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
DeepLearningBook 9.3-9.4
Search
mtjuney
January 16, 2018
Technology
0
47
DeepLearningBook 9.3-9.4
9.3 Pooling
9.4 Convolution and Pooling as an Infinitely Strong Prior
mtjuney
January 16, 2018
Tweet
Share
More Decks by mtjuney
See All by mtjuney
DeepLearningBook 9.5-9.7
mtjuney
0
44
Other Decks in Technology
See All in Technology
AI時代におけるドメイン駆動設計 入門 / Introduction to Domain-Driven Design in the AI Era
fendo181
0
680
[mercari GEARS 2025] Keynote
mercari
PRO
0
210
從裝潢設計圖到 Home Assistant:打造智慧家庭的實戰與踩坑筆記
kewang
0
160
Quarkusで作るInteractive Stream Application
joker1007
0
130
ユーザーストーリー x AI / User Stories x AI
oomatomo
0
190
Lazy Constant - finalフィールドの遅延初期化
skrb
0
180
AWS資格は取ったけどIAMロールを腹落ちできてなかったので、年内に整理してみた
hiro_eng_
0
210
ある編集者のこれまでとこれから —— 開発者コミュニティと歩んだ四半世紀
inao
4
2.4k
Black Hat USA 2025 Recap ~ クラウドセキュリティ編 ~
kyohmizu
0
530
[CV勉強会@関東 ICCV2025] WoTE: End-to-End Driving with Online Trajectory Evaluation via BEV World Model
shinkyoto
0
190
『HOWはWHY WHATで判断せよ』 〜『ドメイン駆動設計をはじめよう』の読了報告と、本質への探求〜
panda728
PRO
5
1.7k
こんな時代だからこそ! 想定しておきたいアクセスキー漏洩後のムーブ
takuyay0ne
4
570
Featured
See All Featured
Raft: Consensus for Rubyists
vanstee
140
7.2k
Chrome DevTools: State of the Union 2024 - Debugging React & Beyond
addyosmani
9
970
Mobile First: as difficult as doing things right
swwweet
225
10k
Improving Core Web Vitals using Speculation Rules API
sergeychernyshev
21
1.2k
Building a Scalable Design System with Sketch
lauravandoore
463
33k
Designing for humans not robots
tammielis
254
26k
The Straight Up "How To Draw Better" Workshop
denniskardys
239
140k
Build The Right Thing And Hit Your Dates
maggiecrowley
38
2.9k
How to train your dragon (web standard)
notwaldorf
97
6.4k
The Web Performance Landscape in 2024 [PerfNow 2024]
tammyeverts
11
930
Fireside Chat
paigeccino
41
3.7k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
367
27k
Transcript
Deep Learning Book 9.3-9.4 mtjuney
• ը૾ೝࣝʹ͓͍ͯ ը૾ͷେ͖͖͕͞มΘͬͨͱ͖Ͱಉ͡ग़ྗ͕ग़ͯཉ͍͠ • ࣮ࡍมΘͬͯ͠·͏ • ͦ͜ͰPooling 9.3 Pooling Apple
• Pooling • ग़ྗΛԿΒ͔ͷ౷ܭྔͰஔ͖͑Δ • ౷ܭྔ: ࠷େɺฏۉɺL2ϊϧϜ etc 9.3 Pooling
{ { { ࠷େ ࠷େ ࠷େ ྫ: ࠷େPooling
9.3 Pooling • ೖྗͷখ͞ͳҠಈʹରͯ͠ෆมʹͳΔ • খ͞ͳεέʔϦϯάɺճసɺҠಈʹରͯ͠ؤ݈ • ग़ྗͷϢχοτ͕ݮ͞ΕΔ • ܭࢉྔݮ
{ { { • ՄมͷೖྗʹରԠͰ͖Δ • ՄมΛݻఆʹམͱ͠ࠐΉ 9.3 Pooling ࠷େ
࠷େ ࠷େ
9.3 Pooling • ՄมͷೖྗʹରԠͰ͖Δ • Global Average Pooling[Lin 2013] •
νϟωϧຖʹฏۉΛग़͢ • શ݁߹ͷΘΓʹΘΕͨΓ͢Δ
• Maxout Network [Goodfellow 2013] • νϟωϧ୯ҐͰͷPooling • ҟͳΔಛʹԠ͢ΔϑΟϧλͷ͏ͪɺ ͲΕ͔ҰͭͰԠ͢Εग़ྗԠ͢Δ
9.3 Pooling
• Infinitely Strong Priorʢແݶʹڧ͍ࣄલʣ • ύϥϝʔλ͕͍͔ͭ͘ͷʹͳΔ֬Λ0ʹͯ͠ શʹ༻ېࢭͱ͢Δ • ΈࠐΈωοτϫʔΫશ݁߹ωοτϫʔΫʹ ແݶʹڧ͍ࣄલΛ༩͑ͨͷͱྨࣅ
• ۙҎ֎ͷॏΈ͕0 • ॏΈΛڞ༗ • Poolingಉ༷ • ೖྗ͕খ͞ͳҠಈʹରͯ͠ෆม 9.4 Convolution and Pooling as an Infinitely Strong Prior
• ͜ΕΒͷࣄલԾఆ͕ଥͰ͋Ε༗ޮ • ଥͰͳ͍߹ɺUnderfittingͷݪҼͱͳΔ • Pooling • ͋Δಛ͕ग़Δݫີͳॴ͕Γ͍ͨ߹ٯޮՌ • GANͷֶशͳͲͷಛఆͷλεΫʹٯޮՌ
• ΈࠐΈ • ΕͨॴͷใΛ͍͍ͨ߹ɺٯޮՌ 9.4 Convolution and Pooling as an Infinitely Strong Prior
• ΈࠐΈϞσϧΈࠐΈϞσϧͷΈͱൺֱ͞ΕΔ͖ • ΈࠐΈϞσϧ • ઃܭऀ͕ۭؒతࣝΛΈࠐΜͰ͍Δ • ΈࠐΈͰͳ͍Ϟσϧ • ۭؒతࣝΛֶशͰ֫ಘ͢Δඞཁ͕͋Δ
• ͜ΕΒΛಉ͡ϕϯνϚʔΫͰൺֱ͢Δ͖Ͱͳ͍ 9.4 Convolution and Pooling as an Infinitely Strong Prior