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
B3_Seminar_05
Search
kakubari
February 24, 2017
Technology
95
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
B3_Seminar_05
ビックデータ解析入門3
kakubari
February 24, 2017
More Decks by kakubari
See All by kakubari
動詞クエリの語間の関係性に基づくクエリマイニング
kakubari
0
130
Neural Modeling of Multi-Predicate Interactions for Japanese Predicate Argument Structure Analysis
kakubari
1
190
Leveraging Crowdsourcing for Paraphrase Recognition
kakubari
0
110
Automatically Acquired Lexical Knowledge Improves Japanese Joint Morphological and Dependency Analysis
kakubari
0
120
Labeling the Semantic Roles of Commas
kakubari
0
100
Integrating Case Frame into Japanese to Chinese Hierarchical Phrase-based Translation Model
kakubari
0
130
Improving Chinese Semantic Role Labelingusing High-quality Surface and Deep Case Frames
kakubari
0
98
Exploring Verb Frames for Sentence Simplification in Hindi
kakubari
0
160
述語項構造と照応関係のアノテーション
kakubari
0
270
Other Decks in Technology
See All in Technology
AI時代のエンジニアキャリアについて今一度考える
sakamoto_582
2
1.5k
End-to-Endで考える信頼性 —LINEアプリにおけるクライアント開発×SRE連携の実践
maruloop
4
3.9k
Making sense of Google’s agentic dev tools
glaforge
1
140
事業価値を⽣み出すSREへ SREが担うべき意思決定の5層
kenta_hi
2
3.2k
Zoom2Youtube.Claude
kawaguti
PRO
3
490
AI時代の EM への処方箋
staka121
PRO
0
110
なぜ人は自分のプロジェクトを 「なんちゃってアジャイル」と 自嘲するのか
kozotaira
0
280
プロンプト_きのこカンファレンス2026_LT
yurufuwahealer
0
150
Keeping applications secure by evolving OAuth 2.0 and OpenID Connect
ahus1
PRO
1
160
誤解だらけの開発生産性 / Myths and Misconceptions about Developer Productivity
i35_267
1
100
AWS Blocks を触ってみた/first-tach-aws-blocks
fossamagna
2
160
CIで使うClaude
iwatatomoya
0
230
Featured
See All Featured
How to Think Like a Performance Engineer
csswizardry
28
2.7k
Agile that works and the tools we love
rasmusluckow
331
22k
Jess Joyce - The Pitfalls of Following Frameworks
techseoconnect
PRO
1
190
Odyssey Design
rkendrick25
PRO
2
730
Navigating the Design Leadership Dip - Product Design Week Design Leaders+ Conference 2024
apolaine
1
370
VelocityConf: Rendering Performance Case Studies
addyosmani
333
25k
Bash Introduction
62gerente
615
220k
Design in an AI World
tapps
1
260
Getting science done with accelerated Python computing platforms
jacobtomlinson
2
260
New Earth Scene 8
popppiees
3
2.4k
Information Architects: The Missing Link in Design Systems
soysaucechin
0
1k
Large-scale JavaScript Application Architecture
addyosmani
515
110k
Transcript
Ԭٕज़Պֶେֶ ిؾిࢠใֶ՝ఔ ֶ෦ɹ֯ுཽ ࣗવݴޠݚڀࣨ ɹ#̏θϛ ʙୈճʙ ϏοΫσʔλղੳೖᶅ
目次 ˔ͱϞʔϝϯτ ˔͖ͱϞʔϝϯτ ˔͍ҙຯͰͷ͖ ˔ཁ౷ܭྔ ˔౷ܭྔͷਪఆ
分布と統計量 ˔ूஂͷதͷͲͷཁૉબΕΔ֬Λಉ͡ʹ͢Δ ɹແ࡞ҝநग़๏ɺϥϯμϜαϯϓϦϯά ˔ಘΒΕͨඪຊ ɹແ࡞ҝඪຊɺϥϯμϜαϯϓϧ நग़ͨ͠ඪຊͷ࣮ଌʹج͍ͮͯɺ ूஂɺฏۉɺࢄΛਪఆ͢Δɻ
શମ͔ΒภΓͳ͘औΓग़ͨ͠Ұ෦͔ΒશମͷಛੑΛΔ
分布とモーメント ˔Ϟʔϝϯτͱ ɹฏۉࢄͷΑ͏ʹΛಛ͚Δྔ ྫ͑ʜ ਖ਼نฏۉͱࢄ͕༩͑ΒΕΕɺ࠶ݱ͕Մೳ ฏۉࢄͦΕͧΕ̍࣍ɺ̎࣍ͷϞʔϝϯτ ɹฏۉɿ ɹࢄɿ
µ = E[x]= x ⋅ f (x)dx ∫ σ 2 = E[(x −µ)2 ]= (x −µ)2 ⋅ f (x)dx ∫
分布とモーメント ˔Ұൠతͳʹରͯ͠ ɹฏۉࢄΑΓߴ࣍ͷϞʔϝϯτ·Ͱߟ͑Δ͜ͱͰ Λಛ͚Δ ˔֬ʹ͓͍ͯ ɹЋΛத৺ͱͨ͠ҰൠԽ͞ΕͨϞʔϝϯτͷఆٛ
E[(x −α)n ]= (x −α)n ⋅ f (x)dx ∫
べき分布とモーメント ˔ϞʔϝϯτʹΑΔͷಛ͚ͮ ཧ্औΓѻ͍͍͢ ࣮ࡍʹଟ͘ͷ౷ܭख๏ͰҊʹԾఆ͞Ε͍ͯΔ ଟ͘ͷ߹ɺਖ਼نࢦͰ͋Δɻ
Ὃ ݱ࣮ͰɺҟͳΔʹै͏֬ม͕͋Δɻ ͦͷΑ͏ͳ֬มʹै͍ͬͯΔσʔλͰɺ ؍ଌ͞Εͨʹରͯ͠౷ܭख๏͕దͰ͋Δ͔ҙ ͖తͳΛ࣋ͭ
べき分布とモーメント ˔͖ͱʜ ɹɾҝସՁ֨ࠩͷ ɹɾॴಘ͕େ͖͍ྖҬͰͷݸਓॴಘͷ ɹɾจষதͷ୯ޠͷස ࣾձݱɺࣗવݱ
べき分布とモーメント P(≥ x) = Ax−α Լهͷྦྷੵؔʹै͏Λ͖ͱ͍͏ɻ
"ن֨Խఆ ͖͕ͦ͢ް͍͜ͱΛಛͱ͢Δ ʢۃʹେ͖ͳΛ࣋ͭݱ͕ਖ਼نΑΓى͜Γ͍͢ʣ (x ≥ A 1 α ) (1)
べき分布とモーメント ˔͖ͷੑ࣭ ଟ͘ͷখ͞ͳͱগͳ͍ܻҧ͍ʹେ͖ͳΛͱΔ ͷΛؚΉ ਖ਼نΑΓߴ͍֬Ͱܻҧ͍ʹେ͖ͳΛͱΔ ྦྷੵؔΛ྆ରͰϓϩοτ͢ΔͱઢʹͳΔɻ ઢͷ͖͖ࢦЋͰ͋Δɻ
Ћ㱡̎ͰࢄɺЋ㱡̍Ͱฏۉ͕ଘࡏ͠ͳ͍ɻʢЋ࣍ Ҏ্ͷϞʔϝϯτ͕ଘࡏ͠ͳ͍ʣ Ћ̍Ͱɺ࠷େͷγΣΞ͕αϯϓϧ/ˠ㱣Ͱ̌ ʹͳΒͳ͍ɻ ɹ S max = max(x 1 , x 2 ,!, x N ) x i k=1 N ∑
べき分布とモーメント ಛʹࡾͭͷੑ࣭ɺσʔλͷ͕͖ʹै͍ͬͯΔ ͔͔֬ΊΔͨΊʹσʔλղੳʹ͏ɻ σʔλͷྦྷੵΛॻ͘ɻ ྆ରϓϩοτ͠ɺઢͰ͋Δ͜ͱΛ͔֬ΊΔɻ ۙࣅઢΛٻΊɺࢦЋΛٻΊΔɻ
広い意味でのべき分布 ͕ࣜݫີʹΓཱͭ͜ͱݱ࣮ʹͳ͍ɻ Ὃ ͷઈର͕େ͖͍ྖҬͰɺ ͖ؔͰۙࣅͰ͖ΔΑ͏ͳ
͖ͷΛ࣋ͭͱ͍͍ɺ૯͖ͯ͡ͱݺͿɻ ྫʣɾٯΨϯϚ ɹɹɾθʔλ ɹɹଞʹଟ͋Δɻ
要約統計量 ˔ཁ౷ܭྔͱ ɹඪຊͷ࣋ͭੑ࣭Λఆྔతʹಛ͚Δྔ ɾҐஔʹؔ͢Δཁ౷ܭྔ ඪຊฏۉɺதԝ ɾईʹؔ͢Δཁ౷ܭྔ ࢄɺඪ४ภࠩ
統計量の推定 ˔ϏοΫσʔλͷॲཧ ؍ଌ͞Εͨσʔλ͔Βཁ౷ܭྔΛ༻͍ͯɺ ɾͦͷ֬ີؔͷύϥϝʔλΛٻΊΔ ɾσʔλ͕ै͏ํఔࣜͷύϥϝʔλΛٻΊΔ ੳʹΑΓɺूஂ͕࣋ͭະͷύϥϝʔλΛಘΔ ඪຊ͔Βਪఆ͢Δ ਪఆํ๏ʹɺ࠷ਪఆ࠷খೋਪఆ͕͋Δɻ
参考文献 ˔ߴ҆ඒࠤࢠฤஶɺాଜޫଠɾࡾӜߤஶɺ ɹʮֶੜɾٕज़ऀͷͨΊͷϏοΫσʔλղੳೖʯ ʢୈ̏ষʣɺ ɹגࣜձࣾຊධࣾɺ݄