[re:Growth 2019] Deep Learning が学べる最高のツール 「DeepComposer」

[re:Growth 2019] Deep Learning が学べる最高のツール 「DeepComposer」

Taewoo Kim
金泰雨（キム テウ）

クラスメソッド株式会社のAWS re:Invent 2019 の振り返り勉強会イベントに登壇しました。

Transcript

1. Deep Learning ֶ͕΂Δ࠷ߴͷπʔϧ ʮDeepComposerʯ Taewoo Kim ΩϜ ς΢ Classmethod re:Growth 2019 ౦ژ
2. ͪΐͬͱ ଴ͬͯʂ
3. ֎ࠃਓͳΜ͔ͩΒ ༏ͯ͘͘͠͠ΕͯͶὑ
4. ࣗݾ঺հ { “name”: “Taewoo Kim”, “company”: “Classmethod, Inc”, “role”: “AWS engineer”, “like”: [ “aws”, “serverless”, “machine learning”, “chaos engineering” ] }
5. ࠷ॳ͔Βਅ໘໨ͳ࿩Λ ͠Α͏͔ͱࢥ͍·͢ɻ
6. Deep Learning ͝ଘ஌ͷํ ͍Βͬ͠Ό͍·͔͢ʁ
7. ίʔυϨϕϧͰ ͝ଘ஌ͷํ ͍Βͬ͠Ό͍·͔͢ʁ
8. ίʔυϨϕϧͰ ৄ͍͠ํ ͍Βͬ͠Ό͍·͔͢ʁ
9. Generative Adversarial Networks (GANs)
10. Generative Adversarial Networks (GANs) CycleGAN: https://junyanz.github.io/CycleGAN/
11. ͓ʂෆࢥٞͩʂ ֶ΅͏͔ʂ
12. Generative Adversarial Networks (GANs)
13. Generative Adversarial Networks (GANs) • Deep Belief Networks (DBNs) • Generative Stochastic Models (GSMs) • Variational Autoencoders (VAEs) • Markov Chain Monte Carlo (MCMC) • …
14. Generative Adversarial Networks (GANs)
15. Generative Adversarial Networks (GANs)
16. ఘΊͨํ͕ྑ͍ͬͯ ࢥ͍·͔ͨ͠ʁ
17. Deep Learning ֶ͕΂Δ࠷ߴͷπʔϧ ʮDeepComposerʯ
18. AWS DeepComposer
19. AWS DeepComposer
20. AWS DeepComposer
22. ͪΐ͏ͲSageMakerपΓͷ ৽αʔϏε͕ ൃද͞Ε·ͨ͠ʂ
23. Jupyter Notebook The Jupyter Notebook is an open-source web application that allows you to create and share documents that contain live code, equations, visualizations and narrative text. Uses include: data cleaning and transformation, numerical simulation, statistical modeling, data visualization, machine learning, and much more. “
24. Jupyter Notebook
25. Jupyter Notebook
26. Jupyter Notebook
27. Jupyter Notebook
28. Jupyter Notebook
29. Jupyter Notebook
30. Jupyter Notebook
31. Jupyter Notebook
32. Generative Adversarial Networks (GANs) • Generative Models Ͱ͢ɻ • AIͷ૑଄׆ಈͱؔ܎ͷਂ͍ϞσϧͰ͢ɻ • Deep Learning ͷੈքͰ defacto ͷϞσϧͷҰͭͰ͢ɻ • ͍ΖΜͳ෼໺Ͱ࢖͑·͢ɻ • ֶΜͰ͓͘ͱ͖ͬͱ໾ʹཱͯ·͢ɻ
33. GANsͷଟ༷ͳࣄྫ zi2zi: Master Chinese Calligraphy with Conditional Adversarial Networks
34. GANsͷଟ༷ͳࣄྫ zi2zi: Master Chinese Calligraphy with Conditional Adversarial Networks
35. GANsͷଟ༷ͳࣄྫ StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks
36. GANsͷଟ༷ͳࣄྫ StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks
37. GANsͷଟ༷ͳࣄྫ High-Quality Face Image Super-Resolution Using Conditional Generative Adversarial Networks
38. GANsͷଟ༷ͳࣄྫ High-Quality Face Image Super-Resolution Using Conditional Generative Adversarial Networks
39. GANsͷίʔυॻ͖ • શͯͷίʔυΛ঺հ͢Δ͜ͱ͸ແཧɻ • ͲΜͳײ͡ͰDeep Learningίʔυ͕ॻ͚Δͷ͔Λ঺հɻ • ਺ֶͷਂ͍஌ࣝແ͠Ͱ΋ίʔυ͕ॻ͚·͢ɻ
 ʢ਺ֶͷਂ͍஌͕ࣝ͋Ε͹΋ͬͱreasonableͳߟ͕͑Մೳ ͳ͚ͩʣ
40. ྫʣGANsͷίʔυॻ͖
41. ྫʣGANsͷίʔυॻ͖
42. 2015೥ ࢲ͕௚઀ॻ͍ͨίʔυ
45. ͜Μͳͷ͸ ΋͏΍Βͳͯ͘ྑ͍
46. Curated list of awesome GAN applications and demo
47. Curated list of awesome GAN applications and demo
48. Curated list of awesome GAN applications and demo
49. Deep Learning ษڧ͠·͠ΐ͏ʂ ʮDeepComposerʯͰ
50. ऴ