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
カーネル関数とは
Search
kichinosukey
January 14, 2019
Science
65
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
カーネル関数とは
kichinosukey
January 14, 2019
More Decks by kichinosukey
See All by kichinosukey
Latin Hypercube Sampling
kichinosukey
0
84
Lasso.pdf
kichinosukey
0
62
NSGAII
kichinosukey
0
59
NSGA
kichinosukey
0
68
基底関数回帰の概要について
kichinosukey
0
89
児童虐待の現状
kichinosukey
0
74
イノベーションのジレンマとフィンテック
kichinosukey
0
80
Paxos made simpleを理解しよう
kichinosukey
0
600
Other Decks in Science
See All in Science
生成AI・プレプリント時代における 研究成果公開の再設計 ― トップカンファレンス文化はどこへ向かうのか / Redesigning the Dissemination of Research Outputs in the Age of Generative AI and Preprints — Where Is the Top-Conference Culture Heading?
ykiyota
0
30k
JSAI2026企画セッションKS-14 インタビュー集『⼈⼯知能と哲学と四つの問い』が提起する⼈⼯知能のこれからの課題 趣旨説明 / JSAI2026 Special Session: A Collection of Interviews, “Artificial Intelligence, Philosophy, and Four Questions”
ykiyota
0
470
GKE上でオセロの強化学習やってみた
akasan
0
110
知能とはなにか -ヒトとAIのあいだ-
tagtag
PRO
0
230
[Relay webinaire eBIS actu n°16] Conclusion
institutdelelevage
PRO
0
210
機械学習 - ニューラルネットワーク入門
trycycle
PRO
0
1.4k
自社のレビュー履歴からAIコードレビュアーをつくる方法
estie
5
1.8k
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
1.8k
Toward Causal Scientific Discovery with AI
sshimizu2006
0
220
機械学習 - DBSCAN
trycycle
PRO
0
2.2k
AI for Phage-Host prediction
michielstock
0
140
生成AI時代のインタラクションサイエンス(招待講演)/interaction-science
okana2ki
0
110
Featured
See All Featured
Become a Pro
speakerdeck
PRO
31
6.3k
Stop Working from a Prison Cell
hatefulcrawdad
274
21k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
25
2.1k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
287
14k
HDC tutorial
michielstock
2
960
Marketing to machines
jonoalderson
1
5.9k
SEO in 2025: How to Prepare for the Future of Search
ipullrank
3
3.9k
Designing for humans not robots
tammielis
254
26k
Test your architecture with Archunit
thirion
2
2.4k
Chasing Engaging Ingredients in Design
codingconduct
0
360
Building a Scalable Design System with Sketch
lauravandoore
464
34k
Ethics towards AI in product and experience design
skipperchong
2
430
Transcript
Χʔωϧؔͱ
ࢀߟॻ੶ w ΄΅ࣸܦ͍͖ͤͯͨͩ͞·͠ ͨ w ͱͯΘ͔Γқ͍༰Ͱײँ Ͱ͢ w IUUQTXXXJXBOBNJDPKQ CPPLCIUNM
ಛநग़ʹ͍ͭͯ ↟ᏣͺͼΝᤎ᷀ᔻṟϧόϰͽͶ͖ວ͚Ν ↟ͥΞᬿᔻ᧣ͼؔΤᬤͶᦉᷯͩ͝ϧόϰặ͟ͼ͖ ↟ΚΜີᮈၞḄᙚΤᑿͥ͘ͺΤວ͚Νͺ ↟EᮈᡯfᓓᙚᑤᔻṟͲ͞fϖϮϦ⑲τ⑲XͽͶ͖ᔻṟ y = wTx f(x) =
d ∑ m=1 wm xm ⋯ (1)
ಛநग़ʹ͍ͭͯ ↟ጽͽYͽ᷁͝ᨷΐ͵ͱᑤᔻṟมỐΤᙇͩີᮈͽࣸᓕͫΝͥͺΤ|ᴰ ᱢᲇ}ͺṺΈ ↟ᴰϞδύϰΤᤎ᷀Κ͘ͽᔉ͡ͺͫΝ ↟ͫΝͺ ᙚᴰϞδύϰᤎ᷀ͺͼΝ ඇઢܗม ೖྗY ಛϕΫτϧП
Y ϕ(x) = (ϕ1 (x), ⋯, ϕd (x))T ϕ(x) = (x, x2, x3, ⋯, xd)T
ಛநग़ʹ͍ͭͯ ↟ᴰᱢᲇͧΞͱ໖ණͽ͖͜fᔻṟϧόϰΤວ͚Νͺ ↟ͥວ͚ፖᔟ͝Λϖτ⑲ϸ᤹ᙣᐠᛯ᤹ᬰͧΞ͟ͱ ↟ᴰᱢᲇϦϯψύΐͺΓΝͺᤎ᷀Ͷ ↟ᏝࡶͼؔؔΤ࣮ḿ͟Ν ↟ᡯ᥍ᅪY࣮͞Ϟδύϰ͔Νᶶ᠓͞ͼ͖ ↟ᴰϞδύϰ࣮͞ϞδύϰͼͱΓY͞የᥙᅾΖεϮϚཋ᪦Δᅔ͖ fw (x) =
wTϕ(x) = d ∑ m=1 wm ϕm (x) ⋯ (2)
Χʔωϧؔͷఆٛ ↟Y Ysͺ͖͘ᤍͶมͽରͩΰ⑲ϒϰؔL Y Ys ͯΞͰΞ ᧤Ẹͺͩᤎ᷀Κ͘ͽᨷᤅͧΞΝ ↟ ᙚĘมၞḄᙚfᓓᙚΤͽᤎ᷀Κ͘ͽဒᘥ͟Ν
k(x, x′) = ϕ(x)Tϕ(x′) = d ∑ m=1 ϕm (x)ϕm (x′) ⋯ (3) k(x, x′) = d ∑ m=1 xm(x′)m ⋯ (4)
֤छͷΧʔωϧಋೖʹ͍ͭͯ ↟᧤ ↟ᩆᨷᲩᕥ ↟ệᖰᐠᵿď࿆ỹᕭᅊĐ
ੵͷΧʔωϧදݱ ↟ȃ Y ͺϖϮϦ⑲τ᧤͔Ν ᙚfᚌᐠၞ͡Y Y hhhΤ᧱ ͽᕅΈͺᤎ᷀ṟ࿄ᑾͫΝͥͺ͟͞Ν ↟ΰ⑲ϒϰؔᴰϞδύϰ᧤fଈͳᤎ᷀ᨷᤅͧΞΝ
↟ͥΞfϖϮϦ⑲τ⑲XΤᤎ᷀ṟͽᷟᨷ͟ΝͥͺΤዳͫΝ f(x) = ∑ i αi k(xi , x) ⋯ (5) f(x) = ∑ i αi ϕ(xi )Tϕ(x) ⋯ (6) w = ∑ i αi ϕ(xi )
ੵͷΧʔωϧදݱ ↟ ᙚ|ᚌᐠၞ͡}ᨨY@Jΰ⑲ϒϰؔᔻṟẸͩ͝ͼ͖ ↟κϸϜϰᨨᔻṟẸṟᔉ͡ͱΓͽĦ ↟͔ΝᴰᎭͼᩆ᳀ặΤḝͥ͘ͺfᤎ᷀Κ͘ͼဒᘥ͞൱ၙͺͼΝ ↟ϯϜϱρϸτ⑲ᨷሑďᕭᅊĐf-ϓϰϥΤᑿ͘ ↟OEᦉᷯfXႚΡΜͽEΤᠭ͖ΝͥͺϖϮϦ⑲τΤවΛͫͥͺ͟͞Ν ↟FY ᮈ༜ṹᥭᠭḄࠐΑၞḄᙚΘ͘ͽᮈᡯ͞ᯭͺͼΝͱΓfޮໝΤൃἿͫΝ
f(x) = n ∑ i=1 αi k(x(i), x) = n ∑ i=1 αi d ∑ m=1 (x(i))mxm ⋯ (7)
ਖ਼ఆੑ͔ΒͷΧʔωϧؔ ಋೖ ↟᧤ᕥ᭳͝Λfΰ⑲ϒϰؔᴰᅗͱͺ͟YͺYsᇮᑾ ႷΤᶡ͖ͩΝͺວ͚ΛΞΝ ↟ჯͪፖḟΤḟ͖͖Νͺ͟ͽႝ͟͡ͼΜfᬿ༜ͫΝͺͽͼΝ ↟ᝂͽYͺYsᇮᑾႷΤᶡͫΚ͘ͼؔL Y Ys ͔͞Νͺ͟ͽfͯΞΤ ΰ⑲ϒϰؔͺͩᑿ͚ͼ͖͝Ħ
↟|ᩆᨷᲩᕥ}ΤຬͱͭZFT
ਖ਼ఆੑ ↟͔ΝؔL Y Ys ͞ᩆᨷᲩ͔Νͺf᥅O෪ᨨY@ hhh Y@O ͝ΛຌᒿͧΞΝᤎ᷀ḝᅾďεϮϥḝᅾĐ͔͞Ν ↟ͥΞͽͶ͖ęᮈṟᙚ͞ᓝͽᑤᐔ͔Νͺ͟fᤎ᷀ᙚ͞᥅O ᮈϞδύϰǯ
ǯ@ hhh ǯ@O ?5ͽͶ͖ᕧΜርͶͥͺ͔Νďᩆͩ ͡ጪᩆᨷᲩᕥĐ K = k(x1 , x1 ) k(x2 , x1 ) ⋯ k(xn , x1 ) k(x1 , x2 ) k(x2 , x2 ) ⋯ k(xn , x2 ) ⋮ ⋮ ⋱ ⋮ k(x1 , xn ) k(x2 , xn ) ⋯ k(xn , xn ) n ∑ i=1 n ∑ j=1 αi αj Kij ≥ 0 ⋯ (8)
ΧʔωϧτϦοΫ ↟طͽᩆᨷᬰᕥΤຬͱͫͥͺ͞Ρ͝͵͖ΝؔΤᑿ͚f᧤ຌᒿ͞Ᏽ᠓ͽͼΝ Ħďᤎ᷀ἦᥥζ⑲ξͽͶ͖Đ ↟ᴰϞδύϰ᧤ͺͩᨷᤅͩͱΰ⑲ϒϰ|ᩆᨷᲩ}͔Ν ↟ᝂͽf᥅ᩧᨷᲩؔL Y Ys ͞ରশ L Y
Ys L Ys Y ͔ΞͯΞ|ᴰϞδύϰ ᧤}͔Ν ↟ͥΞᦉᷯͽΚ͵ᷟᮈᡯᴰϞδύϰ᧤͞ඦ୯ͼؔͽͼΝᦉᷯ͞ ͔Ν ↟ͶΐΜfᩆᨷᲩᕥΤΰ⑲ϒϰᨷᤅͺͫΞຌᒿᅗΤවΛͭΝͥͺ͟͞Ν ↟ͥΞΤ|ΰ⑲ϒϰύϯψδ}ͺ͖͘
ΨεΧʔωϧ ↟༺ମᆚ ↟ᩆྥᐠᵿጓႷؔͺჯͪṟ ↟αΫξΰ⑲ϒϰͺṺΈͥͺΔ͔Ν ↟ͥΞᴰϞδύϰණฌሜ᧤ͺͩᶡḿ͟Ν ↟ᩆᨷᲩᕥΔຬͱͫďᕭᅊĐ k(x, x′) = exp(−β∥x
− x′∥2) ⋯ (9)
ଟ߲ࣜΧʔωϧ ↟αΫξΰ⑲ϒϰͺჯ༷ͽfΚ͡ᑿΡΞΝ ↟Qᥦ D∕ ↟ፖfͥᙚΤᤎ᷀ṟͽ᧾ͫΝͺE?Qί⑲υ⑲ಠርͼḄẸ ͺͩᔉͣΝ k(x, x′) = (xT
x′+ c)P ⋯ (10) k(x, x′) = p ∑ m=0 ( p m) { d ∑ l=1 (xl x′ l ) } m cp−m ⋯ (11)
֤Χʔωϧͷಛ w ϊΠζΛͤͨਖ਼ݭʹର͢ΔϑΟο ςΟϯάΛ֤ΧʔωϧͰ࣮ࢪ w ্ஈͰY< >ͷൣғͷϑΟοςΟ ϯάΛ֬ೝ͍ͯ͠Δ͕େ͖ͳࠩҟ ݟΒΕͳ͍ w
͔͠͠ɺԼஈͰֶशσʔλͷൣ ғ֎Λ༧ଌͤ͞Δͱɺଟ߲ࣜൃ ࢄํɺ3#'ऩଋ͢Δํ ͱ͔͏ w IUUQTHJTUHJUIVCDPN LJDIJOPTVLFZ