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
深層ニューラルネットワークにおける訓練高速化のための自動最適化
Search
Kazuhiro Serizawa
March 07, 2019
Research
95
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
深層ニューラルネットワークにおける訓練高速化のための自動最適化
My slide at "第168回HPC研究会".
http://id.nii.ac.jp/1001/00194707/
Kazuhiro Serizawa
March 07, 2019
More Decks by Kazuhiro Serizawa
See All by Kazuhiro Serizawa
Accelerating Machine Learning I/O by Overlapping Data Staging and Mini-batch Generations
serihiro
1
390
hpc170_slide.pdf
serihiro
0
110
画像解像度別ImageNetの100 iterationの合計処理時間比較
serihiro
0
180
Introduction to Parallel Computing 2.2
serihiro
0
93
My summer internship result at Treasure Data 2018 #td_intern
serihiro
0
2.2k
startupでもrails使うなら これだけはやっとけ的 tips集
serihiro
19
10k
つらくないコードレビューの運用
serihiro
43
21k
5分で分かるかもしれないjava8 Stream API
serihiro
1
2.4k
Other Decks in Research
See All in Research
COMETAを用いたデータ民主化運動の歴史
sazimai
0
280
2026年 オープンキャンパス 研究室紹介
junkurihara
0
250
論文紹介:Doc-to-LoRA: Learning to Instantly Internalize Contexts
yukako_nakano
0
200
Visual SLAM未来予測 / Future Prediction in Visual SLAM
koide3
1
1.3k
Kunai: Toward a Verifier-Safe Layered DSL for Nested Encapsulation Packet Filtering on eBPF
takehaya
1
150
[最先端NLP勉強会2026] Checklists Are Better Than Reward Models For Aligning Language Models
nzw0301
1
330
GLIM とMegaParticles:正規分布近似の限界とタイトカップリング&パーティクルフィルタの進展 / GLIM and MegaParticles : Progress of the distribution representation in SLAM
koide3
0
960
Pesée et pointage par imagerie 3D : PHENO3D entre en phase de test terrains
institutdelelevage
PRO
0
110
Byzantine Eventual Consistency for Reliable Local-First Software
ept
0
220
XDPerf: A High-Performance Traffic Generator Built with WASM and eBPF
takehaya
1
340
2025年度秋葉原ウォーカブルプロジェクト調査報告 「アキバらしいウォーカブル」とは何か
izumiyama_lab
1
250
[WebDB2026]セレンディピティ指向推薦システム再考 ―セレンディピティの原義・類型・発生過程に基づく設計指針―
recsyslab
PRO
0
210
Featured
See All Featured
Noah Learner - AI + Me: how we built a GSC Bulk Export data pipeline
techseoconnect
PRO
0
460
How To Stay Up To Date on Web Technology
chriscoyier
790
250k
My Coaching Mixtape
mlcsv
0
350
Building Flexible Design Systems
yeseniaperezcruz
330
41k
Bash Introduction
62gerente
615
220k
Build The Right Thing And Hit Your Dates
maggiecrowley
39
3.5k
Designing Powerful Visuals for Engaging Learning
tmiket
1
600
世界の人気アプリ100個を分析して見えたペイウォール設計の心得
akihiro_kokubo
PRO
74
43k
A better future with KSS
kneath
240
18k
<Decoding/> the Language of Devs - We Love SEO 2024
nikkihalliwell
1
360
Self-Hosted WebAssembly Runtime for Runtime-Neutral Checkpoint/Restore in Edge–Cloud Continuum
chikuwait
0
890
Building Better People: How to give real-time feedback that sticks.
wjessup
370
20k
Transcript
ਂχϡʔϥϧωοτϫʔΫ ʹ͓͚Δ܇࿅ߴԽͷͨΊͷ ࣗಈ࠷దԽ ۔༸ ݐ෦मݟ ஜେֶେֶӃγεςϜใֶݚڀՊ ஜେֶܭࢉՊֶݚڀηϯλʔ ۔༸ ݐ෦मݟਂχϡʔϥϧωοτϫʔΫʹ͓͚Δ܇࿅ߴԽͷͨ
Ίͷࣗಈ࠷దԽ ใॲཧֶձୈճ)1$ݚڀձใࠂ )1$ 7PM )1$ /P .BS !ੴࢁԹઘྤཨޫ
ൃද༰ w ݚڀഎܠ w ؔ࿈ݚڀʢϋΠύʔύϥϝʔλνϡʔχϯάɼࢄ%//ʣ w ༧උ࣮ݧ w ఏҊख๏ w
ఏҊख๏(16ར༻Λར༻ͨ͠࠷దԽ w ఏҊख๏܇࿅ॲཧ࣌ؒͷਪҠΛར༻ͨ͠࠷దԽ w ධՁ࣮ݧ w ݁ 2
ݚڀͷഎܠʢʣ w ۙɼਂχϡʔϥϧωοτϫʔΫʢҎԼ%//ʣΛ༻͍ ͨػցֶशͷར༻͕׆ൃԽ Ґ
Ґ Ґ ̐Ґ Ґ Ґ Ґ IUUQJNBHFOFUPSHDIBMMFOHFT-473$SFTVMUTIUNMΑΓҾ༻ #FUUFS *-473$ͷ ը૾ೝࣝ෦ͷUPQ Τϥʔ<> "MFY/FU ΈࠐΈχϡʔϥϧ ωοτϫʔΫ Λ࠾༻ 3
ݚڀͷഎܠʢʣ Ϟσϧͷઃܭ ϋΠύʔ ύϥϝʔλͷ ܾఆ ܇࿅ॲཧ ܇࿅ࡁΈ ϞσϧͷධՁ ࣌ؒʙؒ ϑΟʔυόοΫ
܇࿅࣌ؒͷظԽ ༻αʔϏεʹ͓͍ͯ ػձଛࣦͷՄೳੑ͕ݒ೦͞ΕΔ 4
w %//ͷ܇࿅࣌ؒॖͷͨΊͷΞϓϩʔν w (16ͷ࠾༻ w ϛχόον܇࿅ͷ࠾༻ w ࢄॲཧԽ w ઐ༻ϋʔυͷ࠾༻
w Ϟσϧѹॖ ݚڀͷഎܠʢʣ ຊݚڀͰѻ͏ྖҬ 5
ݚڀͷత w ϊʔυ(16ͷڥʹ͓͍ͯɼϛχόονͷύϥϝʔλͰ͋Δ ϛχόοναΠζΛɼ܇࿅࣌ؒΛ࠷খԽ͢ΔΑ͏ʹࣗಈͰ࠷ద Խ͢Δख๏ΛఏҊ͢Δ w ఏҊख๏ΛػցֶशϑϨʔϜϫʔΫ$IBJOFSʹ࣮͠ධՁΛߦ͏ 6
ؔ࿈ݚڀʢʣ w ࢄॲཧΛ༻͍ٖͯࣅతʹେنͳϛχόον܇࿅Λ࣮ݱ w ύϥϝʔλαʔόΛ༻͍ͨඇಉظࢄฒྻ܇࿅ <%FBOFUBM > w .1*ͷ"MM3FEVDFΛ༻͍ͨಉظࢄฒྻ܇࿅ <"LJCBFUBM
> େྔͷܭࢉϊʔυΛલఏͱͨ͠ख๏Ͱ͋Γɼϊʔυ୯ҐͰͷ࠷దԽ·Ͱ ߟྀ͍ͯ͠ͳ͍ ຊݚڀͰϊʔυ(16୯ҐͰͷ܇࿅Λ্ͤ͞ɼ ࢄ܇࿅ʹద༻ͯ͠ߴԽʹد༩Ͱ͖Δख๏ΛఏҊ͢Δ 7
ؔ࿈ݚڀʢʣ w ػցֶशϞσϧͷਫ਼Λ࠷େԽ͢ΔϋΠύʔύϥϝʔλ νϡʔχϯά w 3BOEPNTFBSDI ߏ1BS[FOਪఆثΛ༻ͨ͠ख๏ <#FSHTUSBFUBM > w
ϕΠζ࠷దԽΛ༻͍ͨख๏<4OPFLFUBM > ܇࿅Λ࠷େԽ͢ΔͨΊͷϋΠύʔύϥϝʔλνϡʔχϯάख๏ ͜Ε·ͰఏҊ͞Ε͍ͯͳ͍ ຊݚڀͰ܇࿅Λ࠷େԽ͢ΔͨΊͷϛχόον αΠζͷࣗಈ࠷దԽख๏Λݕ౼ 8
ϛχόον܇࿅ʹ͍ͭͯ ʢʣ ϛχόονΛ࠾༻ͤͣʹ܇࿅͢Δ߹ɼ ܇࿅σʔλͷ͕܇࿅ճʹ݁͢Δ ܇࿅σʔλ͕ ຕ ࠷Ͱ JUFSBUJPOͷ܇࿅͕ඞཁ 9
܇࿅σʔλ͕ ຕ JUFSBUJPO ͷ܇࿅͕ඞཁ ϛχόον͝ͱʹFNCBSSBTTJOHMZQBSBMMFMͳ σʔλฒྻԽ͕Մೳ ϛχόον܇࿅ʹ͍ͭͯ ʢʣ
ϛχόονԽʹΑΓ܇࿅ճͷݮ͕ՄೳʹͳΔ 10
ϛχόοναΠζΛ૿Ճͤ͞ଓ͚Δͱ(16ଆͷԋࢉίετ͕಄ ଧͪʹͳͬͯॲཧ࣌ؒͷݮޮՌ͕ࣦΘΕΔ͜ͱ͕ఆ͞ΕΔ ϛχόοναΠζ ϛχόοναΠζ ܇࿅࣌ͷ߹ܭԋࢉྔ<qPQ> ߹ܭ܇࿅ॲཧ࣌ؒ<TFD> ཧ্ͷ܇࿅ॲཧ࣌ؒͷਪҠ ʢը૾ຕ͋ͨΓͷॲཧ͕࣌ؒ Ұఆͷ߹ʣ ཧ্ͷԋࢉྔͷਪҠ
U U O O ϛχόον܇࿅ʹ͍ͭͯ ʢʣ 11 ҰൠతʹɼͳͲ͕ϛχόοναΠζͱͯ͠༻͞Ε͍ͯΔ͕ ϛχόοναΠζΛܾఆ͢Δ٬؍తͳࢦඪΒΕ͍ͯͳ͍
༧උ࣮ݧʢʣ w ϛχόοναΠζΛͷൣғͰม͑ͳ͕ΒΈࠐΈ χϡʔϥϧωοτϫʔΫʢ7((̍̒ʣΛ܇࿅ͯ͠ɺҎԼͷ ؔΛௐࠪ͢Δ w ϛχόοναΠζͱ܇࿅࣌ؒͷؔ w ϛχόοναΠζͱฏۉ(16ར༻ͷؔ ฏۉ(16ར༻܇࿅தʹOWQSPGͰඵִؒͰܭଌͨ͠ͷฏۉ
12
༧උ࣮ݧʢʣ ϛχόοναΠζΛ૿͢ը૾ຕ͋ͨΓͷ܇࿅͕࣌ؒॖ ϛχόοναΠζ͝ͱͷ܇࿅ॲཧ࣌ؒͷਪҠ ϛχόοναΠζ͕ Λ͑ͨͨΓͰมԽ͕ऩଋ ͠εέʔϧ͠ͳ͘ͳΔ 'BTUFS 13
༧උ࣮ݧʢʣ ϛχόοναΠζΛ૿͢ฏۉ(16ར༻্͕ঢ ϛχόοναΠζ͝ͱͷฏۉ(16ར༻ͷਪҠ ϛχόοναΠζ͕ Λ͑ͨͨΓͰ ฏۉ(ۙʹ౸ୡ 14
༧උ࣮ݧʢʣ ϛχόοναΠζΛաʹ૿͢ϞσϧͷੑೳʹѱӨڹ FQPDI ը૾ೝࣝਖ਼ղ #FUUFS ϛχόοναΠζ͝ͱͷϞσϧੑೳ
15
༧උ࣮ݧ͔Βߟ͑ΒΕΔԾઆ ϛχόοναΠζΛ૿͢ w ը૾ຕ͋ͨΓͷ܇࿅͕࣌ؒݮগ w ฏۉ(16ར༻্͕ঢ w ྆ऀͱҰఆͷαΠζͰऩଋ w աʹ૿͢ͱϞσϧੑೳʹѱӨڹ
ฏۉ(16ར༻ͷมԽ͕ऩଋ࢝͠ΊΔલޙͷ ϛχόοναΠζ͕܇࿅࣌ؒΛ࠷খԽ͢Δ ༧උ࣮ݧΑΓ 16 खಈͰϛχόοναΠζ͝ͱʹ(16ར༻Λܭଌ͠ͳ͕Β ϚχϡΞϧͰϛχόοναΠζΛௐ͢Δͷඇৗʹख͕͔͔ؒΔ
ఏҊख๏ w ҎԼ̎ͭͷख๏ΛఏҊ w ఏҊख๏ ฏۉ(16ར༻͕ࢦఆͨ͠ʹۙͮ͘Α͏ʹ ܇࿅Λߦ͍ͳ͕ΒϛχόοναΠζΛ ࣗಈͰ࠷దԽ w ఏҊख๏
܇࿅ॲཧ࣌ؒͷมԽ͕࠷খʹͳΔΑ͏ʹ܇࿅ Λߦ͍ͳ͕ΒϛχόοναΠζΛࣗಈͰ࠷దԽ 17
ఏҊख๏ʢʣ ᶃฏۉ(16ར༻͕ඪͷ ྫ͑ ʹ ۙͮ͘Α͏ʹϛχόοναΠζΛมԽͤ͞Δ ᶄϛχόοναΠζͷมԽ͕ ऩଋͨ͠Βͦ͜Λ࠷దͱͯ͠ ࠾༻͢Δ ඪͷฏۉ(16ར༻
ܭଌͨ͠ฏۉ(16ར༻ ͕ ࠷খʹͳΔΑ͏ʹ࠷దԽ 18
ఏҊख๏ʢʣ /ճ܇࿅Λߦ͏ ্Ґͷܭଌ͔Βฏۉ(16ར༻Λܭࢉ ඵ͝ͱͷ(16ར༻ ΛOWNMͰඇಉظͰܭଌ ϛχόοναΠζΛௐ ฏۉ(16ར༻͕ඪͷΑΓখ͍͞ ૿Ճݮ ϛχόοναΠζΛॳظԽ ϛχόοναΠζͷมԽ͕ऩଋͨ͠
'BMTF 5SVF ࠷దԽऴྃ ॲཧϑϩʔ 19
ఏҊख๏̎ʢʣ ᶃॳظϛχόοναΠζͷ ॲཧ࣌ؒΛϕʔεϥΠϯͱ͢Δ ᶄϕʔεϥΠϯ͔Βͷॲཧ࣌ؒͷվળ͕ ऩଋͨ͠Βͦͷ࣌ͷϛχόοναΠζΛ ɹ࠷దͱͯ͠࠾༻͢Δ ϕʔεϥΠϯ͔Βͷ܇࿅ॲཧ࣌ؒͷվળ͕มԽ͠ͳ͘ͳΔ·Ͱ ϛχόοναΠζΛ૿Ճͤ͞Δ 20
ఏҊख๏̎ʢʣ FQPDIͨΓͷ܇࿅ॲཧ࣌ؒ<TFD> JUFSBUJPOͨΓͷॲཧ࣌ؒ<TFD>Y FQPDIʹඞཁͳΠςϨʔγϣϯ σʔλαΠζϛχόοναΠζʣ 21 ϕʔεϥΠϯͷ܇࿅ॲཧ࣌ؒ ϛχόον͝ͱͷ܇࿅ॲཧ࣌ؒ ϕʔεϥΠϯͷ܇࿅ॲཧ࣌ؒ
FQPDIͨΓͷ܇࿅ॲཧ࣌ؒͷվળ
ఏҊख๏̎ʢʣ /ճ܇࿅Λߦ͍ɼॲཧ࣌ؒΛܭଌ FQPDI͋ͨΓͷ܇࿅ॲཧ࣌ؒվળΛࢉग़ ϛχόοναΠζΛ૿Ճ ϛχόοναΠζΛॳظԽ FQPDI͋ͨΓͷ܇࿅ॲཧ࣌ؒվળͷมԽ͕ऩଋͨ͠ʁ 'BMTF 5SVF ࠷దԽऴྃ ॲཧϑϩʔ
ॳճͷܭଌΛϕʔε ϥΠϯͱͯ͠อଘ 22
$IBJOFSΛ༻͍࣮ͨํ๏ .PEFMΛΠϯελϯεԽ NPEFM7(( 0QUJNJ[FSΛΠϯελϯεԽͯ͠.PEFMͱώϞ PQUJNJ[FS.PNFOUVN4(% NPEFM ςετσʔλΛϩʔυͯ͠*UFSBUPSΛΠϯελϯεԽ
JUFSBUPS4FSJBM*UFSBUPS HFU@DJGBS 6QEBUFSΛΠϯελϯεԽͯ͠JUFSBUPS PQUJNJ[FSͱώϞ VQEBUFS4UBOEBSE6QEBUFS JUFSBUPS PQUJNJ[FS 5SBJOFSΛΠϯελϯεԽͯ͠VQEBUFSͱώϞ USBJOFS5SBJOFS VQEBUFS ఏҊख๏Λ࣮ͨ͠&YUFOTJPOΛՃ USBJOFSFYUFOE .JOJCBUDI4J[F0QUJNJ[FS ܇࿅ϧʔϓΛ࣮ߦ USBJOFSSVO $IBJOFSΛ༻͍ͨ܇࿅ॲཧεΫϦϓτͷ࣮ྫͱओཁΫϥεͷॴ༗ؔ 23
ධՁ࣮ݧ֓ཁʢʣ w ఏҊख๏ Λ༻͍ͯҎԼͷ༰ΛධՁ͢Δ w ҙਤͨ͠ͱ͓Γʹ࠷దԽ͕ਐΈऩଋ͢Δ͔Ͳ͏͔ w ऩଋ݅ w ఏҊख๏ϛχόοναΠζͷมԽ͕Ҏ
w ఏҊख๏FQPDI͋ͨΓͷॲཧ࣌ؒվળ͕Ҏ 24
ධՁ࣮ݧ֓ཁʢʣ ༻͢ΔσʔληοτͱωοτϫʔΫͷΈ߹Θͤ σʔληοτʢը૾ղ૾<QJYFM>ʣ ωοτϫʔΫ $JGBSʢYʣ 7(( $JGBSʢYʣ 3FT/FU *NBHF/FULʢYʣ 7((
*NBHF/FULʢYʣ 3FT/FU 25
ධՁ࣮ݧ֓ཁʢʣ ධՁڥ TQFD $16 9FPO 3 $16&W!()[Y .FNPSZ (J# (16
/7*%*"5FTMB7(J# 04 $FOU04 1ZUIPO $IBJOFS B GPSL࣌ $6%" DV%// W 26
ఏҊख๏̍ʹ͓͚Δ $JGBSͷධՁ݁Ռ $JGBSͷ ฏۉ(16ར༻ऩଋͷ༷ࢠ 7(( 3FT/FU w ࠷దԽͷਐߦͱڞʹฏۉ(16ར༻͕૿Ճ w ϛχόοναΠζͱฏۉ(16ར༻ͱͷؒʹਖ਼ͷ૬ؔੑ͕ݟΒΕΔ
27 $JGBSͷϛχόοναΠζͱ ฏۉ(16ར༻ͷؔ 7(( 3FT/FU )JHIFS ࠷దԽਐḿ େ খ
*NBHF/FULͷ ฏۉ(16ར༻ऩଋͷ༷ࢠ 7(( 3FT/FU $JGBSͱಉ༷ͷ͕ݟΒΕΔ͕ܭଌ͞Εͨ(16ར༻ͷࢄ͕େ͖͍ ఏҊख๏̍ʹ͓͚Δ *NBHF/FULͷධՁ݁Ռ 28 7(( 3FT/FU
*NBHF/FULͷϛχόοναΠζͱ ฏۉ(16ར༻ͷؔ )JHIFS ࠷దԽਐḿ େ খ
ఏҊख๏̎ʹ͓͚Δ $JGBSͷධՁ݁Ռ $JGBSͷ܇࿅ॲཧ࣌ؒ վળऩଋͷ༷ࢠ #FUUFS 29 7(( 3FT/FU
$JGBSͷϛχόοναΠζͱ ܇࿅ॲཧ࣌ؒվળͷؔ w ࠷దԽͷਐߦͱڞʹFQPDIͨΓͷॲཧ࣌ؒվળ͕૿Ճ w ϛχόοναΠζͱվળͱͷؒʹਖ਼ͷ૬ؔੑ͕ݟΒΕΔ ࠷దԽਐḿ େ খ
ఏҊख๏̎ʹ͓͚Δ *NBHF/FULͷධՁ݁Ռ *NBHF/FULͷ܇࿅ॲཧ࣌ؒ վળऩଋͷ༷ࢠ $JGBSͷධՁ݁Ռͱ΄΅ಉͷΛ͕ࣔ͢ $JGBSΑΓվળ͕͍ 30 7((
3FT/FU *NBHF/FULͷϛχόοναΠζͱ ܇࿅ॲཧ࣌ؒվળͷؔ #FUUFS ࠷దԽਐḿ େ খ
ಘΒΕͨ࠷దΛ༻͍ͯ FQPDI܇࿅ͨ݁͠Ռ $JGBSͷܭଌ݁Ռ *NBHF/FULͷܭଌ݁Ռ #FUUFS ಘΒΕͨϛχόονͷ࠷దΛ༻͍ͯFQPDIͷ܇࿅Λߦ͍ ܇࿅࣌ؒΛܭଌͨ݁͠ՌɼͲͷέʔεʹ͓͍ͯϕʔεϥΠϯ͔Βվળ͕ݟΒΕͨ CBTF QSPQPTFE QSPQPTFE
CBTF QSPQPTFE QSPQPTFE #FUUFS 31
ධՁ݁Ռ·ͱΊ ධՁύλʔϯ ఏҊख๏ ఏҊख๏̎ ύλʔϯ ύλʔϯ
ύλʔϯ ύλʔϯ ࠷దͱͯ͠ಘΒΕͨϛχόοναΠζ ࠷దͱͯ͠ಘΒΕͨϛχόοναΠζΛൺֱ͢Δͱ ࠷దԽख๏ؒຖͰ͕ࠩݟΒΕͨ 32
$JGBSͰͷධՁ݁Ռ ʹ͓͚Δߟʢʣ ύλʔϯ ʢ$JGBSʣʹ͓͚Δ྆ख๏ͰಘΒΕͨ࠷దͱվળͷൺֱ 7(( 3FT/FU ఏҊख๏Ͱ ಘΒΕͨ࠷ద ఏҊख๏̎Ͱ ಘΒΕͨ࠷ద
#FUUFS $JGBSͰಘΒΕͨ࠷దఏҊख๏̎ख๏ͷํ͕࠷ద 33
$JGBSͰͷධՁ݁Ռ ʹ͓͚Δߟʢʣ #FUUFS 7(( 3FT/FU ఏҊख๏ ఏҊख๏ ฏۉ(16ར༻͕ۙ͘ʹͳͬͨҎ߱ ϛχόονʹΑΔॲཧ࣌ؒͷݮޮՌ͕֬ೝ͞ΕΔ FQPDIͨΓͷ܇࿅ॲཧ͕࣌ؒ
ϛχόοναΠζͷมԽʹൺྫ͢ Δ߹ͷཧۂઢ ఏҊख๏̎ 34 (16 ϛχόοναΠζ
ύλʔϯ̏ ̐ʢ*NBHF/FULʣʹ͓͚Δ྆ख๏ͰಘΒΕͨ࠷దͱվળͷൺֱ 7(( 3FT/FU #FUUFS *NBHF/FULͰಘΒΕͨ࠷ద྆खํؒͰେ͖ͳࠩݟΒΕͳ͍ ఏҊख๏ͰಘΒΕͨ࠷ద ఏҊख๏̎Ͱ ಘΒΕͨ࠷ద *NBHF/FULͰͷධՁ݁Ռ
ʹ͓͚Δߟʢʣ 35
*NBHF/FULͰͷධՁ݁Ռ ʹ͓͚Δߟʢʣ 7(( 3FT/FU #FUUFS ఏҊख๏ ఏҊख๏ (16ʹΑΔߴԽϛχόονʹΑΔॲཧݮޮՌͷ྆ํ͕ ͋·ΓޮՌతͰͳ͍ վળͷਪҠ͕ॳظ͔Βͷ
ϛχόοναΠζͷมԽʹൺྫ ͢Δ߹ͷվળਪҠ վળͷਪҠ͕ॳظ͔Βͷ ϛχόοναΠζͷมԽʹͷΈ ൺྫ͢Δ߹ͷཧۂઢ 36
ฏۉ(16ར༻ͷࢄʹ ؔ͢Δߟʢʣ $JGBS 7(( ύ$JGBS 3FT/FU *NBHF/FUL 7(( *NBHF/FUL 3FT/FU
w ฏۉ(16ར༻ͷࢄ*NBHF/FUL$JGBS w σʔληοτͷαΠζࠩʹΑͬͯ(16ͷར༻ঢ়گ͕ ҟͳΔՄೳੑ͕ߟ͑ΒΕΔ 37
*NBHF/FULΛ༻ͨ͠߹ɼJUFSBUJPO͝ͱʹ ඵ(16Χʔωϧ͕࣮ߦ͞Ε͍ͯͳ͍࣌ؒଳ͕ଘࡏ͢Δ ܇࿅ॲཧதʢJUFSBUJPOʣͷ(16ར༻ঢ়گΛώʔτϚοϓͱͯ͠ՄࢹԽͨ͠ਤ $JGBS *NBHF/FUL ܇࿅ॲཧதͷ(16Χʔωϧ࣮ߦঢ়گ ฏۉ(16ར༻ͷࢄʹ ؔ͢Δߟʢʣ 38 (16͕ར༻͞Ε͍ͯͳ͍࣌ؒଳ
ը૾ಡΈࠐΈ࣌ͷ%JTL*0ը૾͔Βߦྻͷมॲཧ͕ߦΘΕ͍ͯΔͱߟ͑ΒΕΔ
݁ w ຊݚڀͰɼ(16ར༻ͱ܇࿅ͷվળʹணͨ͠ϛχόονα Πζͷ࠷దԽख๏ΛఏҊͨ͠ w $JGBSΛ༻͍ͨ܇࿅ʹ͓͍ͯɼϛχόοναΠζͱൺֱͯ͠࠷ େͰFQPDIͨΓͷ܇࿅ॲཧ࣌ؒΛվળ͢Δϛχόονα ΠζΛࣗಈͰ୳ࡧ͢Δ͜ͱ͕Ͱ͖ͨ w *NBHF/FULΛ༻͍ͨ܇࿅ʹ͓͍ͯɼϛχόοναΠζͱൺֱͯ͠
࠷େͰFQPDIͨΓͷ܇࿅ॲཧ࣌ؒΛ࠷େվળ͢Δϛχ όοναΠζΛࣗಈͰ୳ࡧ͢Δ͜ͱ͕Ͱ͖ͨ w *NBHF/FULΛ༻͍ͨ܇࿅Ͱ܇࿅ॲཧͷ(16Λར༻͍ͯ͠Δ࣌ؒͷ ׂ߹͕$JGBSͱൺ͍ͯ͜ͱ͕֬ೝ͞Εͨ 39
ࠓޙͷ՝ w JUFSBUJPOؒͷ(16Χʔωϧ͕࣮ߦ͞Ε͍ͯͳ͍࣌ؒͷ ݮํ๏Λݕ౼͢Δ w ܇࿅σʔλϩʔυ࣌ͷ%JTL*0ͷߴԽ w ܇࿅σʔλͷมॲཧͳͲͷલॲཧͷӅṭ 40
ҎԼ༧උεϥΠυ 41
ఏҊख๏̎ͷ ධՁ݁Ռͷݕূ ධՁύλʔϯ ఏҊख๏̎ख๏Ͱਪଌ͞Εͨ࠷ద ʹ͓͚Δվળ<> ࣮ࡍʹܭଌ͞Εͨվળ<> ύλʔϯ ύλʔϯ
ύλʔϯ ύλʔϯ ಘΒΕͨϛχόονͷ࠷దΛ༻͍ͯFQPDIͷ܇࿅Λߦ͍ɼ ॳظ͔ΒͷվળΛܭଌͨ͠ͱ͜Ζɼਪଌͱܭଌ΄΅ಉΛࣔͨ͠ 42
ఏҊख๏ͷ ධՁ݁Ռͷݕূʢৄࡉʣ ධՁύλʔϯ ਪଌ͞Εͨվળ <> ܭଌ͞Εͨվળ <> ॳظͰͷܭଌ <TFD> ࠷దͰͷܭଌ
<TFD> ύλʔϯ ύλʔϯ ύλʔϯ ύλʔϯ ಘΒΕͨϛχόονͷ࠷దΛ༻͍ͯFQPDIͷ܇࿅Λߦ͍ ॲཧ࣌ؒͱॳظ͔ΒͷվળΛܭଌͨ͠ 43
ఏҊख๏ͱఏҊख๏ͷ ࠷దൺֱ ධՁύλʔϯ ॳظͰͷܭଌ<TFD> ఏҊख๏ͷ࠷దͰͷ ܭଌ<TFD> ఏҊख๏Ͱͷ࠷దͰ ͷܭଌ<TFD> ύλʔϯ
ύλʔϯ ύλʔϯ ύλʔϯ ಘΒΕͨϛχόονͷ࠷దΛ༻͍ͯFQPDIͷ܇࿅Λߦ͍ ॲཧ࣌ؒΛܭଌͨ͠ 44
ఏҊख๏ͷ ධՁ݁Ռͷݕূ σʔληοτ ωοτϫʔΫ ϕʔεϥΠϯ<TFD> ఏҊख๏ʹ͓͚Δ࣮ଌ <TFD> $JGBS 7((
3FT/FU *NBHF/FUL 7(( 3FT/FU ఏҊख๏ͰಘΒΕͨ࠷దΛ༻͍ͯFQPDIܭଌͨ݁͠Ռ 45
ఏҊख๏ͷ ධՁ݁Ռͷݕূ σʔληοτ ωοτϫʔΫ ϕʔεϥΠϯ<TFD> ఏҊख๏ʹ͓͚Δ࣮ଌ <TFD> $JGBS 7((
3FT/FU *NBHF/FUL 7(( 3FT/FU ఏҊख๏ͰಘΒΕͨ࠷దΛ༻͍ͯFQPDIܭଌͨ݁͠Ռ 46
$JGBS 7((ʹ͓͚Δ ϛχόοναΠζ͝ͱͷਫ਼ NBJOBDDVSBDZ FQPDI WBMJEBUJPOBDDVSBDZ FQPDI
47
$JGBS 3FT/FUʹ͓͚Δ ϛχόοναΠζ͝ͱͷਫ਼ NBJOBDDVSBDZ FQPDI WBMJEBUJPOBDDVSBDZ FQPDI
48
FQPDI͋ͨΓͷॲཧ࣌ؒվળ ͷཧۂઢͷܭࢉϩδοΫ 49
ධՁதͷܭଌྫ
DIFDL DIFDL DPOWFSHFODF DIFDL লུ DIFDL DPOWFSHFODF ఏҊख๏ ఏҊख๏̎ ϛχόον αΠζ ϛχόον αΠζ 50