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
Variational Auto Encoderでの Disentangled表現
Search
Kosuke Miyoshi
November 03, 2016
Research
650
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Variational Auto Encoderでの Disentangled表現
※ Speaker Deck内ではURLリンクがクリックできないので、クリックしたい場合はpdfをダウンロードしてください
Kosuke Miyoshi
November 03, 2016
More Decks by Kosuke Miyoshi
See All by Kosuke Miyoshi
Representation Learning with Contrastive Predictive Coding
miyosuda
1
240
Sutton "Reinforcement Learning" 2nd Edition Ch13: Policy Gradient Methods
miyosuda
0
230
Sutton "Reinforcement Learning" 2nd Edition Ch7: n-step Bootstrapping
miyosuda
0
120
Sutton "Reinforcement Learning" 2nd Edition Ch6: TD-learning
miyosuda
0
120
SCAN
miyosuda
0
900
TensorFlow & DeepMind Lab & UNREAL
miyosuda
1
2.6k
Other Decks in Research
See All in Research
コーディングエージェントとABNを再考
hf149
2
830
260624_NLP-colloquium: Hubness
de9uch1
1
170
Cross-Media Information Spaces and Architectures
signer
PRO
0
330
IA for theory
gpeyre
1
340
Unified Audio Source Separation (Defense Slides)
kohei_1979
1
640
Geometric calculations on probability manifolds from reciprocal relations in Master equations
lwc2017
0
110
Anthropic が提案する LLM の内部状態を自然言語で説明可能にした Natural Language Autoencoders / Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
shunk031
0
160
第12回人と環境にやさしい交通をめざす全国大会/熊本都市圏「車1割削減、渋滞半減、公共交通2倍」をめざして
trafficbrain
0
140
Data Visualization Tools in the Age of AI
flekschas
0
180
【中間報告】国会議員の立法・政策実務を支える環境を巡る現状と課題
polipoli
0
420
Sleuthcon Keynote - How Cybercriminals (ab)use AI
fr0gger
0
290
ScoreMatchingRiesz for Automatic Debiased Machine Learning and Policy Path Estimation with an Application to Japanese Monetary Policy Evaluation
masakat0
0
310
Featured
See All Featured
B2B Lead Gen: Tactics, Traps & Triumph
marketingsoph
0
200
Bioeconomy Workshop: Dr. Julius Ecuru, Opportunities for a Bioeconomy in West Africa
akademiya2063
PRO
1
210
Marketing Yourself as an Engineer | Alaka | Gurzu
gurzu
0
270
Exploring the relationship between traditional SERPs and Gen AI search
raygrieselhuber
PRO
2
4.2k
Avoiding the “Bad Training, Faster” Trap in the Age of AI
tmiket
0
200
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
25
2k
What Being in a Rock Band Can Teach Us About Real World SEO
427marketing
0
1.1k
Unlocking the hidden potential of vector embeddings in international SEO
frankvandijk
0
890
Leading Effective Engineering Teams in the AI Era
addyosmani
9
2.3k
Code Review Best Practice
trishagee
74
20k
[RailsConf 2023] Rails as a piece of cake
palkan
59
6.9k
Data-driven link building: lessons from a $708K investment (BrightonSEO talk)
szymonslowik
1
1.2k
Transcript
7BSJBUJPOBM"VUP&ODPEFSͰͷ %JTFOUBOHMFEදݱ OBSSBUJWFOJHIUTגࣜձࣾࡾ߁༞ શΞʔΩςΫνϟୈճ4*(8#"Φϑձ
&BSMZ7JTVBM$PODFQU-FBSOJOHXJUI 6OTVQFSWJTFE%FFQ-FBSOJOH Irina Higgins, Loic Matthey, Xavier Glorot, Arka Pal,
Benigno Uria, Charles Blundell, Shakir Mohamed, Alexander Lerchner Google DeepMind, 2016 ͱΓ͋͛Δจ ˞όʔδϣϯʹΑͬͯύϥϝʔλ͕େ͖͘ҧ͍ͬͯΔɻ࠷৽ͷWΛར༻
%JTFOUBOHMFE SFQSFTFOUBUJPO w ҙຯۭؒͷදݱΛɺҐஔɺαΠζɺFUDʜͳͲͷߏ ཁૉͷύϥϝʔλʹ͔Εͨঢ়ଶͰ֫ಘ͍ͨ͠ w 7BSJBUJPOBM"VUP&ODPEFSͷજࡏมͰ͜ΕΒͷ දݱΛڭࢣͳ͠Ͱ֫ಘͰ͖ͳ͍͔ʁ
7BSJBUJPOBM"VUP&ODPEFS 9 [ 9` FODPEFS EFDPEFS αϯϓϦϯά
7BSJBUJPOBM"VUP&ODPEFS zͷࣄޙۙࣅq(z|x) (=Τϯίʔμͷग़ྗ ͱ N(0,I)ͱͷڑΛ࠷খԽ ෮ݩޡࠩΛ࠷খԽ
Ќͩͱzͷࣄޙۙࣅq(z|x)ΛΑΓ N(0,I)ʹ͚ۙͮΑ͏ͱ͢Δ ਖ਼ଇԽ߲ʹЌΛಋೖ Ќͷࢹ֮࿏ʹ͋ΔύϥϝʔλʁΛۙࣅ
w Y࠲ඪ Z࠲ඪͦΕͧΕύλʔϯ w TDBMFύλʔϯ w ճసύλʔϯ w ΦϒδΣΫτͷܗ ପԁɺϋʔτɺ࢛֯
ύλʔϯ Yͷը૾ݸΛજࡏมݸͷ7"&Ͱֶश [ FODPEFS EFDPEFS αϯϓϦϯά ͷμ,σ
Ќ Ќ
N(0, I)ʹٵ͍ࠐ·Εͯ͠·͍ͬͯΔ જࡏมݸͷ͏ͪͷͻͱͭͷЖ͚ͩΛม͑ͯΓΛݻఆ ֤જࡏม͕ύϥϝʔλ͝ͱʹ ͔Ε͍ͯΔ જࡏมͷࢄ જࡏมͷЖΛdͰ ಈ͔͢
%FFQ.JOEʹΑΔͦͷଞͷ࣮ݧ݁Ռ IUUQUJOZVSMDPNKHCZ[LF "UBSJ %໎࿏FUD…
7"&Ͱ%JTFOUBOHMF͞ΕΔʹ w ม͕࿈ଓͰຶʹαϯϓϦϯά͞Ε͍ͯΔ w ཁૉ͕౷ܭతʹಠཱ w ੑ͕ݮΒ͞Ε͍ͯΔ
w ີʹαϯϓϦϯά͞Ε͍ͯΕɺೋ͕ͭผͷΦϒδΣΫτͩͱೝࣝͰ͖Δ
ಈ͖͕࿈ଓͰີʹαϯϓϦϯά͞Ε͍ͯΕɺ ಉ͡ΦϒδΣΫτͩͱೝࣝͰ͖Δ ͜ΕΒೋͭͷύυϧಉ͡ΦϒδΣΫτͩΖ͏͔ʁ
࠶ݱݕূ ETIBQF
࠶ݱݕূ #SFBLPVU ΫϦοΫͰ࠶ੜ wϒϩοΫͷมԽ͕ͭͬͨજࡏมʹ·͕ͨͬͯऔΕ͍ͯΔ จͷ༷ͳύυϧͷಈ͖είΞͷมԽ·ͩऔΕ͍ͯͳ͍ wЌΛڧ͘͢Δͱͬͯ͘Δજࡏมݮͬͯ͘Δ͕ɺಉ࣌ʹϘέ ۩߹͕େ͖͘ͳͬͯύυϧ͕ফ͑ͯ͠·͏ [ [
[