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
機械学習ゼミ_正則化.pdf
Search
Sponsored
·
Your Podcast. Everywhere. Effortlessly.
Share. Educate. Inspire. Entertain. You do you. We'll handle the rest.
→
RyoMaekura
August 15, 2018
21
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
機械学習ゼミ_正則化.pdf
RyoMaekura
August 15, 2018
More Decks by RyoMaekura
See All by RyoMaekura
データサイエンス初心者講座_Kaggle入門.pdf
ryomaekura
0
54
HANDAI-JK_LT
ryomaekura
0
48
Featured
See All Featured
Why You Should Never Use an ORM
jnunemaker
PRO
61
10k
Fashionably flexible responsive web design (full day workshop)
malarkey
409
67k
Data-driven link building: lessons from a $708K investment (BrightonSEO talk)
szymonslowik
1
1.4k
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
Joys of Absence: A Defence of Solitary Play
codingconduct
1
550
Neural Spatial Audio Processing for Sound Field Analysis and Control
skoyamalab
0
570
Why Your Marketing Sucks and What You Can Do About It - Sophie Logan
marketingsoph
0
430
Rails Girls Zürich Keynote
gr2m
96
14k
We Are The Robots
honzajavorek
0
400
Raft: Consensus for Rubyists
vanstee
142
7.8k
Code Review Best Practice
trishagee
74
20k
Documentation Writing (for coders)
carmenintech
77
5.6k
Transcript
//ͷਖ਼ଇԽ ੑཧՊֶίʔεɹલଂྒྷ
ਖ਼ଇԽͱ ֶशΞϧΰϦζϜΛमਖ਼͢Δ͜ͱͰ ܇࿅ޡࠩͰͳ͘൚Խޡ͕ࠩͰ͖Δ ݶΓ࠷খԽ͞ΕΔΑ͏ʹ͢Δ͜ͱ ܇࿅ޡࠩ ฏۉೋޡࠩͷ߹ E(w) = 1 N
N ∑ n=1 1 2 ( ̂ y(xn ; w) − yn )2 ൚Խޡࠩ ฏۉೋޡࠩͷ߹ Egen (w) = Edata ( 1 2 ( ̂ y(x; w) − y)2 ܇࿅ޡࠩͷ࠷খԽաֶशͷڪΕ
ਖ਼ଇԽͱ ൚Խޡࠩશͯͷσʔλʹରͯ͠Ͳͷ΄Ͳͷਫ਼͕͋Δͷ͔ΛධՁ͢Δࣜ ѻ͑Δσʔλ༗ݶͳͷͰݫີͳ൚ԽޡࠩΛΔ͜ͱෆՄೳ ൚Խޡࠩςετ༻σʔλͰ༻ ςετޡࠩ ฏۉೋޡࠩͷ߹ Etest (w) = 1
Ntest N ∑ m=1 1 2 ( ̂ y(xm ; w) − ym )2
ਖ਼ଇԽͷํ๏ ɾॏΈݮਰ ɾ&BSMZ4UPQQJOH
ॏΈݮਰ ޮՌ ɾޡࠩؔʹਖ਼ଇԽΛՃ͑աֶशΛ͙ ɾ-ਖ਼ଇԽ -BTTPਖ਼ଇԽ ͱ-ਖ਼ଇԽ 3JEHFਖ਼ଇԽ ɾਖ਼ଇԽ߲ʹ-ͱ-ͷ྆ํΛ͢߹͋Δ ɾϩδεςΟοΫճؼ //
47.ͳͲͷ໘Ͱ༗ޮͰ͋Δ -ਖ਼ଇԽ ɾॏΈ͕େ͖͍΄Ͳʹ͍͍ۙͮͯ͘Α͏ʹॏΈΛߋ৽͢Δಇ͖͕͋Δ ɾଟ//ͰॏΈ͕ൃࢄ͍͢͠ͷͰ Α͘ΘΕΔ
ॏΈݮਰ ޡࠩؔΛ͋Δ࠷খۙʹͯ࣍ؔ͠Ͱۙࣅ͢Δ·ͨ γΞϯ ର֯ߦྻͰ͋ΔͱԾఆͯ͠ਐΊΔ w0 E(w) ≃ E0 + 1
2 ∑ i hi (wi − w0 i )2 ͜͜Ͱ ॏΈݮਰͷ߲ΛՃ͑ͨ৽ͨͳޡࠩؔΛఆٛ͢Δ Ereg (w) = E(w) + α w2 2 ࣍ ͜ͷޡࠩؔͷඍ͕ʹͳΔ͕࠷খͱͳΔ
ॏΈݮਰ ਖ਼ଇԽޙͷ৽ͨͳ࠷খ w* i = hi hi + α w0
i = 1 1 + α hi w0 i ЋͱൺͯγΞϯͷର͕֯ۃΊͯখ͍͞ ߹ hi ≪ α w* i = 1 1 + α hi w0 i ≃ 0 ͜ΕʹΑΓॏΈݮਰͷޮՌΛࣔͨ͠
&BSMZ4UPQQJOH ޮՌ ɾ࣮͕ඇৗʹ؆୯ ɾਖ਼ଇԽͷޮՌ͕ߴ͍ ɾཧ୯७ ํ๏ աڈͷΤϙοΫͰߋ৽ͨ͠ॏΈΛอଘ ςετޡࠩͷ૿Ճ͕࿈ଓͨ͠߹ ֶशΛఀࢭ ςετޡ͕ࠩݮগ͔Β૿Ճʹసͨ͡ΤϙοΫ·ͰḪΔ
ͦͷ࣌ͰͷύϥϝʔλΛ࠾༻
&BSMZ4UPQQJOH
ࢀߟจݙ 13.-্ר χϡʔϥϧωοτϫʔΫͷਖ਼ଇԽ ਂֶशೖ ਂֶशͷਖ਼ଇԽ