w1 = 1, w2 = 1, b = − 1 ϧʔϧ͕ͯ·Βͳ͍जᙾ ӈͷϧʔϧͷํ͕ɺ܇࿅σʔλʹͯ·Δ
/39 ػցֶशͷ͘͠Έɿଛࣦ ϧʔϧͷྑ͠ѱ͠ΛͰද͢ 30 ଛࣦɿେ ଛࣦɿখ େ͖͞ Ͱ͜΅͜ େ͖͞ Ͱ͜΅͜
w ଛࣦɿϧʔϧͷѱ͞ΛͰදͨ͠ͷ w ܇࿅σʔλʹͯ·Βͳ͍ϧʔϧଛࣦ͕େ͖͘ͳΔ
/39 ػցֶशͷ͘͠Έɿଛࣦ w ʮޡఆͷ݅ʯΛଛࣦͱͯ͠͏ͱϧʔϧͷࡉ͔͍ҧ͍͕Θ͔Βͳ͍ w ϧʔϧʹʮѱੑͷ֬ʯΛग़ྗͤ͞ɺͦΕʹͱ͖ͮଛࣦΛࢉग़ ϧʔϧͷྑ͠ѱ͠ΛͰද͢ 31 ૯Λϧʔϧͷଛࣦͱ͢Δ ଛࣦɿ
܇࿅σʔλ [ ѱੑͷ֬ ଛࣦ ೖྗ ग़ྗ େ͖͞ Ͱ͜΅͜ जᙾ 0.252 0.014 ྑੑ -0.542 0.336 0.410 जᙾ 0.327 0.340 ྑੑ -0.172 0.457 0.611 जᙾ 1.000 0.371 ѱੑ 1.208 0.770 0.261 जᙾ 0.223 0.369 ѱੑ -0.348 0.414 0.882 ʜ ʜ ʜ ʜ a = σ(z)
/39 ػցֶशͷ͘͠Έɿޯ๏ w ଛࣦ͕࠷খͷϧʔϧɺ ͭ·Γ܇࿅σʔλʹ࠷ͯ·ΔϧʔϧΛݟ͚ͭΔͷֶ͕श w ޮతʹͦͷΑ͏ͳϧʔϧΛݟ͚ͭΔͨΊʹޯ๏͕༻͍ΒΕΔ w ޯ๏Ͱɺଛࣦ͕࠷খ͘͞ͳΔํʹগ͚ͩ͠ਐΉ ʢX
X CͷΛগ͚ͩ͠มԽͤ͞Δʣ͜ͱΛ܁Γฦ͢ ଛࣦ͕࠷খͱͳΔϧʔϧʢX X CͷʣΛݟ͚ͭΔ 32
/39 ػցֶशͷ͘͠Έɿ൚Խೳྗ w ֶशͷͦͦͷత ʮ܇࿅σʔλʹͳ͍ະͷೖྗʹରͯ͠ਖ਼͍͠ग़ྗΛฦ͢ʯϧʔϧͷֶश w ܇࿅σʔλͷͯ·Γ͚ͩΛߟྀ͢Δͱ ϧʔϧ͕܇࿅σʔλʹͯ·Γ͗ͯ͢ະͷೖྗͰؒҧ͑ΔڪΕ w X
Xͷ͕ۃʹେ͖͍ϧʔϧաద߹͍ͯ͠ΔڪΕ͕͋ΔͷͰ ͦͷΑ͏ͳϧʔϧʹϖφϧςΟΛ༩͑ΔΑ͏ʹଛࣦΛઃܭ͢Δ w ʮϧʔϧ͕ະͷೖྗʹର͠ਖ਼͍͠ग़ྗΛฦ͢ೳྗʯΛ ൚Խೳྗͱ͍͏ ܇࿅σʔλʹͯ·Γ͗͢Δϧʔϧ൚Խೳྗ͕͍ 33
/39 w ઢͰදݱ͞ΕΔϧʔϧೖྗͱग़ྗͷରԠ͚͕ͮෳࡶͳ߹ʹෆे w ਂֶशΛ༻͍ΔͱෳࡶͳϧʔϧΛֶशͰ͖Δ େ͖͞ Ͱ͜΅͜ େ͖͞ Ͱ͜΅͜ ୯७ͳϧʔϧ
ػցֶशͷ͘͠Έɿਂֶश ෳࡶͳϧʔϧͷֶशʹ༻͍ΒΕΔͷ͕ਂֶश 34 େ͖͞ Ͱ͜΅͜ ਂֶशʹΑΔϧʔϧ
/39 w ୯७ͳϧʔϧͷࣜਤɿ ػցֶशͷ͘͠Έɿਂֶश ਂֶशͰෳͷඇઢܗؔΛଟʹॏͶΔ 35 େ͖͞ Ͱ͜΅͜ ྑੑ·ͨѱੑ w1
w2 a=σ(z) େ͖͞ Ͱ͜΅͜ ྑੑ·ͨѱੑ w ਂֶशʹΑΔϧʔϧͷࣜਤɿ
/39 ػցֶशͷ࣮ w ػցֶशͷ࣮ʹɺϓϩάϥϛϯάݴޠ1ZUIPO͕͘ΘΕ͍ͯΔ w ྫ͑TDJLJUMFBSOͱ͍͏1ZUIPOϥΠϒϥϦΛ͏ͱɺ ߦͷίʔυͰػցֶशΛ࣮͢Δ͜ͱ͕Ͱ͖Δ w ͍͔ͭ͘ͷαϯϓϧσʔλTDJLJUMFBSOͰఏڙ͞Ε͍ͯΔ ༷ʑͳϥΠϒϥϦ͕ެ։͞Ε͓ͯΓ؆୯ʹ࣮͕Ͱ͖Δ
36 https://scikit-learn.org/ ϧʔϧͷֶश ϧʔϧͷద༻
/39 ػցֶशͷ࣮ w ਂֶशͷ1ZUIPOϥΠϒϥϦ༷ʑͳͷ͕ఏڙ͞Ε͍ͯΔɿ 5FOTPS'MPX ,FSBT 1Z5PSDI $IBJOFS w (PPHMF$PMBCPSBUPSZΛ͏ͱԾϚγϯΛͬͯ
ਂֶशϥΠϒϥϦΛΣϒϒϥβ͔Β؆୯ʹ͏͜ͱ͕Ͱ͖Δ ༷ʑͳϥΠϒϥϦ͕ެ։͞Ε͓ͯΓ؆୯ʹ࣮͕Ͱ͖Δ 37 https://colab.research.google.com
/39 ػցֶशͷ࣮ w ,BHHMFͳͲͷػցֶशίϯϖςΟγϣϯͰ ࣮ફతͳػցֶशͷ՝ɾσʔλ͕ఏڙ͞Εଞऀͱڝ͏͜ͱ͕Ͱ͖Δ ػցֶशίϯϖςΟγϣϯͰྗࢼ͠Λ͢Δ͜ͱ͕Ͱ͖Δ 38 https://www.kaggle.com/c/titanic/leaderboard
/39 ػցֶशͷ·ͱΊ w େྔͷೖग़ྗྫͷσʔλΛ܇࿅σʔλͱ͍ͯ͠ ೖྗͱग़ྗΛରԠ͚ͮΔϧʔϧΛֶश w ܇࿅σʔλʹͰ͖Δ͚ͩͯ·ΔΑ͏ͳϧʔϧ ʢଛࣦͷখ͍͞ϧʔϧʣΛޯ๏Λ༻͍ͯݟ͚ͭΔ ܇࿅σʔλΛ༻͍ͯೖྗͱग़ྗΛରԠ͚ͮΔϧʔϧΛֶश 39