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
iclr2020deepsemi-supervisedanomalydetectionyamatookamoto-200531022507.pdf
Search
Yamato.OKAMOTO
June 14, 2020
Technology
0
47
iclr2020deepsemi-supervisedanomalydetectionyamatookamoto-200531022507.pdf
Yamato.OKAMOTO
June 14, 2020
Tweet
Share
More Decks by Yamato.OKAMOTO
See All by Yamato.OKAMOTO
いまAI組織が求める企画開発エンジニアとは?
roadroller
1
980
Slide ICCV2023 Constructing Image Text Pair Dataset from Books
roadroller
0
16
第11回 全日本コンピュータビジョン勉強会 CVPR2022 "A Self-Supervised Descriptor for Image Copy Detection"
roadroller
0
430
第9回 全日本コンピュータビジョン勉強会 発表資料
roadroller
0
520
第七回全日本コンピュータビジョン勉強会 A Multiplexed Network for End-to-End, Multilingual OCR
roadroller
1
850
部下のマネジメントはAI開発に学べ
roadroller
0
79
Domain Generalization via Model-Agnostic Learning of Semantic Features NeurIPS’19 読み会 in 京都
roadroller
0
150
ICML’2019 読み会in京都 Federated Learningの研究動向
roadroller
0
33
CVPR2019@Long Beach 参加速報(本会議)
roadroller
0
64
Other Decks in Technology
See All in Technology
Four Keysだけじゃ足りなくない? 〜俺たちだけのFour Keysを探して〜
rinchsan
2
2.6k
プロダクト拡大フェーズでプロダクト検証サイクル効率化を目指す過程で見えたもの / Streamlining Product Validation in Growth Phase
kakehashi
5
5.6k
Bedrockに触ってみよう - 【#Bedrock開発入門 出版記念】AWSで生成AIアプリ開発をはじめよう!
moritalous
0
190
Oracle Modern Data Platform Reference Architecture (MySQL HeatWave Lakehouse編)
oracle4engineer
PRO
2
290
5W1H ~LLM活用プロジェクトを推進するうえで考えるべきこと~
natsuume
0
130
プラットフォーム開発の実例と撤退から学ぶ / Learning from examples of platform development and withdrawal
kaminashi
2
430
Lernen durch Schmerzen! Mit Reinforcement Learning selbstlernende Systeme entwickeln
joergneumann
0
220
現場の失敗から学ぶ!プロダクトバックログアイテムの改善/Learn_from_On-Site_Failures!_Improving_Product_Backlog_Items
m_iyama
3
920
Bring your app’s core features to users with App Intents とか App Intents 関連の要約
ryomm
1
260
IaCツールのいろいろ
takesection
0
220
えにしテックさん15周年に寄せて〜万葉と私のこれまでの学び〜
nay3
44
27k
AWSの生成AIサービス入門 & 最強の書籍紹介!
minorun365
PRO
5
600
Featured
See All Featured
BBQ
matthewcrist
80
8.9k
How To Stay Up To Date on Web Technology
chriscoyier
784
250k
Designing for Performance
lara
603
67k
[RailsConf 2023] Rails as a piece of cake
palkan
31
4.3k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
34
6.2k
XXLCSS - How to scale CSS and keep your sanity
sugarenia
244
1.2M
RailsConf 2023
tenderlove
12
670
A better future with KSS
kneath
231
17k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
14
1.2k
"I'm Feeling Lucky" - Building Great Search Experiences for Today's Users (#IAC19)
danielanewman
222
21k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
275
13k
Learning to Love Humans: Emotional Interface Design
aarron
269
39k
Transcript
2020/6/14 Yamato OKAMOTO ICLRΦϯϥΠϯಡΈձ Deep Semi-supervised Anomaly Detection
ࣗݾհʢ͘!!ʣ ɹԬຊେʢ͓͔ͱ·ͱʣ • ژେֶඒೱݚڀࣨͰύλʔϯೝࣝΛݚڀͯ͠म࢜՝ఔमྃ • ΦϜϩϯͰ৽نࣄۀΛܦݧޙɺ͍·ࣾձγεςϜࣄۀ෦ͷݚڀॴϦʔμʔ • ເژΛϙετɾγϦίϯόϨʔʹ͢Δ͜ͱɺؔͷίϛϡχςΟΛڧԽ͍ͨ͠ ɹ㱺 ژͷมਓύϫʔΛੈքʹΒ͠Ί͍ͨ
Twitter : RoadRoller_DESU ҆৺҆શͳࣾձͷ࣮ݱʹ͚ͯɺ ࠷ۙ Anomaly Detection ʹڵຯΞϦ
Anomaly Detection ͋Δ͋Δ ఆٛࠔ • ҟৗʹ༷ʑͳόϦΤʔγϣϯ͕͋Δ • ҟৗݕग़͍͚ͨ͠ͲʮWhat is ҟৗʁʯ͕ఆٛͰ͖ͳ͍
ֶशσʔλ͕ೖखࠔ • ҟৗ໓ଟʹൃੜ͠ͳ͍ʢ※ සൟʹൃੜ͢ΔΠϕϯτҟৗ͡Όͳ༷ͯ͘ʣ • ѹతʹҟৗσʔλ͕ෆͯ͠ػցֶश͕ࠔ ैདྷख๏ɿਖ਼ৗΛఆٛ͢Δ • ʮWhat is ҟৗʁʯͷఆٛΛఘΊΔɺҟৗσʔλͷֶशఘΊΔ • ͦͷΘΓʮWhat is ਖ਼ৗʁʯͷఆٛΛֶशͯ͠ɺʮNot ਖ਼ৗʯΛҟৗͱఆ͢Δ
Anomaly Detection ͷैདྷݚڀ Deep One-Class Classification (ICML’18) • ਖ਼ৗσʔλͷΈΛ༻͍ͯɺClassifierͳΓAutoEncoderͳΓΛैདྷ௨Γʹֶश •
͜ͷͱ͖ɺಛྔ͕࣍ݩ෦ۭؒʹऩଋ͢ΔΑ͏LOSSΛՃ͑Δ • ਖ਼ৗσʔλͳΒٿʹ͢ΔͣͳͷͰɺٿ͔Β֎ΕͨσʔλΛҟৗͱఆ͢Δ ୈҰ߲ʹΑͬͯٿʹ͕ԡ͠ࠐ·ΕΔ cɿ ٿͷத৺ʢͨͩ͠≠0ʣ nɿֶश͢Δਖ਼ৗσʔλͷ
Anomaly Detection ͷධՁ؍ ͲΕ͚ͩਖ਼֬ʹҟৗΛݕͰ͖͔ͨʁ • ਖ਼ৗσʔλΛਖ਼ৗͱఆͯ͠ɺҟৗσʔλΛҟৗͱఆ͢Δਫ਼ ԼྲྀλεΫΛअຐ͠ͳ͍͔ʁ • ԼྲྀλεΫ͕͋Δ߹ɺҟৗݕػೳͷՃʹΑͬͯѱӨڹ͕ͳ͍͔Ͳ͏͔ •
ྫ͑ɺ10ΫϥεͷࣈࣝผثʹɺਤܗͳͲࣈҎ֎͕ೖྗ͞Εͨͱ͖ҟৗͱఆ͢Δػ ೳΛ͚Ճ͍͑ͨͤͰɺैདྷͷ10Ϋϥεࣝผੑೳ͕Լ͢ΔͱࠔΔ ad-hoc͔post-hoc͔ʁ • ҟৗݕ͢ΔͨΊʹϞσϧߏֶशํ๏·Ͱม͑Δඞཁ͕͋Δ͔ʁ • ·ͨɺLOSSΛޙ͔Β͚͚̍ͭͩͯ͠Ճֶश͢Δ͚ͩͰOK͔ʁ • ͲͪΒ͕ྑ͍ѱ͍ͳͲҰ֓ʹݴ͑ͳ͍͕ɺpost-hocͷํ͕ѻ͍͍͢ɻ
հจͷ֓ཁ ʮSemi-supervisedʹֶश͠Α͏ʂʯ Anomaly Detection ͷݚڀUnsupervised͕ओྲྀͷΑ͏ͩ Ͱɺֶश༻ͷҟৗσʔλ͕ೖखࠔͩͱͯ͠ɺ ӡ༻Λଓ͚ͯͨΒҟৗσʔλʹ͍ͣΕग़ձ͏ͣ ͳΒɺͦΕΒগྔͷҟৗσʔλΛͬͯɺ Semi-supervisedʹֶशͨ͠ํ͕ྑ͍ͷͰʁ ※Semi-supervisedͷAnomaly
Detectionݚڀඇৗʹগͳ͍
ఏҊख๏ ʮLOSSʹ߲Λ̍ͭՃ͠·ͨ͠ʯ Deep One-Class Classification (ICML’18) ͷLOSSʹSemi-supervisedͷ߲Λ̍ͭՃ • ࣮ಉ͡ஶऀͰͨ͠ɻࣗͷݚڀΛࣗͰΞοϓσʔτͨ͠ܗʹͳΔɻ ͠ҟৗσʔλʹग़ձͬͨΒɺ
ٿͷ֎ଆʹߦ͘Α͏ֶश͢Δ mɿsemi-supervisedʹֶश͢Δσʔλ yj ɿਖ਼ৗorҟৗͷϥϕϧ
࣮ݧ݁Ռ ॎ࣠ɿҟৗσʔλͷݕग़ੑೳ ʢHigher is Betterʣ Unsupervised Semi-supervised ԣ࣠ɿSemi-supervisedͰڭࢣ͖ͷҟৗσʔλΛֶशׂͨ͠߹ ఏҊख๏ MNISTɺFashion-MNISTɺCIFAR-10ͷσʔληοτͰධՁ
• ̍Ϋϥεͱਖ਼ৗͱఆٛͯ͠ɺAutoEncoderʴఏҊख๏ͰಛྔදݱΛֶश • Γͷ̕ΫϥεΛೖྗͨ͠ͱ͖ɺҟৗͱఆͰ͖Δ͔Ͳ͏͔ධՁ ੑೳվળΛ֬ೝ
·ͱΊͱߟ ਂֶशʹΑΔ Semi-supervised ͳ Anomaly Detection ख๏ΛఏҊ • ॳΊͯͰͳ͍ͱࢥ͏͕ɺਂֶशʹΑΔAnomaly DetectionͰsemi-supervised͍͠
• ͔ͨ͠ʹࣾձ࣮Λߟ͑Δͱɺ͜ͷઃఆద • ख๏γϯϓϧͰɺpost-hocͳͷͰѻ͍͍͢ • ࠓճԼྲྀλεΫ͕AE͕ͩͬͨɺClassificationͩͱͲ͏ͳΔ͔ʁ • Anomaly DetectionͷධՁσʔληοτͬͯଞʹͳ͍ͷ͔ͳɺɺɺɺ ʢ͍ͭ·ͰMNISTʹΑΔධՁ͕ଓ͘ͷͩΖ͏͔ʣ