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
若手カンストしたので基盤がんばるぞい
Search
Hiroki (REO) Kashiwazaki
August 27, 2021
Science
0
69
若手カンストしたので基盤がんばるぞい
Hiroki (REO) Kashiwazaki
August 27, 2021
Tweet
Share
More Decks by Hiroki (REO) Kashiwazaki
See All by Hiroki (REO) Kashiwazaki
自律システムの運用を実践する環境 “AS DOJO” の設計
reokashiwa
0
6
Decentralized resource allocation for Beyond 5G era
reokashiwa
0
14
Resource-transparent wide-area distributed computing environment with a functional paradigm in the Beyond 5G era.
reokashiwa
0
12
Resource-transparent wide-area distributed computing environment with a functional paradigm in the Beyond 5G era.
reokashiwa
0
25
Distcloud update 2022Q1~Q2 And Beyond 5G (B5G) update
reokashiwa
0
34
distcloud: distributed virtualization platform
reokashiwa
0
160
Brief discussion of container leveraged cloud
reokashiwa
0
47
DESTCloud now and the future
reokashiwa
1
500
Other Decks in Science
See All in Science
People who frequently use ChatGPT for writing tasks are accurate and robust detectors of AI-generated text
rudorudo11
0
120
academist Prize 4期生 研究トーク延長戦!「美は世界を救う」っていうけど、どうやって?
jimpe_hitsuwari
0
400
データベース11: 正規化(1/2) - 望ましくない関係スキーマ
trycycle
PRO
0
960
Collective Predictive Coding as a Unified Theory for the Socio-Cognitive Human Minds
tanichu
0
110
Gemini Prompt Engineering: Practical Techniques for Tangible AI Outcomes
mfonobong
2
180
機械学習 - K-means & 階層的クラスタリング
trycycle
PRO
0
1.1k
Ignite の1年間の軌跡
ktombow
0
160
点群ライブラリPDALをGoogleColabにて実行する方法の紹介
kentaitakura
1
470
AI(人工知能)の過去・現在・未来 —AIは人間を超えるのか—
tagtag
0
110
Quelles valorisations des logiciels vers le monde socio-économique dans un contexte de Science Ouverte ?
bluehats
1
550
生成検索エンジン最適化に関する研究の紹介
ynakano
2
1.4k
機械学習 - pandas入門
trycycle
PRO
0
330
Featured
See All Featured
Become a Pro
speakerdeck
PRO
29
5.6k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
46
7.7k
What’s in a name? Adding method to the madness
productmarketing
PRO
24
3.7k
Context Engineering - Making Every Token Count
addyosmani
7
290
Product Roadmaps are Hard
iamctodd
PRO
55
11k
I Don’t Have Time: Getting Over the Fear to Launch Your Podcast
jcasabona
34
2.5k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
31
2.7k
Imperfection Machines: The Place of Print at Facebook
scottboms
269
13k
Building Adaptive Systems
keathley
44
2.8k
Writing Fast Ruby
sferik
629
62k
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
32
1.7k
GitHub's CSS Performance
jonrohan
1032
470k
Transcript
Hiroki Kashiwazaki (NII) 27th Aug. 2021 एखΧϯετͨ͠ͷͰ ج൫͕ΜΔ͍ͧ Reaching the
limit to apply for Early-Career Scientists, so I will challenge the Scienti fi c Research.
ಈػ͚ ӡ༻࣭ʹҼΔΠϯηϯςΟϒʹؔ͢ΔݚڀΛߦதɻ ߴ͍ӡ༻࣭ ސ٬͕ཁٻ͢Δੑೳ͕ຬ͢Δ֬ ʹߴ͍ ใुΛ༩͑ɺͦͦ͜͜ͷӡ༻࣭ʹͦͦ͜͜ͷใुΛɻ ճઢΛԽͨ͠Β Խͯ͠దʹӡ༻͞ΕΕ ӡ༻ ্࣭͢ΔɻͲΕ͙Β্͍͢Δͷ͔
ԽճઢҰఆͷ֬Ͱো͕ൃੜ͢Δɻଟॏো͋ Γ͑Δ ྫ4*/&5 ɻ ଟॏো͕ൃੜͯ͠ɺ͕ͳ͍߹͋Δɻ࣏Ո ඒஊʹ͕ͨ͠Δ͚ΕͲɺݸʑͷোಉ͡ॏΈͰධՁ͞ ΕΔͷͰͳ͘ɺൃੜ͢Δ֬ʹԠͯ͡ධՁ͞ΕΔͷͰ ͳ͔Ζ͏͔ɻ
ϨδϦΤϯε ؤڧ͞ɺͶΓͮΑ͞ɻ ো͕ൃੜͯ͠ʮେৎʯɻ ͋Δܥ͕͋Γɺͦͷ্ͰΞϓϦέʔγϣϯ͕ಈ࡞͢Δ࣌ɺ ͋Δܥͷ্Ͱಈ࡞͢ΔΞϓϦέʔγϣϯͷཁٻ༷͕ɺ ҙͷোঢ়ଶʹ͓͍ͯɺຬͨ͞ΕΔ͔Ͳ͏͔ɻ ͢ͳΘͪɺ͋Γͱ͋ΒΏΔΈ߹Θͤͷো͕ൃੜ͢Δ֬ Λ1ͱ͠ɺͦͷ͏ͪͰཁٻ͕ຬͨ͞Εͳ͍֬Λ2 ͋Δ ͍ຬͨ͞ΕΔ֬12
ͱ͢Δͱ͖ɺϨδϦΤϯε3 ͱͯ͠දݱ͞ΕΔͷͰͳ͔Ζ͏͔ ͠ΒΜ͚Ͳ ɻ R = Q P ͋Δ͍ P − Q P
G(V, E) V = {v1, v2, ..., vn } E
= {e1, e2, ..., em } v1 v2 v3 e1 e2 e3 P = {p1, p2, ..., pm } p1 p2 p3 WFSUFY ͷू߹7 ล FEHF ͷू߹& ҙͷลFJ ͷোൃੜ֬QJ ͷू߹1 ఆࣜԽͯ͠ΈΜɻ
&ͷத͔ΒลΛऔΓग़͢Έ߹Θͤͷू߹' F1 = {F1 1 , F1 2 , ...,
F1 mC1 } = {{e1 }, {e2 }, {e3 }} &ͷத͔ΒลΛऔΓग़͢Έ߹Θͤͷू߹' F2 = {F2 1 , F2 2 , ..., F2 mC2 } = {{e1, e2 }, {e1, e3 }, {e2, e3 }} F3 = {F3 1 , F3 2 , ..., F3 mC3 } = {{e1, e2, e3 }} &ͷத͔ΒลΛऔΓग़͢Έ߹Θͤͷू߹' G(V, E) v1 v2 v3 e1 e2 e3 p1 p2 p3 m ลͷ૯
G(V, E) v1 v2 v3 e1 e2 e3 p1 p2
p3 ҰൠԽ͢Δɻ ཁૉNͷू߹&ͷҙͷKݸͷཁૉͷΈ߹Θͤͷू߹' K Fj = {Fj 1 , ..., Fj mCj } ͜͜ͰಥʹɺҙͷลFJΛɺରԠ͢Δোൃੜ֬QJʹ ม͢Δࣸ૾G FJ Λఆٛ͢Δɻ f(ei) = pi ཁૉू߹ʹؚ·ΕΔL൪ͷཁૉΛҎԼͷ௨Γදݱ͢Δɻ ͍ͩ͞ Fj i Fj i [k] m ลͷ૯ j ಉ࣌োൃੜ
Fj i G(V, E) v1 v2 v3 e1 e2 e3
p1 p2 p3 ಉ࣌ൃੜ֬ ཁૉू߹ʹؚ·ΕΔKݸͷลͰো͕ಉ࣌ൃੜ͢Δ֬ Pj i = j f(Fj i [k]) Ώ͑ʹಉ࣌ʹKݸͷลͰো͕ൃੜ͢Δ֬ Ώ͑ʹ͋ΒΏΔো͕ൃੜ͢Δ֬ɺ Pj = mCj i Pj i = mCj i j k f(Fj i [k]) P = m j Pj = m j mCj i j k f(Fj i [k]) m ลͷ૯ j ಉ࣌োൃੜ
ϨδϦΤϯεͷఆٛ ͋ΔΞϓϦέʔγϣϯͷཁٻ༷Λຬͨ͞ͳ͍ಉ࣌K࣍োͷ Έ߹ΘͤΛཁૉͱؚͯ͠Ή෦ू߹4 G(V, E) v1 v2 v3 e1 e2
e3 p1 p2 p3 K Sj = {Sj 1 , ..., Sj σj }(Sj ⊆ Fj) ɺ'ಉ༷ʹಉ࣌ൃੜ֬Λಋ͘͜ͱ͕Ͱ͖Δɻ Q = σj j Sj = m j σj i j k f(Sj i [k]) ͜͜ͰɺϨδϦΤϯεΛҎԼͷΑ͏ʹఆٛ͢Δɻ R = − log Q P ͋ɺ͏ͦɺMPH͍Βͳ͍ͳʜɻ ͏͏ʔΜɺ͔͠͠ʜɻ ͓ؾ࣋ͪΘΔͱࢥ͍·͢ʜɻ m ลͷ૯ j ಉ࣌োൃੜ
͋ΔΞϓϦέʔγϣϯͷཁٻ༷Λຬͨ͞ͳ͍͔Ͳ͏͔ͷ அΛ͢Δʹ͋ͨΓɺҎԼͷ͕͋Δɻ ᶃ શͯͷಉ࣌O࣍োͷΈ߹Θͤ૯ɺลNͷͱ͖ Ͱ͋Γɺࢦؔతʹ૿େ͢ΔɻͦͷશͯͷΈ߹Θͤ Ͱཁٻ༷ͷຬ͔ͨ͢Ͳ͏͔ఆΛߦ͏ඞཁ͕͋Δɻ
ᶄ ཁٻ༷Λຬ͔ͨ͢Ͳ͏͔ͷఆʹཁ͢Δ࣌ؒଟ༷ɻ ؆୯ͳྫ ࿈݁άϥϑఆ m n=1 mCn = 2m − 1
#%%;%% #JOBSZ%FDJTJPO%JBHSBN;FSPTVQQSFTTFECJOBSZ %FDJTJPO%JBHSBN άϥϑΛྻڍࡧҾ͢ΔͨΊͷσʔλ ߏɻ ҙͷ( 7 & ͕༩͑ΒΕͨͱ͖ʹɺ&`Λҙͷ&ͷ෦ू ߹ͱ͢Δͱ͖ɺશͯͷ(
7 &` Λแׅతʹࢉग़͢Δ͜ͱͰɺ ଟ༷ͳ੍݅ʹ͓͚ΔϨδϦΤϯεΛܭࢉ͢Δɻ ͭ·ΓఏҊ͢ΔϨδϦΤϯεͷఆٛʹ߆ΔݶΓʹ͓͍ͯ ᶃͷͲ͏ʹͳΒΜ ͔ɺฒྻܭࢉͳͲͷྗٕ͔͠ ͳͦ͞͏ͳ ͷͰɺᶃºᶄͱ͍͏ঢ়گΛ࡞Γग़͠ɺᶄͷܭ ࢉΛ#%%;%%Ͱݮ͢Δ͜ͱͰɺ݁ՌతʹᶃͷΛ͝ ·͔ͦ͏ͱ͢ΔΞϓϩʔνͰ͋Δ υਖ਼ ɻ
Պݚඅ ج൫# #%%;%%Λ༻͍ͨଟ༷ͳ੍݅ԼͷϨδϦΤϯεධ Ձख๏ͷఏҊͱͦͷԠ༻ ɾݚڀදऀ#%%;%%Λ༻͍ͨϨδϦΤϯεͷධՁख ๏ͷઃܭͱ࣮ɺԠ༻ൣғͷ֦ு യવͱ͍͗ͯ͢͠Δ ͷͰΑΓৄࡉΛ٧ΊΔ ɻ
ྫ .&$ͰΩϟϦΞճઢͭ૿ͨ͠ΒͲΕ͙Β͍Ϩδ ϦΤϯεߴ·Δ අ༻ରޮՌࢉग़Ͱ͖ΔͶϋοϐʔɻ ɾݚڀ୲ऀϨδϦΤϯεΛఆྔతʹධՁͰ͖Δ͜ͱΛ ར༻ͨ͠Ԡ༻ݚڀͷఏҊɺ༗ޮੑͷධՁ ืूத ɻ ໊͙Β͍ूΊΔ͜ͱ͕Ͱ͖Ε ͳγφδʔ͕ ੜ·ΕΔΜ͡Όͳ͍͔ͱ͔ͦ͏͍͏͍ߟ͑ɻ