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

Generative Adversarial Networks by Hasanov Vagi...

Generative Adversarial Networks by Hasanov Vagif - TMLS #4

In recent years Generative Adversarial Networks were shown to be one of the best generative models available. In this talk we will follow the development of this idea starting from the moment of its appearance in 2014, and up to the most recent state of the art results.

More Decks by Tokyo Machine Learning Society

Other Decks in Technology

Transcript

  1. Tokyo Machine Learning Society Presentation Reactive (FOFSBUJWF"EWFSTBSJBM/FUXPSLT  *BO(PPEGFMMPXFUBM 

    VQMPBEFEJO+VOF    5JNFMJOFPGTFMFDUFE(FOFSBUJWF"EWFSTBSJBM/FUXPSLTSFTFBSDI
  2. Tokyo Machine Learning Society Presentation Reactive 3BOEPN WFDUPS (FOFSBUPS NPEFM

    3BOEPNTBNQMF GSPNUIFEBUBTFU %JTDSJNJOBUPS NPEFM  QSPCBCJMJUZ  QSPCBCJMJUZ JGSFBM  JGHFOFSBUFE (FOFSBUPSJTUSZJOHUPNJOJNJ[FUIFNBYJNVNQPTTJCMF QFSGPSNBODFPGEJTDSJNJOBUPS (FOFSBUJWF"EWFSTBSJBM/FUXPSL
  3. Tokyo Machine Learning Society Presentation Reactive (FOFSBUJWF"EWFSTBSJBM/FUXPSLT  *BO(PPEGFMMPXFUBM 

    VQMPBEFEJO+VOF    %FFQHFOFSBUJWFJNBHFNPEFMTVTJOHB -BQMBDJBOQZSBNJEPGBEWFSTBSJBMOFUXPSLT  &NJMZ%FOUPOFUBM  VQMPBEFEJO+VOF 5JNFMJOFPGTFMFDUFE(FOFSBUJWF"EWFSTBSJBM/FUXPSLTSFTFBSDI
  4. Tokyo Machine Learning Society Presentation Reactive JNBHFTPVSDFIUUQBSYJWPSHQEGWQEG (FOFSBUFEEJ⒎FSFODF JNBHF "WFDUPSPGSBOEPN

    OVNCFSTJO< > (FOFSBUPSOFUXPSL %JTDSJNJOBUPS OFUXPSL "DUVBMEJ⒎FSFODF JNBHF 5SBJOJOHUIFNPEFM
  5. Tokyo Machine Learning Society Presentation Reactive (FOFSBUJWF"EWFSTBSJBM/FUXPSLT  *BO(PPEGFMMPXFUBM 

    VQMPBEFEJO+VOF    %FFQHFOFSBUJWFJNBHFNPEFMTVTJOHB -BQMBDJBOQZSBNJEPGBEWFSTBSJBMOFUXPSLT  &NJMZ%FOUPOFUBM  VQMPBEFEJO+VOF 6OTVQFSWJTFESFQSFTFOUBUJPOMFBSOJOHXJUI EFFQDPOWPMVUJPOBMHFOFSBUJWFBEWFSTBSJBMOFUXPSLT  "MFD3BEGPSEFUBM  VQMPBEFEJO/PWFNCFS 5JNFMJOFPGTFMFDUFE(FOFSBUJWF"EWFSTBSJBM/FUXPSLTSFTFBSDI
  6. Tokyo Machine Learning Society Presentation Reactive 5SJDLT 3FQMBDFQPPMJOHXJUITUSJEFEDPOWPMVUJPOTJOEJTDSJNJOBUPS 6TFGSBDUJPOBMTUSJEFEDPOWPMVUJPOT EFDPOWPMVUJPOT

    JOHFOFSBUPS 6TFCBUDIOPSNBMJ[BUJPOJOCPUIUIFHFOFSBUPSBOEUIFEJTDSJNJOBUPS 3FNPWFGVMMZDPOOFDUFEIJEEFOMBZFSTGPSEFFQFSBSDIJUFDUVSFT 6TF3F-6BDUJWBUJPOJOHFOFSBUPSGPSBMMMBZFSTFYDFQUGPSUIFPVUQVU  XIJDIVTFT5BOI 6TF-FBLZ3F-6BDUJWBUJPOJOUIFEJTDSJNJOBUPSGPSBMMMBZFST 6TF"%".PQUJNJ[BUJPONFUIPEGPSPQUJNJ[BUJPO BOEUXFBLUIF IZQFSQBSBNFUFST
  7. Tokyo Machine Learning Society Presentation Reactive (FOFSBUJWF"EWFSTBSJBM/FUXPSLT  *BO(PPEGFMMPXFUBM 

    VQMPBEFEJO+VOF    %FFQHFOFSBUJWFJNBHFNPEFMTVTJOHB -BQMBDJBOQZSBNJEPGBEWFSTBSJBMOFUXPSLT  &NJMZ%FOUPOFUBM  VQMPBEFEJO+VOF 6OTVQFSWJTFESFQSFTFOUBUJPOMFBSOJOHXJUI EFFQDPOWPMVUJPOBMHFOFSBUJWFBEWFSTBSJBMOFUXPSLT  "MFD3BEGPSEFUBM  VQMPBEFEJO/PWFNCFS *NQSPWFE5FDIOJRVFTGPS5SBJOJOH("/T  5JN4BMJNBOTFUBM  6QMPBEFE+VOF 5JNFMJOFPGTFMFDUFE(FOFSBUJWF"EWFSTBSJBM/FUXPSLTSFTFBSDI
  8. Tokyo Machine Learning Society Presentation Reactive .PSFUSJDLTGPSUSBJOJOH("/T 'FBUVSFNBUDIJOH .JOJCBUDIEJTDSJNJOBUJPO )JTUPSJDBMBWFSBHJOH

    0OFTJEFEMBCFMTNPPUIJOH 7JSUVBMCBUDIOPSNBMJ[BUJPOVTFBpYFESFGFSFODFCBUDIGPSUIF CBUDIOPSNBMJ[BUJPOTUBUJTUJDTDBMDVMBUJPO5IJTQSFWFOUTUIFOFUXPSL PVUQVUGPSBHJWFOTBNQMFCFJOHEFQFOEFOUPOUIFPUIFSTBNQMFT XJUIJONJOJCBUDI 
  9. Tokyo Machine Learning Society Presentation Reactive +VOF +VOF 1SPHSFTTPGUIFQBTUUXPZFBST "NB[JOH

    JNBHFT`TPVSDFT IUUQTQBQFSTOJQTDDQBQFSHFOFSBUJWFBEWFSTBSJBMOFUTQEG  IUUQTBSYJWPSHQEGWQEG
  10. Tokyo Machine Learning Society Presentation Reactive 4FNJTVQFSWJTFEMFBSOJOHXJUI("/T 6TFDMBTTJpFSBTBEJTDSJNJOBUPSCZBEEJOHBlHFOFSBUFEEBUBz , 

    DMBTTUPJUTMJTUPGDMBTTFTGPSDMBTTJpDBUJPO 0QUJNJ[FTVQFSWJTFEBOEVOTVQFSWJTFEMPTTFTKPJOUMZ "DIJFWFTTUBUFPGUIFBSUGPSTFNJTVQFSWJTFEMFBSOJOHPO ./*45 $*'"3 47)/
  11. Tokyo Machine Learning Society Presentation Reactive $PODMVTJPO 6TF("/TGPSHFOFSBUJOHTNBMMJNBHFT PSMBSHFCVUOPU DPNQMFYJNBHFT

    6TF("/TUPUSBJODMBTTJpFSTXIFOZPVIBWFTNBMMBNPVOU PGMBCFMFEJNBHFT BOEMBSHFBNPVOUPGVOMBCFMFEJNBHFT
  12. Tokyo Machine Learning Society Presentation Reactive • Generative adversarial nets

    
 (Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio, NIPS, 2014) • Deep generative image models using a Laplacian pyramid of adversarial networks 
 (Emily Denton, Soumith Chintala, Arthur Szlam, Rob Fergus, NIPS, 2015) • Unsupervised representation learning with deep convolutional generative adversarial networks
 (Alec Radford, Luke Metz, Soumith Chintala, ICLR, 2016) • Improved Techniques for Training GANs 
 (Tim Salimans, Ian J. Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, Xi Chen, arXiv: 1606.03498, 2016) 3FGFSFODFT