Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Sign up for free
Menu
Search
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Pricing
Search
Sign in
Sign up for free
公平性に配慮した学習とその理論的課題
Search
Kazuto Fukuchi
November 06, 2018
Research
110
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
公平性に配慮した学習とその理論的課題
第21回情報論的学習理論ワークショップ, 2018.11.4〜7, 札幌(かでる2.7・北大)の企画セッション:学習理論 で発表した講演のスライドです.
Kazuto Fukuchi
November 06, 2018
More Decks by Kazuto Fukuchi
See All by Kazuto Fukuchi
Harnessing the Power of Vicinity-Informed Analysis for Classification under Covariate Shift
nanofi
3
580
機械学習アルゴリズムに潜む不公平なバイアスとその理論
nanofi
0
120
公平性を保証したAI/機械学習アルゴリズムの最新理論
nanofi
0
120
Other Decks in Research
See All in Research
ros2-perf-multihost: 分散システムにおける客観的なアーキテクチャ評価フレームワーク
takasehideki
0
290
VRID: View-Invariant Representation through Dual-Axis Transformation for Cross-iew Pose Estimation
satai
3
120
Evaluating LLM Reliability Across Facts, Evidence, and Cultures
yukiar
0
200
Détecter les vulnérabilités dans un environnement souverain
hellosct1
0
110
La génomique au service de la fromageabilité du lait grâce aux spectres MIR
institutdelelevage
PRO
0
180
Anthropic が提案する LLM の内部状態を自然言語で説明可能にした Natural Language Autoencoders / Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
shunk031
0
330
PHTalks Bengaluru - SSRF When All Else Fails
dk999
0
1.2k
Fukui Shibiten 39 - AI Art
butchi
0
260
CVPR2026論文紹介_VLMにとって良いvision encoderとは何か?Rethinking Model Selection in VLM Through the Lens of Gromov-Wasserstein Distance
kobayashi31
1
260
就労支援の現場での生成AI活用
enochan
0
130
Adapter les bergeries aux fortes chaleurs
institutdelelevage
PRO
0
130
データサイエンティストの就労意識~2015 → 2026 一般(個人)会員アンケートより
datascientistsociety
PRO
0
1k
Featured
See All Featured
HU Berlin: Industrial-Strength Natural Language Processing with spaCy and Prodigy
inesmontani
PRO
0
750
Principles of Awesome APIs and How to Build Them.
keavy
128
18k
Accessibility Awareness
sabderemane
1
240
Producing Creativity
orderedlist
PRO
348
41k
Bash Introduction
62gerente
615
220k
Art, The Web, and Tiny UX
lynnandtonic
304
22k
So, you think you're a good person
axbom
PRO
2
2.2k
Evolving SEO for Evolving Search Engines
ryanjones
0
320
Intergalactic Javascript Robots from Outer Space
tanoku
273
27k
Game over? The fight for quality and originality in the time of robots
wayneb77
1
300
How to build a perfect <img>
jonoalderson
1
6.1k
Six Lessons from altMBA
skipperchong
29
4.5k
Transcript
ެฏੑʹֶྀͨ͠शͱͦͷ ཧత՝ *#*4 اըηογϣϯֶशཧ ɹҰే ཧݚ"*1 LB[VUPGVLVDIJ!SJLFOKQ 1
֓ཁ ػցֶशʹ͓͚Δެฏੑ 'BJSOFTT w ެฏੑ͕͞Ε͍ͯΔ !2 '"5.- ެฏੑͷϫʔΫγϣοϓ JO/*14
*$.- ,%% "$.'"5 ެ ฏ ੑ ͷ ࠃ ࡍ ձ ٞ *OWJUFEUBMLT w*$.- -4XFFOFZ w/*14 ,$SBXGPSE w,%% $%XPSL w,%% +.8JOH
֓ཁ w ຊߨԋͷ༰ w ެฏੑʹ·ͭΘΔٙ w ެฏੑͱʁ w ՝
Կ͕Ͱ͖Δͱخ͍͠ͷ͔ʁ w ಛʹཧతղੳʹ͓͚Δ՝Λத৺ʹհ !3
࣍ w ެฏੑͱ w ެฏੑͷݪҼͱެฏੑఆٛ w ूஂެฏੑ w ूஂެฏੑͷཧత՝
w ݸਓެฏੑ w ݸਓެฏੑͷཧత՝ w ࠷ۙͷల։ͱͦͷཧత՝ !4
ެฏੑ 'BJSOFTT w ػցֶश͕ҙࢥܾఆʹ༻͍ΒΕ͍ͯΔ w ҙࢥܾఆ͕ࠩผΛੜΉՄೳੑ͕͋Δ !5 ࠾༻ ৴༻είΞ ೖࢼ
อݥྉ ݸਓ உੑ ঁੑ
ࠩผతࠂ<4XFFOZ> w lHPPHMFDPNͱSFVUFSTDPNͰਓ໊ͷݕࡧͰදࣔ͞ΕΔࠂΛूܭ w ظ݄݄ؒ w ݸͷࠂΛऩू ΞϑϦΧܥͷ໊લ Ϥʔϩούܥͷ໊લ ωΨςΟϒͳࠂ
தཱతͳࠂ "SSFTUFE -PDBUFE !6
ࠩผతࠂ<4XFFOZ> w ͷࠂ͕lJOTUBOUDIFDLNBUFz ൜ࡑྺݕࡧαΠτ w JOTUBOUDIFDLNBUFͷࠂͷ༰ͱਓछͷಠཱੑΛݕఆ ਓछʹґଘͯ͠ࠂ༰͕ωΨςΟϒʹͳΔ͔ܾ·Δ TJHOJpDBODF Ͱ༏Ґʹैଐ
TJHOJpDBODF Ͱ༏Ґʹैଐ !7
.BDIJOF#JBT<"OHXJO > w $0.1"4ΞϧΰϦζϜ w नਓͷྦྷ൜ʹର͢ΔείΞΛ༩͑ͯ͘ΕΔ w είΞ͕ߴ͍΄Ͳྦྷ൜ϦεΫ͕ߴ͍ !8 ϦεΫ͍͕
ճͷྦྷ൜ ϦεΫߴ͍͕ ྦྷ൜͍ͯ͠ͳ͍
.BDIJOF#JBT<"OHXJO > !9 είΞͷਖ਼֬ੑͱਓछͷؔੑ w ޡͬͯࠇਓͷείΞΛߴ͘ਪఆ w ޡͬͯനਓͷείΞΛ͘ਪఆ നਓ͕༏۰͞ΕͨείΞ͕͞Ε͍ͯΔ ޡͬͯϦεΫ͕͍ͱਪఆ
ޡͬͯϦεΫ͕ߴ͍ͱਪఆ
ػց༁ʹ͓͚Δࠩผ w (PPHMF༁ w ਓশͷੑࠩͷͳ͍τϧίޠ͔Βӳޠͷ༁ w l൴൴ঁ<৬ۀ>Ͱ͋Δɽzͱ͍͏จΛ<৬ۀ>Λม͑ͯ༁ !10 (PPHMF༁IUUQTUSBOTMBUFHPPHMFDPN &㵽BSCBL`TGBDFCPPLQPTU
IUUQTXXXGBDFCPPLDPNQIPUPQIQ GCJETFUB UZQFUIFBUFS
ެฏੑͷࣾձతཁٻ w ػցֶशͷެฏੑࣾձ͔Βͷཁٻڧ͍ w #JH%BUBʹؔ͢Δ8IJUF)PVTF3FQPSU w 8IJUF)PVTF3FQPSU<1PEFTUB > lΞϧΰϦζϜʹΑΔજࡏతͳࠩผΛࢹ͠ͳ͚Ε ͳΒͳ͍z
w ಉ༷ͳ༰͕8IJUF)PVTF3FQPSU<.VOP[ > !11
ެฏੑͷ๏తཁٻ w 5JUMF7** w ਓछɼഽͷ৭ɼफڭɼੑผɼग़ࠃʹΑΔޏ༻ࠩผͷېࢭ w உঁޏ༻ػձۉ๏ w ৬ʹ͓͚Δஉঁࠩผͷېࢭ w
(%13 w "SUJDMFݸਓσʔλॲཧʹؔ͢Δنఆ w lద๏ɺެฏ͔ͭಁ໌ੑͷ͋ΔखஈͰॲཧ͠ͳ͚ΕͳΒͳ͍z !12 ޏ༻ʹػցֶशΛ͏߹ରॲ͕ඞཁෆՄܽ (%13ͲΜͳλεΫͰެฏੑ͕ཁٻ͞ΕΔՄೳੑ͕͋Δ
ෆެฏͷݪҼ ͳͥػցֶश͕ෆެฏͳग़ྗΛ͢Δͷ͔ w σʔλऩू͔Βֶशͷաఔʹ͓͍ͯόΠΞε͕Δ͜ͱ͕ݪҼ w ༷ʑͳόΠΞε͕ͷΔݪҼ͕ٞ͞Ε͍ͯΔ<#BSPDBT > w େ͖ͭ͘ʹΘ͚Δ
!13 ෆެฏ σʔλऩूʹ͓͚Δ όΠΞε w ࠩผతϥϕϧ͚ w αϯϓϦϯάόΠΞε ֶशʹ͓͚Δ όΠΞε w ֶशϞσϧͷઃܭ w গάϧʔϓͷແࢹ σʔλ ֶश
σʔλऩूʹ͓͚ΔόΠΞε w ֶशऀ͕ಘΒΕΔαϯϓϧ͕ෆެฏ !14 ࠩผతͳϥϕϧ wϥϕϧਓͷखʹΑ͚ͬͯΒΕΔ wྫ ޏ༻ͷ࠾൱աڈਓ͕அͨ͠ wϥϕϧ͚ʹࠩผతͳࢥ͍ࠐΈΛө͢ΔՄೳੑ͋Γ wҙࣝతແҙࣝతʹؔΘΒͣ
σʔλऩूʹ͓͚ΔόΠΞε w ֶशऀ͕ಘΒΕΔαϯϓϧ͕ෆެฏ !15 αϯϓϦϯάόΠΞε wภͬͨαϯϓϧ͔͠ಘΒΕͳ͍͕࣌͋Δ wྫ ͓ۚͷିΛͨ͠ਓ͕࠴ෆཤߦʹؕΔ͔Ͳ͏͔Θ͔Βͳ͍ wաڈ͓ۚΛିͨ͠ਓΛࠩผతʹબ͍ͯ͠ΔՄೳੑ͋Γ
"EVMUEBUB<$BMEFST > w 64ͷશ݅ௐࠪσʔλ w ݸਓͷऩೖ͕LҎ্͔ҎԼ͔༧ଌ͢Δ !16 .BMF 'FNBMF )JHIJODPNF
-PXJODPNF ͕ ߴऩೖ ͕ ߴऩೖ σʔλʹஉঁؒͷόΠΞε͕͋Δ
ֶशʹ͓͚ΔόΠΞε w ֶशʹΑͬͯෆެฏͳྨث͕ಘΒΕΔ !17 গάϧʔϓͷແࢹ wσʔλͷஉঁ͕ͻͲ͘ภ͍ͬͯΔͱ͢Δ wྫ ΄ͱΜͲͷσʔλ͕உੑͷͷͰঁੑͷσʔλ͕΄ͱΜͲͳ͍ w༧ଌੑೳΛ্͛ΔͨΊগΛϊΠζͱΈͳ͢Մೳੑ͕͋Δ
"EVMUEBUB<$BMEFST > w /BJWF#BZFTͰֶश͠ྨ͢Δ !18 .BMF 'FNBMF )JHIJODPNF
-PXJODPNF ͕ ߴऩೖ ͕ ߴऩೖ ΑΓࠩผతͳ༧ଌ݁ՌʹͳΔ
ֶशʹ͓͚ΔόΠΞε w ֶशʹΑͬͯෆެฏͳྨث͕ಘΒΕΔ !19 ֶशͷ༧ଌϞσϧͷઃܭʹΑΔࠩผ wػցֶशͰϞσϧઃܭΛͦ͠ͷޙϞσϧͷύϥϝʔλΛσʔλ͔ΒಘΔ wϞσϧઃܭͷํʹΑͬͯࠩผΛট͘ wྫ உঁͷΛͬͯ༧ଌΛߦ͏ϞσϧΛઃܭ wࠩผ
EJTQBSBUFUSFBUNFOU CMJOEOFTT Ϟσϧઃܭ ͳΜͷಛྔΛ͏͔ ༧ଌؔ ઢܗɼଟ߲ࣜɼ3),4ɼ%FFQ//
3FEMJOJOHF⒎FDU<$BMEFST > wஉঁͷΛΘͳͯࠩ͘ผ͕ى͜Δ w உঁਓछͳͲͱڧ͘ґଘͨ͠σʔλ Λ͏͜ͱͰؒతʹࠩผ͕ൃੜ w ྫ ֶྺΛͬͯ࠾൱ΛܾΊΔͱ ੑ͕ࠩੜ·ΕΔՄೳੑ͋Γ
w ྫ ॅॴΛͬͯ࠾൱ΛܾΊΔͱ ਓछͷ͕ࠩੜ·ΕΔՄೳੑ͋Γ !20 'SPNXJLJQFEJB
"EVMUEBUB<$BMEFST > w ੑผΛऔΓআ͍ͯ࠶/BJWF#BZFTͰֶश͠ྨ͢Δ !21 .BMF 'FNBMF )JHIJODPNF
-PXJODPNF ͕ ߴऩೖ ͕ ߴऩೖ ΑΓࠩผతͳ༧ଌ݁ՌʹͳΔ
ෆެฏͷݪҼ w ̎ͭͷঢ়گઃఆ σʔλ ֶशͰόΠΞε͕ͷΔՄೳੑ͕͋Δ ֶशͰόΠΞε͕ͷΔՄೳੑ͕͋Δ !22 ෆެฏ
σʔλऩूʹ͓͚Δ όΠΞε w ࠩผతϥϕϧ͚ w αϯϓϦϯάόΠΞε ֶशʹ͓͚Δ όΠΞε w ֶशϞσϧͷઃܭ w গάϧʔϓͷແࢹ σʔλ ֶश
ެฏੑఆٛ ूஂެฏੑ (SPVQGBJSOFTT w ηϯγςΟϒଐੑʹΑΔάϧʔϓؒͰ ͷࠩҟ !23 ݸਓެฏੑ *OEJWJEVBMGBJSOFTT
w ݸਓؒͰͷࠩҟ ࠾༻ ඇ࠾༻ ࠾༻ ඇ࠾༻ உੑ ঁੑ ࠾༻ ඇ࠾༻ ࠾༻ ඇ࠾༻ ≈ ≈ ⟹ =
ूஂݸਓ σʔλֶश !24 όΠΞε JOσʔλ ֶश όΠΞεJOֶश ूஂެฏੑ
ݸਓެฏੑ
ूஂݸਓ σʔλֶश !25 όΠΞε JOσʔλ ֶश όΠΞεJOֶश ूஂެฏੑ
ݸਓެฏੑ
ઃఆ w ؆୯ͷͨΊʹڭࢣ͋ΓྨͷΈΛߟ͑Δ w ɹɹɹɹɹɹɹɹɹɹֶྺɼ৬ྺɼࢿ֨ͳͲ w ɹɹɹɹɹɹɹɹɹɹੑผɼਓछɼफڭɼ࣏ࢤɼྸͳͲ w ɹɹɹɹɹɹɹɹɹɹ༧ଌ͍ͨ͠ͷ FH
࠾൱ w ɹɹɹɹɹɹɹɹɹɹΞϧΰϦζϜʹΑͬͯ༧ଌ͞Εͨϥϕϧ ೖྗ X ϥϕϧ Y ༧ଌϥϕϧ ̂ Y !26 ผͷೖྗ X S = உੑ S = ঁੑ ೖྗ X ϥϕϧ Y ηϯγςΟϒଐੑ S ༧ଌϥϕϧ ̂ Y ֶश
ूஂݸਓ σʔλֶश !27 όΠΞε JOσʔλ ֶश όΠΞεJOֶश ूஂެฏੑ
ݸਓެฏੑ
%FNPHSBQIJDQBSJUZ w ηϯγςΟϒଐੑͰ͚݅ͮΒΕͨ༧ଌϥϕϧͷ͕Ұக w Ͱͳ͘༧ଌਫ਼ِཅੑِӄੑͷҰகΛࢦ͢ͷ͋Γ !28 %FNPHSBQIJDQBSJUZ ℙ{ ̂ Y
∈ 𝒜|S = s} = ℙ{ ̂ Y ∈ 𝒜|S = s′} ҙͷ𝒜, s, s′ʹ͍ͭͯ ࠾༻ ඇ࠾༻ ࠾༻ ඇ࠾༻ உੑ ঁੑ = ̂ Y|S = உੑ ̂ Y|S = ঁੑ
%FNPHSBQIJDQBSJUZ w σʔλʹόΠΞε͕ͷ͍ͬͯΔՄೳੑ͕͋ΔͨΊ ϥϕϧɹͱ༧ଌϥϕϧɹҟͳΔ͖ w ϥϕϧͱҧ͏༧ଌΛ༩͑Δͱ༧ଌੑೳ͕Լ͕Δ ༧ଌੑೳͱެฏੑͷτϨʔυΦϑͷޮԽ͕త !29 %FNPHSBQIJDQBSJUZ ℙ{
̂ Y ∈ 𝒜|S = s} = ℙ{ ̂ Y ∈ 𝒜|S = s′} ҙͷ𝒜, s, s′ʹ͍ͭͯ Y ̂ Y
ूஂݸਓ σʔλֶश !30 όΠΞε JOσʔλ ֶश όΠΞεJOֶश ूஂެฏੑ
ݸਓެฏੑ
&RVBMJ[FEPEET<)BSEU > w ϥϕϧɹʹΑΔ༧ଌϥϕϧมԽࠩผΛੜ͡ͳ͍ ͱ৴͍ͯ͡Δ w ແཧΓ%1Λอো͠Α͏ͱ͢Δͱٯࠩผ !31 &RVBMJ[FEPEET
ℙ{ ̂ Y ∈ 𝒜|Y = y, S = s} = ℙ{ ̂ Y ∈ 𝒜|Y = y, S = s′} ҙͷ𝒜, y, s, s′ʹ͍ͭͯ Y உੑ உੑ ঁੑ ঁੑ σʔλ ༧ଌ ϥϕϧ %1ͷอূͷͨΊஉੑΛඇ࠾༻ʹ͢Δ ࠾༻͢ΔΑ͏ʹͨ͠ঁੑΑΓ ඇ࠾༻ʹͨ͠உੑͷํ͕ೳྗ͕ߴ͍
&RVBMJ[FEPEET<)BSEU > w ɹͱɹΛҰகͤ͞ΔΑ͏ʹֶशͰ͖Δ w %FNPHSBQIJDQBSJUZͰͰ͖ͳ͍ w ཧతʹτϨʔυΦϑͳ͍ গάϧʔϓແࢹͷࢭ͕తɹ !32
&RVBMJ[FEPEET ℙ{ ̂ Y ∈ 𝒜|Y = y, S = s} = ℙ{ ̂ Y ∈ 𝒜|Y = y, S = s′} ҙͷ𝒜, y, s, s′ʹ͍ͭͯ Y ̂ Y
%1WT&0 %1 w 1SPTόΠΞε͕ͷͬͨσʔλʹରԠ͍ͯ͠Δ w $POTٯࠩผͷ͋Γ &0 w 1SPT %1ʹൺͯ
༧ଌੑೳ͕Α͍ɼٯࠩผى͖ͳ͍ w $POTෆެฏͳόΠΞε͕ͬͨσʔλʹ͑ͳ͍ ͲͪΒಉ࣌ʹୡ͢Δ͜ͱෆՄೳ !33
ूஂެฏੑͷཧత՝ ൚Խతͳެฏੑͷอূ w ֶश࣌ʹςετ࣌ͷެฏੑͷอূ w طଘ݁Ռ w ૬ؔΛجʹͨ͠%1ͷެฏੑࢦඪͷҰ༷ऩଋ<'VLVDIJ > w
Ϋϥεͷେ͖͞ʹґଘ͠ͳ͍&0ͷެฏੑࢦඪͷऩଋ<8PPEXPSUI > w ޙॲཧܕͷΞϧΰϦζϜʹ͓͚Δ%1·ͨ&0ͷ൚Խެฏੑࢦඪόϯυ <"HBSXBM > w Ϋϥεͷେ͖͞ʹґଘ͠ͳ͍%1ͷެฏੑࢦඪͷऩଋ<$PUUFS > !34 ֶशσʔλͷ༧ଌ݁Ռ ςετσʔλͷ༧ଌ݁Ռ ެฏʹͳΔ Α͏ʹֶश ͢Δ ެฏ
ूஂެฏੑͷཧత՝ w ී௨ͷ൚Խόϯυ༧ଌϞσϧͷෳࡶ͞Λ༻͍Δ w ެฏੑͷ൚Խόϯυ༧ଌϞσϧͷෳࡶ͞ʹґଘ͠ͳ͘Ͱ͖Δ <8PPEXPSUI $PUUFS > w
ΘΓʹϥϕϧͷछྨºηϯγςΟϒଐੑͷͷछྨʹґଘ !35 ཧతʹ࠷దੑΛֶͬͨशํ๏ͷൃݟ w ूஂެฏੑʹ͓͚Δ࠷దੑͷղੳ·ͩͳ͍
ूஂެฏੑͷཧత՝ ֶशΞϧΰϦζϜͰ͋Δ࠷దԽͷ࠷దԽΞϧΰϦζϜ w ެฏੑͷ੍Λݴ͑ΕΔͱඇತͳ࠷దԽ͕ݱΕΔ w ۙࣅ ࠷దղΛಘΒΕΔ͜ͱΛอূ͢Δඞཁ͋Γ w طଘݚڀ w
ճؼ ࿈ଓηϯγςΟϒଐੑʹద༻Մೳͳඇತ࠷దԽʹΑͬͯఆࣜԽ͞ΕΔ ΞϧΰϦζϜͷ࠷దอূͷ͍ͭͨ࠷దԽΞϧΰϦζϜ<,PNJZBNB > w ͋ΔछͷΦϥΫϧͷଘࡏͷԾఆͷͱۙࣅ࠷దղΛଟ߲ࣜ࣌ؒͰٻΊΒΕΔ ࠷దԽΞϧΰϦζϜ<"MBCJ > !36
ूஂݸਓ σʔλֶश !37 όΠΞε JOσʔλ ֶश όΠΞεJOֶश ूஂެฏੑ
ݸਓެฏੑ
*OEJWJEVBMGBJSOFTT<%XPSL> w ੑผҎ֎શ͘ಉ͡ਓ͕͍Ε࠾൱ಉ͡ʹ͢Δ͖ w ࣅͨΑ͏ͳਓࣅͨ݁ՌΛड͚औΔ͖ w ֬త༧ଌؔ w !38 -JQTDIJU[QSPQFSUZ
ҙͷx, x′ʹ͍ͭͯ D( f(x), f(x′)) ≤ d(x, x′) ≈ ⟹ f : 𝒳 → Δ(𝒴) ݁Ռͷؒͷ ڑ
ݸਓެฏੑͷཧత՝ ൚Խతͳެฏੑͷอূ w ݸਓެฏੑʹؔͯ͠൚Խతͳੑೳͷղੳ͕ඞཁ w طଘ݁Ռ w σʔλ ֶशʹ͓͚ΔόΠΞεΛআڈ͍ͨ͠ઃఆͷͱ 1"$MFBSOJOHͷΈͰ1"ͳެฏੑͷ੍Լʹ͓͚Δαϯϓϧෳࡶͷղੳ
<3PUICMVN > !39 ࠷దੑΛֶͬͨशํ๏ͷൃݟେ͖ͳ՝
'BJSCBOEJU<+PTFQI > ࠷దੑͷূ໌͕Ͱ͖͍ͯΔ͋Δ w όϯσΟοτʹ͓͍ͯΞʔϜͷબʹެฏੑͷ੍ w σʔλʹόΠΞε͕ೖ͍ͬͯͳ͍ w ֶशʹ͓͚ΔόΠΞεͷআڈ͕త w
ఢରతόϯσΟοτͷઃఆͰ࠷దͳΞϧΰϦζϜΛఏҊ w จ຺͖όϯσΟοτʹ͓͚Δ݁Ռ͋Δ͕࠷దੑͳ͠ !40
࠷ۙͷల։ w طଘͷެฏੑͷఆٛʹٙ w ཧతʹެฏੑఆٛͷਖ਼ԽΛ͍ͨ͠ w طଘͷఆٛͷ w %FNPHSBQIJDQBSJUZٯࠩผ w
&RVBMJ[FEPEETσʔλʹؚ·ΕΔόΠΞεΛऔΓআ͚ͳ͍ w *OEJWJEVBMGBJSOFTTڑവͷఆٛ !41
%FMBZFE&⒎FDU<-JV > w ֶशͱςετͷؒʹ࣌ؒతִͨΓ͕͋Δ w ͦͷؒʹαϯϓϧͷ͕มԽ͢Δ w %FNPHSBQIJDQBSJUZͷਖ਼ੑ ೖࢼ
w ශࠔͷֶੜΛऔΒͳ͍͜ͱͰকདྷශࠔ͕֦େ͢Δ͜ͱͷࢭ w %1 &0ͷ੍Λ͚ͭͨ࣌༧ଌ࣌ͷੑೳͲ͏ͳΔ͔ !42 ࣌ࠁ σʔλऩू ֶश ༧ଌ αϯϓϧͷ͕มԽ
·ͱΊ w ެฏੑʹ͓͚Δཧత՝ w ൚ԽੑೳͷղੳΒΕ࢝Ί͍ͯΔ w ࠷దੑͷূ໌͕େ͖ͳ՝ͱ͍ͯͬͯ͠Δ w ཧతͳެฏੑఆٛͷਖ਼ԽҰ൪େ͖ͳ՝ !43