Upgrade to Pro — share decks privately, control downloads, hide ads and more …

FTMA18まとめ

soneo1127
January 15, 2019

 FTMA18まとめ

好きに読んだ論文をまとめたスライドです。画風変換などGANs系が多いです。

soneo1127

January 15, 2019
Tweet

More Decks by soneo1127

Other Decks in Technology

Transcript

  1. Differentiable Plasticity: A New Method for Learning to Learn- arXiv

    2018 Thomas Miconi, Jeff Clune, Kenneth O. Stanley (Uber AI Labs) ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ CBDLQSPQͱޯ഑߱Լ๏Λ࢖ͬͯՄ઼ੑʢ)FCCJBO ͷ ͭͳ͕ΓΛσβΠϯ͢Δɽ %//ͰϔϒଇΛద༻ͨ͠΋ͷ͸͜Ε·Ͱͳ͔ͬͨɽ Y@K U ͕χϡʔϩϯͷग़ྗɽX@J K͸௨ৗͷ//ͷ܎਺ɽͦΕʹ )FCCଇΛද߲͢ΛՃ͑ͨɽ)FCC͸୯७ʹલޙͷϨΠϠʔͷɼ χϡʔϩϯͷग़ྗͷੵɽ(SBEJFOUEJTDFOUʹΑͬͯX@J KͱЋ@J K ΛUVOF͍ͯ͘͠ɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ ஋ͷύλʔϯهԱΛগͳ͍χϡʔϩϯͰͰ͖ΔΑ͏ʹͳͬͨɽ ը૾ͷ൒෼͚ͩݟͤͯ࠶ߏ੒͢Δ࣮ݧͰطଘͷ-45.͸Ͱ͖ͳ ͔ͬͨͷ͕Ͱ͖ΔΑ͏ʹͳͬͨɽYͷ໎࿏ͰɼڧԽֶशʹΑ Γ໨ҹͷҐஔʹͨͲΓண࣮͘ݧͰྑ͍Ϧβϧτɽ OFVSPNPEVMBUJPOͷํ๏͸༗ޮͱݴΘΕ͍ͯΔͷͰɼ Б΍Ћ΋ɼਆܦ׆ಈʹԠͯ͡มԽͤ͞Δͱྑ͍͔΋͠ Εͳ͍ɽ -FBSOJOHUPMFBSOXJUICBDLQSPQBHBUJPOPG)FCCJBO QMBTUJDJUZɼ.JDPOJ FUBM &WPMVUJPOBSZBEWBOUBHFTPGOFVSPNPEVMBUFEQMBTUJDJUZJO EZOBNJD SFXBSECBTFETDFOBSJPT4PMUPHHJP FUBM େીࠜ޺޾ ਓؒίʔε   
  2. World Models- arXiv 2018 David Ha (Google Brain) Jurgen Schmidhuber

    (NNAISENSE, Swiss AI Lab, IDSIA (USI & SUPSI)) ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ը໘σʔλ͔Β಺෦ϞσϧʹΑͬͯѹॖ৘ใͷ[ͱະདྷ ༧ଌͷ̷Λग़ྗ͠ɼͦͷग़ྗͱใुͷΈ͔ΒࣗΒͷߦ ಈΛֶश͢Δɽ 4DPSFͷ޲্ɽσʔλͷલॲཧ͕͍Βͳ͍ɽ Vision Modelʢߴ࣍ݩ৘ใͷѹॖʣ(VAE)ͱɼMemory RNNʢѹॖ৘ใͷ࣌ܥྻσʔλ ʢهԱʣ͔Βະདྷͷ༧ଌʣ(Mixture Density Network -RNN)ʢ͜ͷ2͕ͭ಺෦Ϟσϧͷ ੜ੒ʣͱɼControlerʢߦಈͷܾఆʣʢ࠷దԽʹCMA-ESΛ࢖༻ʣͷ3ͭΛ૊Έ߹Θͤ ͨɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ ΧʔϨʔεήʔϜͷ4DPSFɽ7J[%PPN ஄Λආ͚ͯੜ͖࢒Δήʔ Ϝ ͷ4DPSFɽ ࠶ֶशͳ͠Ͱ7"&Λ৽͍͠λεΫʹޮՌతʹ࠶ར༻Ͱ͖ͳ͍ɽ খ͞ͳ-45.ϕʔεͷੈքϞσϧͰ͸ɼه࿥Λશͯύϥϝʔλத ʹอଘ͢Δ͜ͱ͕Ͱ͖ͳ͍ͷͰେ͖ͳϞσϧԽ͔֎෦هԱ͕ඞཁɽ $PSUJDBMJOUFSOFVSPOTUIBUTQFDJBMJ[FJOEJTJOIJCJUPSZDPOUSPMç <MJOL> )1J #)BOHZB %,WJUTJBOJ +4BOEFST ;)VBOH ",FQFDT /BUVSF େીࠜ޺޾ ਓؒίʔε   
  3. Towards the Automatic Anime Characters Creation with Generative Adversarial Networks

    (make girls moe) Yanghua Jin School of Computer Science Fudan University Jiakai Zhang School of Computer Science Carnegie Mellon University Minjun Li School of Computer Science Fudan Univerisity Yingtao Tian Department of Computer Science Stony Brook University Huachun Zhu School of Mathematics Fudan Univerisity Zhihao Fang Department of Architecture Tongji Univerisity ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ Ξχϝإը૾ੜ੒ɽ"$("/Λ༻͍ͯɼ൅ܕ΍ϝΨω ͳͲͷಛ௃ΛදͤΔɽ ΑΓߴղ૾౓Խɽ જࡏม਺Λม͑Δ͜ͱͰࣗࡏʹը૾ੜ੒Ͱ͖Δɽ (FOFSBUPSʹ%FDPOWPMVUJPOͷ୅ΘΓʹ1JYFM 4IVGGMFSΛ࢖͍ɼ433FT/FU MJLFͷϞσϧʹͨ͠ɽ ໨తؔ਺ʹϖφϧςΟ߲ΛՃ͑Δ%3"("/Λ༻͍ͨɽ 8FC%//ͰɼXFC্Ͱߴ଎ʹը૾ੜ੒͕Ͱ͖Δɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ ରԠ͢Δϥϕϧͷը૾͕ੜ੒Ͱ͖͔ͨ໨ࢹͰ֬ೝͨ͠ɽ '*%TDPSFΛൺֱͨ͠ɽ ϥϯμϜϊΠζ෦෼͕඼࣭ʹେ͖͘࡞༻͢Δɽϥϕϧ ͕σʔληοτʹۉ౳ʹ෼ࢄ͞Ε͍ͯͳ͍ͨΊɼϥϕ ϧͷ૊Έ߹ΘͤʹΑͬͯ͸ྑ޷ͳը૾͕ಘΒΕͳ͍ɽ 130(3&44*7&(308*/(0'("/4'03*.1307&% 26"-*5: 45"#*-*5: "/%7"3*"5*0/ *$-3  େીࠜ޺޾  ਓؒίʔε  
  4. PROGRESSIVE GROWING OF GANS FOR IMPROVED QUALITY, STABILITY, AND VARIATION

    - ICLR2018 Tero Karras Timo Aila Samuli Laine Jaakko Lehtinen NVIDIA NVIDIA NVIDIA NVIDIA and Aalto University ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ௿ղ૾౓͔ΒͩΜͩΜߴղ૾౓ը૾Λ࡞͍ͬͯ͘ख๏ Ͱը૾ΛQYͰੜ੒ɽ ֶशʹࠇຐज़తͳ͜ͱΛ͠ͳͯ͘΋ൺֱత҆ఆֶͯ͠ शͰ͖ɺ׌ͭଟ༷ੑΛ͍࣋ͬͯΔ ੜ੒ͨ͠ը૾͕ຊ෺ͱݟ෼͚͕͔ͭͳ͍΄Ͳ៉ྷ QYQYʜͱஈ֊తʹੜ੒ͨ͠ɽ σʔληοτͷਖ਼نԽΛͨ͠ɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ *ODFQUJPOTDPSFͰΛ௒͑ͨɽʢࣝผ͞ΕΔϥϕ ϧͷόϦΤʔγϣϯ͕๛෋Ͱ͋Δ΄ͲείΞ͕ߴ͘ͳ ΔΑ͏ʹઃܭ͞ΕͨείΞʣ ֶश͕࣌ؒ௕͍ɽ ·ͩඍࡉͳͱ͜Ζʹվળͷ༨஍͕͋Δɽ 1SPHSFTTJWF/FVSBM/FUXPSLTGPS*NBHF$MBTTJGJDBUJPO ;IBOHFUBM 'VMMCPEZ)JHISFTPMVUJPO"OJNF(FOFSBUJPOXJUI 1SPHSFTTJWF4USVDUVSFDPOEJUJPOBM(FOFSBUJWF "EWFSTBSJBM/FUXPSLT  େીࠜ޺޾  ਓؒίʔε  
  5. Full-body High-resolution Anime Generation with Progressive Structure-conditional Generative Adversarial Networks

    – DeNA Blog Tero Karras Timo Aila Samuli Laine Jaakko Lehtinen NVIDIA NVIDIA NVIDIA NVIDIA and Aalto University ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ QYͷΞχϝը૾ͱϙʔζͷϞʔγϣϯ͔Βɼಈ͖ ͷ͋ΔΞχϝը૾Λੜ੒ɽ إ͚ͩͰͳ͘શ਎ը૾Λੜ੒ɽ ੜ੒ը૾ʹϞʔγϣϯ͕෇͚ΒΕΔɽ Ϟʔγϣϯը૾΋ಉ࣌ʹஈ֊తʹֶशɽ 0QFOQPTFͰϙʔζͷLFZQPJOUΛ఺ੜ੒ɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ ιʔεը૾Λλʔήοτϙʔζͷߏ଄Λ༗͢Δը૾ʹ ม׵͢Δ1(͓Αͼ%1(ͳͲͷख๏ͱൺ΂ͯಉఔ౓ ʹม׵Ͱ͖ͨɽ কདྷσʔληοτΛ࡞ΔΒ͍͠ɽ 1PTF(VJEFE1FSTPO*NBHF(FOFSBUJPO 1( <.B B> BOE%JTFOUBOHMFE1FSTPO*NBHF(FOFSBUJPO %1(  େીࠜ޺޾  ਓؒίʔε  
  6. CGANS WITH PROJECTION DISCRIMINATOR – ICLR 2018 Takeru Miyato1, Masanori

    Koyama2 [email protected] [email protected] 1Preferred Networks, Inc. 2Ritsumeikan University ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ֬཰࿦తϞσϧͷҰൠతͳϑΝϛϦʔʹΑͬͯಈػ෇͚ΒΕͨ D("/ͷ%JTDSJNJOBUPSͷϞσϧΛఏҊͨ͠ɽ৚݅෇͖ը૾ੜ੒͓Α ͼ௒ղ૾ԽͰɼ(FOFSBUPSͷੑೳΛେ෯ʹվળ͢Δ͜ͱ͕Ͱ͖ͨ ΑΓଟ༷ੑ͕޿͘ɼϞʔυ่յ͠ʹ͍͘ੜ੒͕Ͱ͖ͨɽ ͜Ε·Ͱͷ$POEJUJPOBM("/Ͱ͸୯ʹϥϕϧϕΫτϧZΛ (FOFSBUPSͷೖྗ૚͔தؒ૚ʹ࿈݁͢Δ͚͕ͩͩͬͨɼ ("/ͷ໨తؔ਺Λ෼ղ͠ɼ৚݅ϕΫτϧZͱಛ௃ϕΫτϧ Λ಺ੵ͢Δࣜ  ʹؼணͤͨ͞ɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ Ϋϥεͷ*-473$σʔληοτͷϥϕϧ৚݅෇͖ը૾ੜ ੒ͷ඼࣭Λ"$("/ͱ'*%είΞͰൺֱͨ͠ɽD("/ͷϞσϧΛ௒ղ ૾λεΫʹద༻ͨ͠৔߹ɼDPODBUʹجͮ͘D("/ B ΑΓ΋ϥϕϧ ෼ྨثͷਫ਼౓ͷ఺ͰΑΓ༏Εͨߴ඼࣭ͷ௒ղ૾ը૾Λੜ੒͢Δ͜ͱ ͕Ͱ͖ͨɽ 1 ZcY Λ୯७ͳ෼෍ͱԾఆ͍ͯ͠Δ͕ɼ࣮ࡍ͸ͦ͏Ͱ ͸ͳ͍͔΋͠Εͳ͍ͷͰཧ࿦తͳߟ࡯͕ඞཁɽ 5BLFSV .JZBUP 5PTIJLJ ,BUBPLB .BTBOPSJ,PZBNB BOE:VJDIJ:PTIJEB 4QFDUSBMOPSNBMJ[BUJPOGPSHFOFSBUJWFBEWFSTBSJBMOFUXPSLT*O*$-3  *TIBBO(VMSBKBOJ 'BSVL "INFE .BSUJO"SKPWTLZ 7JODFOU%VNPVMJO BOE"BSPO $PVSWJMMF*NQSPWFEUSBJOJOHPGXBTTFSTUFJO ("/TBS9JW QSFQSJOU BS9JW  +BF)ZVO-JNBOE+POH$IVM :F(FPNFUSJD("/BS9JW QSFQSJOU BS9JW   େીࠜ޺޾  ਓؒίʔε  
  7. SPECTRAL NORMALIZATION FOR GENERATIVE ADVERSARIAL NETWORKS- ICLR 2018 Takeru Miyato1

    , Toshiki Kataoka1 , Masanori Koyama2 , Yuichi Yoshida 1 Preferred Networks, Inc. 2 Ritsumeikan University 3 National Institute of Informatics ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ("/ͷֶशΛ҆ఆͤ͞ΔͨΊɼॏΈΛTQFDUSBM OPSNBMJ[BUJPOͨ͠ɽ ϋΠύʔύϥϝʔλͱͯ͠ɼϦϓγοπఆ਺Λௐ੔͢ Δ͚ͩͰྑ͘ͳͬͨɽ·ͨɼௐ੔͠ͳͯ͘΋ྑ͍ੑೳ ͕ग़Δɽܭࢉίετ΋খ͍͞ɽ ϨΠϠʔ͝ͱͷॏΈͷεϖΫτϧϊϧϜΛਖ਼نԽ͢Δ͜ͱͰɼ ϨΠϠʔ͝ͱͱɼϨΠϠʔશମͷϦϓγοπ੍໿ΛͰ཈͑Β ΕΔɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ $*'"3ͱ45-ɼ*NBHF/FUͰը૾ੜ੒ɽଞͷਖ਼نԽख๏ ͱൺ΂ͯੑೳΛൺֱɽ 8FJHIUOPSNBMJ[BUJPOΑΓ͸΄Ͳ஗͍ɽ Tim Salimans and Diederik P Kingma. Weight normalization: A simple reparameterization to accelerate train-ing of deep neural networks. In NIPS, pp. 901–909, 2016. େીࠜ޺޾ ਓؒίʔε   
  8. Improved training of wasserstein GANs – NIPS 2017 Ishaan Gulrajani

    et al. Montreal Institute for Learning Algorithms ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ 8("/ʹ͓͚Δ8FJHIUDMJQQJOH ॏΈΛҰఆͷൣғ಺ʹऩΊΔ ͷ ໰୊Λࢦఠ͠ɼղܾ͢ΔHSBEJFOUQFOBMUZ 8("/(1 ΛఏҊ 3FT/FUϕʔεͷਂ͍ωοτϫʔΫͰ΋ɼ8("/ͳͲ ΑΓ8("/(1ͷํ͕Ϟʔυ่յ͠ʹ͍͘ੜ੒͕Ͱ͖ Δɽ -PTTʹHSBEJFOUQFOBMUZͷ߲ΛՃ͑ͨɽ  Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ ࣮ݧͷͨΊʹ࡞ͬͨUPZ෼෍ʹ͓͍ͯɼߴ͍ϞʔϝϯτΛଊ͑ΒΕͨɽ ޯ഑രൃফࣦ͕ى͖ͳ͍ɽ *ODFQUJPO4DPSF͸ %$("/ͱൺ΂ͯऩଋ͸஗͍͕ɼऩଋޙ͸҆ఆ͍ͯ͠ Δɽ :8V :#VSEB 34BMBLIVUEJOPW BOE3(SPTTF0OUIFRVBOUJUBUJWFBOBMZTJT PGEFDPEFS CBTFEHFOFSBUJWFNPEFMTBS9JW QSFQSJOUBS9JW   େીࠜ޺޾  ਓؒίʔε  
  9. Learning Face Age Progression: A Pyramid Architecture of GANs -

    CVPR2018 Hongyu Yang1 Di Huang1 Yunhong Wang1 Anil K. Jain2 1 Beihang University, China2 Michigan State University, USA
 ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ೖྗը૾ͷਓ෺ͷ࿝ԽإΛग़ྗɽ ೥ྸม׵ͷਖ਼֬͞ͱɼਓ෺ͷಉҰੑͷอଘͰੑೳ͕ྑ ͘ͳͬͨɽ DiscriminatorͰɼ೥ྸͷಛ௃நग़Λஈ֊తʹͨ͠ɽਓ෺ͷಉҰੑΛอͭͨΊʹ deep face descriptor [20] Λ༻͍ͯɼݸਓͷإಛ௃Λಛ௃ۭؒʹམͱ͠ࠐΜͩ ͱ͖ͷloss΋ܭࢉͨ͠ɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ .031)ͱ$"$%σʔληοτΛ༻͍ͯ࿝Խͷਐߦ౓ɼਫ਼౓ɼ ਓ෺ͷಉҰੑΛ࣮ݧɽ঎༻ΞϓϦͳͲͱ΋ൺֱɽ 94IV +5BOH )-BJ --JV BOE4:BO1FSTPOBMJ[FEBHF QSPHSFTTJPOXJUIBHJOHEJDUJPOBSZ*O*$$7 QBHFTr     
 େીࠜ޺޾ ਓؒίʔε   
  10. High-Resolution Image Synthesis and Semantic Manipulation with Conditional GAN -

    CVPR2018 Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao, Jan Kautz, Bryan Catanzaro ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ߴղ૾౓ʢYʣͰϥϕϧม׵ͱ෺ମฤू ΛͰ͖ΔΑ͏ʹͨ͠ɽ QJYQJYΑΓߴղ૾౓ԽɼςΫενϟͳͲͷΠϯλϥ ΫςΟϒͳฤू͕Ͱ͖Δɽ ஈ֊తʹੜ੒(global generatorͱenhancer. enhancerΛ͔͚͍͚ͯ͹ແݶ ʹେ͖͘Ͱ͖Δʣɼஈ֊తʹࣝผ͢Δɽperceptual lossΛdiscriminatorʹ Ճ͑ͨɽdiscriminatorʹڥքϚοϓɼηάϝϯςʔγϣϯϚοϓɼ͓Αͼ࣮ /߹੒ը૾ͷνϟωϧϫΠζ࿈݁Λೖྗͨ͠ɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ ఆྔతʹ͸ɼϥϕϧը૾͔Βม׵͞Εͨࢢ֗ը૾Λطଘख๏Ͱηάϝ ϯςʔγϣϯ͠ɼJOQVUͱൺֱ͢Δɽ ఆੑతʹ͸ɼ"NB[PO.FDIBOJDBM5VSLͰ$JUZTDBQFTEBUBTFU<>͔ Βੜ੒ͨ͠ը૾ΛධՁͯ͠΋ΒͬͨɽʢͲ͕ͬͪੜ੒ը૾͔ΫΠζʣ ҩྍը૾΍ੜ෺ֶʹ΋࢖͑ͦ͏ ++PIOTPO ""MBIJ BOE-'FJ'FJ1FSDFQUVBMMPTTFTGPSSFBMUJNFTUZMFUSBOTGFS BOETVQFSSFTPMVUJPO*O&VSPQFBO$POGFSFODFPO$PNQVUFS7JTJPO &$$7      ɹ "#--BSTFO 4,4”OEFSCZ )-BSPDIFMMF BOE08JOUIFS"VUPFODPEJOH CFZPOEQJYFMTVTJOHBMFBSOFETJNJMBSJUZNFUSJD*O*OUFSOBUJPOBM$POGFSFODFPO .BDIJOF-FBSOJOH *$.-  େીࠜ޺޾ ਓؒίʔε   
  11. Pose Guided Person Image Generation – NIPS 2017 Liqian Ma1

    Xu Jia2 Qianru Sun3 Bernt Schiele3 Tinne Tuytelaars2 Luc Van Gool1,4 1KU-Leuven/PSI, TRACE (Toyota Res in Europe) 2KU-Leuven/PSI, IMEC 3Max Planck Institute for Informatics, Saarland Informatics Campus 4ETH Zurich ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ը૾ͷਓͷϙʔζΛɼೖྗͨ͠ϙʔζʹม׵ɽ ΑΓߴը࣭Ͱม׵Մೳɽ ૈ͍ը૾Ͱͷϙʔζม׵ͱɼը૾ͷߴղ૾౓Խ ຊ෺ϖ Ξͱੜ੒ϖΞͷࣝผ ͷஈ֊ʹ෼͚ͨɽ ࠷৽ͷϙʔζਪఆثΛ࢖༻ͯ͠ੜ੒ͨ͠ ݸͷLFZQPJOUΛϙʔζ৘ใͱͯ͠࢖༻ɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ %FFQ'BTIJPOͱ.BSLFUσʔληοτΛ࢖༻͠ ͯଞख๏ͱൺֱɽ44*.ͱ*ODFQUJPOTDPSFͰͷධՁ͸ ଞख๏ͱಉఔ౓ɽϢʔβʔελσΟͰߴධՁɽ உঁͰͷσʔληοτ਺ʹภΓ͕͋ΓɼੑผΛؒҧ ͑΍͍͢ɽ෰ͷڥքͷಛ௃Λ௫ΈͮΒ͍ɽ %JTFOUBOHMFE1FSTPO*NBHF(FOFSBUJPO %1(  େીࠜ޺޾  ਓؒίʔε  
  12. Looking to listen at the cocktail party a speaker-independent audio-visual

    model for speech separation - SIGGRAPH 2018 ARIEL EPHRAT, Google Research and The Hebrew University of Jerusalem, Israel INBAR MOSSERI, Google Research
 ORAN LANG, Google Research
 TALI DEKEL, Google Research KEVIN WILSON, Google Research AVINATAN HASSIDIM, Google Research WILLIAM T. FREEMAN, Google Research MICHAEL RUBINSTEIN, Google Research ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ෳ਺ͷ࿩ऀ͔Βɼը૾ͱԻ੠Λ΋ͱʹϑΟϧλΛ࡞੒ ͠ɼ֤ʑͷԻ੠Λநग़Ͱ͖Δɽ "741&&$)%"5"4&5 إಈըͱԻ੠ ͷߏஙɽ Ի੠͕֤࿩ऀ͝ͱʹಠཱ͍ͯ͠ͳ͍σʔληοτ͔Β ֶश͕Ͱ͖Δɽ ֤ʑͷإͷඵؒͷը૾ʢGSBNF GQT ͱ45'5 TIPSUUFSN ϑʔϦΤม׵ ͨ͠Ի੠͝ͱʹֶशɽ࿩ऀ͝ͱͷϚεΫΛग़ྗ ͠ɼ*4*'5ͯ͠Ի੠ʹม׵ɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ TJHOBMUPEJTUPSUJPOSBUJP 4%3 ͱ͍͏ࢦඪΛ࢖༻ɽ ଞͷख๏Ͱ͸࿩ऀ͝ͱʹಠཱͨ͠σʔληοτΛ༻͍͍ͯͨʹ ΋͔͔ΘΒͣɼຊख๏ͷ΄͏্͕ճͬͨɽ إͷಈ͖͕ͲΕ͚ͩد༩͔ͨ͠෼ੳͨ͠ɽޱҎ֎΋ॏ ཁΒ͍͠ɽ David F. Harwath, Antonio Torralba, and James R. Glass. 2016. Unsupervised Learning of Spoken Language with Visual Context. In NIPS. େીࠜ޺޾ ਓؒίʔε   
  13. Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks - PMLR

    2017 ARIEL EPHRAT, Google Research and The Hebrew University of Jerusalem, Israel INBAR MOSSERI, Google Research
 ORAN LANG, Google Research
 TALI DEKEL, Google Research KEVIN WILSON, Google Research AVINATAN HASSIDIM, Google Research WILLIAM T. FREEMAN, Google Research MICHAEL RUBINSTEIN, Google Research ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ޯ഑߱Լ๏Λ࢖͍ͬͯΔ͢΂ͯͷख๏ʹ࢖͑ΔNFUB MFBSOJOHख๏.".-ͷఏҊɽ Ϟσϧ΍ύϥϝʔλʹ੍໿ΛՃ͑ͳ͍͍ͯ͘ɽ λεΫू߹ʹ͓͍ͯ'JOF5VOJOH͢Δͱ͍͍஋ʹߦ͘Α͏ͳॳ ظ஋Λֶशɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ ճؼσʔλϙΠϯτ͕গͳ͍ͱ͖Ͱ΋ɼਖ਼ݭ೾ΛֶशՄೳɽ ෼ྨଞͷϝλख๏ͱൺ΂ͯߴਫ਼౓ ڧԽֶशΑΓૣֶ͘शՄೳ ϚϧνλεΫʹର͢ΔॳظԽΛਂ૚ֶश΍ڧԽֶशͷ ඪ४తͳཁૉͱ͢Δ͜ͱ͕͜Ε͔ΒॏཁʹͳΓͦ͏ɽ Al-Shedivat, Maruan, et al. "Continuous adaptation via meta- learning in nonstationary and competitive environments." arXiv preprint arXiv:1710.03641 (2017). େીࠜ޺޾ ਓؒίʔε   
  14. Deep Image Prior - CVPR 2018 Dmitry Ulyanov Skolkovo Institute

    of Science and Technology, Yandex Andrea Vedaldi University of Oxford [email protected] Victor Lempitsky Skolkovo Institute of Science and Technology (Skoltech) [email protected] ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ʮը૾ͱ͸͜͏͍͏΋ͷʯͱ͍͏ࣄલ৘ใ͕$//ͷߏ଄ࣗ ମʹͦ΋ͦ΋උΘ͍ͬͯΔͱԾఆ͠ɼ୯Ұը૾ͷΈΛ༻͍ͨ ֶशͰɼͦͷը૾ͷEFOPJTJOH IJHISFTPMVUJPO JOQBJOUJOH Λ࣮ݱɽ େྔͷσʔληοτ͕͍Βͳ͍ɽ ϥϯμϜϊΠζ͔Βɼ៉ྷʹ͍ͨ͠ը૾Λۙࣅ͢ΔΑ͏ʹੜ੒ ͤ͞Δͱɼӄʹ  ࣜͷ੍໿͕ຬͨ͞Εɼ͖Ε͍ͳը૾͕ੜ੒ ͞ΕΔɽʢϒϥοΫϘοΫεײʣ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ EFOPJTJOHͰ͸14/3͕405"ͱಉఔ౓   ߴղ૾౓ԽͰ΋14/3͕405"ͱಉఔ౓ 4FU<>ͱ4FU <>σʔληοτͰͱ  JOQBJOUͰ͸ $//ͷߏ଄্ɼ("/ͱಉ༷ɼ֨ࢠঢ়ʹͳΓ΍͍͢ʁ [24] V. Papyan, Y. Romano, and M. Elad. Convolutional neural networks analyzed via convolutional sparse coding. Journal of Machine Learning Research, 18(83):1–52, 2017. େીࠜ޺޾ ਓؒίʔε   
  15. A Neural Representation of Sketch Drawings - arXiv 2017 David

    Ha Google Brain [email protected] Douglas Eck Google Brain [email protected] ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ 3//ͰεέονΛඳ͘ɽ QJYFM͔Βֶश͢Δ΋ͷ͕ଟ͘ɼWFDUPS͔Βֶश͢Δͷ ͸গͳ͔ͬͨɽ જࡏม਺Λ͍͡ΕΔΑ͏ʹͨ͠ɽ FODPEF͸CJEJSFDUJPOBM3// જࡏม਺͸7"& EFDPEF͸ (..ͱTPGUNBY Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ EFOPJTJOHͰ͸14/3͕405"ͱಉఔ౓   ߴղ૾౓ԽͰ΋14/3͕405"ͱಉఔ౓ 4FU<>ͱ4FU <>σʔληοτͰͱ  JOQBJOUͰ͸ σʔλ఺͸఺͘Β͍͕ݶքɽ ෳࡶͳ΋ͷ͸͏·͍͔͘ͳ͍ɽ Ϋϥεͱ͔͸ແཧɽ Rosca, Mihaela, et al. "Variational approaches for auto-encoding generative adversarial networks." arXiv preprint arXiv:1706.04987 (2017). େીࠜ޺޾ ਓؒίʔε   
  16. Toward Multimodal Image-to-Image Translation - NIPS 2017 Jun-Yan Zhu UC

    Berkeley Trevor Darrell UC Berkeley Richard Zhang UC Berkeley Alexei A. Efros Oliver Wang Deepak Pathak UC Berkeley Eli Shechtman Adobe Research UC Berkeley Adobe Research ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ը૾ͷը෩ม׵ΛύλʔϯͰ͸ͳ͘ෳ਺Ͱ͖Δɽ QJYQJYͰ͸ੜ੒࣌ʹ༩͑ΔTFFEϊΠζΛ޻෉ͯ͠΋ NPEFDPMMBQTFͯ͠͠·͕ͬͨɼ͠ͳ͘ͳͬͨɽ D7"&("/ͱD-3("/Λ·ͱΊͨɼ#JDZDMF("/ͱ͍͏Ϟσ ϧɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ edges → photos , Google maps → satellite, labels → images, and outdoor night → day images ͷม׵Λpix2pix+noise, cAE-GAN, cVAE-GAN, cVAE-GAN++, cLR-GAN, BicycleGAN Ͱ΍ͬͯΈͨɽGoogle maps → satellites ͰLPIPSͱ͍͏ଟ༷ੑΛଌΔείΞΛൺֱɽ Amazon ϝΧχΧϧλʔΫͰὃ͞Εͨ཰Λൺֱɽ ଟ༷ੑΛଌΔ-1*14ͱ͍͏ࢦඪͷείΞ͸ը૾͕ෆࣗવͳͱ͖΋ ߴ͘ͳΔɽ MBUFOUͷ࣍ݩ͸গͳ͗͢ΔͱNPEFDPMMBQTF͢Δ͕ɼଟ͗͢Δͱ ΑΓଟ͘ͷ৘ใྔΛ࣋ͬͯ͠·͏ͨΊྑ͘ͳ͍ɽ • A. B. L. Larsen, S. K. Sønderby, H. Larochelle, and O. Winther. Autoencoding beyond pixels using a learned similarity metric. In ICML, 2016. େીࠜ޺޾ ਓؒίʔε   
  17. Autoencoding beyond pixels using a learned similarity metric - ICML

    2016 Anders Boesen Lindbo Larsen1 Søren Kaae Sønderby2 Hugo Larochelle3 Ole Winther1,2 1 Department for Applied Mathematics and Computer Science, Technical University of Denmark 2 Bioinformatics Centre, Department of Biology, University of Copenhagen, Denmark 3 Twitter, Cambridge, MA, USA ͲΜͳ΋ͷʁ 7"&ͱ("/Λ૊Έ߹Θͤͨɽ("/ͷEJTDSJNJOBUPSͰ ֶश͞Εͨಛ௃දݱΛ7"&ͷPCKFDUJWFʹ࢖༻Ͱ͖Δɽ େીࠜ޺޾ ਓؒίʔε    Vincent Dumoulin1, Ishmael Belghazi1, Ben Poole2
 Olivier Mastropietro1, Alex Lamb1, Martin Arjovsky3
 Aaron Courville1†
 1 MILA, Université de Montréal, fi[email protected].
 2 Neural Dynamics and Computation Lab, Stanford, [email protected]. 3 New York University, [email protected]. †CIFAR Fellow. ͲΜͳ΋ͷʁ Y͔Β(FOFSBUPSʹΑͬͯਪ࿦ͨ͠[ͱɼ[͔Β (FOFSBUPSʹΑͬͯਪ࿦ͨ͠ϖΞಉ࢜Λ%JTDSJNJOBUPS ʹΑͬͯൺ΂Δɽ ADVERSARIALLY LEARNED INFERENCE - ICLR 2017
  18. InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets

    - NIPS 2016 • X. Chen, Y. Duan, R. Houthooft, J. Schulman, I. Sutskever, and P. Abbeel. ͲΜͳ΋ͷʁ ڭࢣͳֶ͠शͰੜ੒ը૾Λ੍ޚͰ͖Δ જࡏม਺ͱ؍࡯ͷؒͷ૬ޓ৘ใྔΛ࠷େԽ͢Δ େીࠜ޺޾ ਓؒίʔε    • R. Zhang, J.-Y. Zhu, P. Isola, X. Geng, A. S. Lin, T. Yu, and A. A. Efros. ͲΜͳ΋ͷʁ നࠇࣸਅΛϦΞϧλΠϜͰண৭Ͱ͖ΔɽϢʔβʔΠϯ ϓοτܕͱσʔλυϦϒϯܕͷண৭ख๏ͷ༥߹ɽ  ૄͳೖྗ͔ΒશମΛਪଌɽ  దͨ͠ΧϥʔύϨοτΛαδΣετɽ  ෼Ͱྑ͍ண৭͕Ͱ͖Δ͔࣮ݧɽ  VOVTVBMͳண৭΋ੜ੒Ͱ͖Δɽ  Real-Time User-Guided Image Colorization with Learned Deep Priors - SIGGRAPH 2017
  19. CartoonGAN: Generative Adversarial Networks for Photo Cartoonization - CVPR 2018

    Yang Chen Tsinghua University, China Yu-Kun Lai Cardiff University, UK [email protected] Yong-Jin Liu∗ Tsinghua University, China ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ࣸਅΛອը෩ʹม׵͢Δ$BSUPPO("/ΛఏҊɽ ରσʔλ͕ඞཁͳ͍ɽ ଛࣦؔ਺ʹɼ௨ৗͷࣝผଛࣦʹՃ͑ͯɼֶशࡁΈ7((͔Βͷग़ ྗΛ༻͍ͯ(͕ੜ੒͢ΔֆͷҙຯΛࣦΘͳ͍Α͏͢Δϩε  ͱɼ ΤοδΛࡍཱͨͤΔϩε ී௨ͷࣸਅ͔ɼΤοδΛऔΓআ͍ͨ ອը͔ɼΤοδͦͷ··ͷອը͔ɼΛࣝผ Λ࡞ͬͨ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ ৽ւ੣෩ɼٶ࡚ॣ෩ͳͲʹม׵ɽNSTɼCycleGAN౳ͱൺֱɽ ΑΓອըͬΆ͘ɼΤοδ΋ࡍཱͭΑ͏ʹͳͬͨɽ কདྷతʹਓ෺ը΍ɼಈըʹ΋ద༻͍ͨ͠ɽ • [38] J.-Y.Zhu,T.Park,P.Isola,andA.A.Efros. Unpairedimage- to-image translation using cycle-consistent adversarial net- works. In International Conference on Computer Vision, 2017. 
 • [6] L. Gatys, A. Ecker, and M. Bethge. Image style transfer using convolutional neural networks. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2414–2423, 2016. େીࠜ޺޾ ਓؒίʔε   
  20. Unpairedimage- to-image translation using cycle-consistent adversarial net- works. - ICCV

    2017 J.-Y.Zhu,T.Park,P.Isola,andA.A.Efros. ͲΜͳ΋ͷʁ $ZDMF("/ ରͷը૾ϖΞ͕ͳͯ͘΋ը૾ͷม׵͕Ͱ͖Δɽ େીࠜ޺޾ ਓؒίʔε    L. Gatys, A. Ecker, and M. Bethge. ͲΜͳ΋ͷʁ 7((ͰऔΓग़ͨ͠ϨΠϠʔ͝ͱͷTUZMFදݱͱDPOUFOU දݱΛݩʹɼͦΕͧΕΛอ࣋͢ΔΑ͏ͳϩεؔ਺Λ࡞ Γɼը෩ม׵͢Δɽ Image style transfer using convolutional neural networks - CVPR 2016
  21. Rectifier nonlinearities improve neural network acoustic models . - ICML

    2013 A. L. Maas, A. Y. Hannun, and A. Y. Ng. ͲΜͳ΋ͷʁ -FBLZ3F-6ͷఏҊɽ Ի੠ೝࣝγεςϜͰTJHNPJEܥͳͲͱൺֱͯ͠ˋྑ͘ ͳͬͨɽ େીࠜ޺޾ ਓؒίʔε    • Y. Chen, Y.-K. Lai, and Y.-J. Liu. ͲΜͳ΋ͷʁ άϨΠεέʔϧͰɼࣸਅΛອը෩ʹม׵͢Δɽ (BUZͷٕज़Λ୯७ʹອըʹ࢖༻ͨ͠ͷʹՃ͑ͯɼࣸਅ ͔ອըՈΛࣝผ͢ΔΑ͏ʹϩεΛՃ͑ͨɽ Transforming photos to comics using convolutional neural networks. - ICIP 2017
  22. Visual A ribute Transfer through Deep Image Analogy - TOG

    2017 J. Liao, Y. Yao, L. Yuan, G. Hua, and S. B. Kang. ͲΜͳ΋ͷʁ lEFFQJNBHFBOBMPHZzͱ͍͏ɼҙຯతʹ͍ۙΑ͏ʹ ม׵͢ΔΑ͏ͳ࢓૊ΈΛఏҊɽ SFDPOTUSVDUJPOʹ1BUDI.BUDIΛ࠾༻ͨ͠ɽ େીࠜ޺޾ ਓؒίʔε   
  23. Neural scene representation and rendering - Science 2018 June 15

    S. M. Ali Eslami*†, Danilo Jimenez Rezende†, Frederic Besse, Fabio Viola,
 Ari S. Morcos, Marta Garnelo, Avraham Ruderman, Andrei A. Rusu, Ivo Danihelka, Karol Gregor, David P. Reichert, Lars Buesing, Theophane Weber, Oriol Vinyals, Dan Rosenbaum, Neil Rabinowitz, Helen King, Chloe Hillier, Matt Botvinick,
 Daan Wierstra, Koray Kavukcuoglu, Demis Hassabis ͲΜͳ΋ͷʁ ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ٞ࿦͸͋Δʁ ࣍ʹಡΉ΂͖࿦จ͸ʁ ෳ਺ͷࢹ఺ͷը૾͔Βɺର৅ͷࡾ࣍ݩੈքΛූ߸Խ͠ ҟͳΔࢹ఺͔ΒͷࢹքΛ෮ݩ͢ΔɻۭؒϞσϧΛજࡏ ม਺ͱ͢ΔજࡏϞσϧͰ࠷దԽ͢Δɻ ਓؒͷϥϕϧ෇ͳͲͷࣄલ஌ࣝແ͠ͰTDFOFΛղऍ͠ɼදݱͰ͖Δɽ ैདྷͷ%ͰͷֶशͰ͸ѻΘͳ͔ͬͨɼςΫενϟ΍ϥΠςΟϯάͳͲ ந৅తͳ΋ͷ΋ɼදݱۭؒʹམͱ͠ࠐΜͰֶशͰ͖Δɽ SFQSFTFOUBUJPOOFUXPSL͕ଟࢹ఺Χϝϥ W ͔ΒදݱϕΫτϧ SΛֶश͢Δɽ֤ࢹ఺͔ΒͷදݱΛཁૉ࿨ͯ͠දݱϕΫτϧʹ ͢ΔɽHFOFSBUJPOOFUXPSL͕3//Ͱଞࢹ఺ ෆఆ ͔Βͷө૾ Λ༧ଌɽ Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ কདྷతʹ࣌ؒతɼۭؒతͳRVFSZ͕Ͱ͖ΔΑ͏ͳɼTDFOFͷਂ͍ ཧղΛͰ͖ΔΑ͏ʹ͍ͨ͠ɽ େીࠜ޺޾ ਓؒίʔε    ϩϘοτΞʔϜΛ৭෇͖ͷ෺ମʹ౸ୡͤ͞ΔλεΫΛɼݻఆΧϝ ϥͷ৔߹ΑΓૣֶ͘शͰ͖Δɽ ෦෼؍ଌʢҰਓশࢹ఺ʣͷΈͷ໎࿏ 5,BSSBT 5"JMB 4-BJOF +-FIUJOFO BS9JW<DT/&> 0DUPCFS  5%,VMLBSOJ 1,PIMJ +#5FOFOCBVN 7.BOTJOHILB JO1SPDFFEJOHTPGUIF *&&&$POGFSFODFPO$PNQVUFS7JTJPOBOE1BUUFSO3FDPHOJUJPO $713  *&&&  
 QQr
  24. Conditional image generation with PixelCNN decoders - NIPS 2016 A.

    van den Oord et al. ͲΜͳ΋ͷʁ લ·ͰͷϐΫηϧ͔Β࣍ͷϐΫηϧΛ༧ଌ͢Δ 1JYFM$//ʹʮ໡఺ CMJOETQPU ʯΛઃ͚ͨ(BUFE 1JYFM$//ͱɼը૾ͷ৚݅Ϟσϧ͕Ͱ͖Δ$POEJUJPOBM 1JYFM$//ΛఏҊɽ7"&ͷEFDPEFͷ࣭Λ޲্ɽ େીࠜ޺޾ ਓؒίʔε    A. R. Zamir et al. ͲΜͳ΋ͷʁ %0'Χϝϥͷϙʔζਪఆ ࣄલͷಛ௃ྔͳ͠ʹɼGFBUVSFNBUDIJOHͷ405" Generic 3D representation via pose estimation and matching - ECCV 2016
  25. Unsupervised learning of 3D structure from images - NIPS 2016

    D. J. Rezende et al ͲΜͳ΋ͷʁ ը૾͔Β%ͷ෺ମΛੜ੒ɽ %ີ౓ϞσϦϯάΛॳΊͯߦͬͨɽ େીࠜ޺޾ ਓؒίʔε    A. A. Rusu et al ͲΜͳ΋ͷʁ ௿Ϩϕϧͷࢹ֮ಛ௃ΛߴϨϕϧͷϙϦγʔʹҠ͍ͯ͠ ͘QSPHSFTTJWFOFUXPSLʹΑͬͯγϛϡϨʔγϣϯͰ ͷֶशΛ࣮؀ڥʹҠ͢ɽ Sim-to-real robot learning from pixels with progressive nets. - arXiv 2016
  26. Stochastic back-propagation and variational inference in deep latent Gaussian models

    - ICML 2014 D. J. Rezende, S. Mohamed, D. Wierstra, ͲΜͳ΋ͷʁ Ϟσϧʹࣄલ෼෍Λஔ͍ͯσʔλͷ֬཰తΤϯίʔυ Λ͢Δɽੜ੒ϞσϧͱࣝผϞσϧͰ#BDLQSPQΛ֬཰ ม਺Ͱߦ͏ɽ େીࠜ޺޾ ਓؒίʔε