Deep learning applications - for fun and profit
?
2 likes?762 views
This document provides an overview of deep learning applications across various domains including art, music, images, language, healthcare, agriculture, sports and more. It discusses how deep learning is used for tasks like image generation, style transfer, speech generation, machine translation, disease prediction, crop yield prediction, and game strategies. The document also briefly discusses the future of social AI and concludes that deep learning will revolutionize many fields while noting resources for learning more about the topic.
Deep learning applications - for fun and profit
- 1. Deep Learning Applications for fun and profit
- 2. Hello!I am Roberto Silveira EE engineer, ML enthusiast rsilveira79@gmail.com @rsilveira79
- 3. ¡°Fearing a rise of killer robots is like worrying about overpopulation on Mars.¡± (Andrew Ng)
- 4. { A short intro }
- 5. Deep Learning (1) Artificial Intelligence e.g. knowledge bases Source: Deep Learning (Goodfellow, Bengio, Courville) Machine Learning e.g. logistic regression Representation Learning e.g. shallow autoencoders Deep Learning e.g. MLPs
- 6. Source: Deep Learning (Goodfellow, Bengio, Courville) Rule-based INPUT OUTPUT CAT Classic ML Deep Learning Hand designed program CAT Hand designed features Mapping from features CAT CAT Features Mapping from features Simple Features + more abstract features Mapping from features Representation Learning Deep Learning (2)
- 7. Visualizing distributed representation Source: Convolutional networks and applications in vision (Yan LeCun et al., 2010)
- 8. Visualizing distributed representation Source: Visualizing and Understanding Convolutional Networks (Zeiler et al., 2013)
- 9. Visualizing distributed representation Source: Visualizing and Understanding Convolutional Networks (Zeiler et al., 2013)
- 10. Visualizing distributed representation Source: Visualizing and Understanding Convolutional Networks (Zeiler et al., 2013)
- 11. Start of a revolution - why now? Availability of data Processing power (GPUs) Improved models/ techniques (activations, regularization, initialization, attention)
- 12. { Art }
- 13. A Neural Algorithm of Artistic Style Source: A Neural Algorithm of Artistic Style (Gatys et al., 2015) Style Content
- 14. La?a-Style
- 15. La?a-Style
- 16. Deep Dream Low-level layer enhancement High-level layer enhancement Source: https://goo.gl/U4xhM6
- 17. Deep Dream
- 18. La?a-Dreams
- 19. La?a-Dreams
- 20. La?a-Dreams
- 21. La?a-Dreams
- 22. La?a-Dreams
- 23. La?a-Dreams
- 24. { Music, Sound }
- 25. WaveNet - Speech Generation Parametric Concatenative WaveNet Music #2 Music #1 Source: WaveNet: A Generative Model for Raw Audio (Oord et al., 2016)
- 26. Music recommendation Collaborative filtering vs content based Genre Instruments Mood Year Geolocation of artist Lyrical themes Low level (max) - Vibrato singing Low level (avg) - noise, distortion Low level (avg) - Chinese pop Similarity Playlists - Fleet Foxes Source: Deep content-based music recommendation (Oord et al., 2013)
- 27. { Image, Photos }
- 28. { GANs
- 29. Generative Adversarial Networks Source: Thomas Paula (slideshare) Goal: produce counterfeit money that is as similar as real money Goal: distinguish between real and counterfeit money Generated from Gaussian or Normal distribution (usually) Generated instance Training image generator discriminator
- 30. }
- 31. GANs-Image Generation Source: Generative Visual Manipulation on the Natural Image Manifold (Zhu et al., 2016)
- 32. GANs-Super Resolution Source: Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network, (Zhu et al., 2016) bicubic SRResNet SRGAN Original (HR)
- 33. GANs-Super Resolution Source: Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network, (Zhu et al., 2016) bicubic SRResNet SRGAN Original (HR)
- 34. GANs-Super Resolution Source: Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network, (Zhu et al., 2016) bicubic SRResNet SRGAN Original (HR)
- 35. Semantic/Instance Segmentation Source: Mask R-CNN (He et al., 2017)
- 36. Semantic/Instance Segmentation Source: Mask R-CNN (He et al., 2017)
- 37. Semantic/Instance Segmentation Source: Mask R-CNN (He et al., 2017)
- 38. { Language, Text }
- 39. Language Translation Non-monotonic alignment Source: Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation (Cho et al., 2014)
- 40. Text Summarization Source: A Deep Reinforced Model for Abstractive Summarization (Paulus et al., 2017) Extractive vs Abstractive
- 42. Conversational Bots Source: A Neural Conversational Model (Vinyals et al., 2015) Google Smart Reply
- 43. Visual Questioning and Answering Sources: (1)Dynamic Memory Networks for Visual and Textual Question Answering (Xiong et al., 2016), (2)Inferring and Executing Programs for Visual Reasoning , (Johnson et al., 2017) 2-level input mechanism (visual + text)
- 44. { Healthcare }
- 45. Biomedical image segmentation Source: U-Net: Convolutional Networks for Biomedical Image Segmentation (Ronneberger et al., 2015)
- 46. Disease prediction Source: https://goo.gl/jGK2Jp ML improvements: Random forest: +1.7 % Logistic regression: +3.2 % Gradient boosting: +3.3 % Neural nets: +3.6%
- 47. { Agriculture }
- 49. Crop yield prediction Source: http://www.descarteslabs.com/ RGB image Infrared image
- 50. Crop yield prediction Source: http://www.descarteslabs.com/ Corn Prediction (State)
- 51. Crop yield prediction Source: http://www.descarteslabs.com/ Soy Prediction (State)
- 52. Kaggle - Understanding the Amazon Source: https://goo.gl/5eBxnr
- 53. Kaggle - Understanding the Amazon Source: https://goo.gl/5eBxnr Cloudy
- 54. Kaggle - Understanding the Amazon Source: https://goo.gl/5eBxnr Rain forest
- 55. Kaggle - Understanding the Amazon Source: https://goo.gl/5eBxnr Water (rivers & lakes)
- 56. Kaggle - Understanding the Amazon Source: https://goo.gl/5eBxnr Habitation
- 57. Kaggle - Understanding the Amazon Source: https://goo.gl/5eBxnr Agriculture
- 58. { Sports }
- 59. Ghosting with Deep Imitation Learning Source: Data-Driven Ghosting using Deep Imitation Learning (Le et al., 2017) Role and formation discovery 4-5-1
- 60. { Future }
- 61. Social AI ¡ú AI + emotions understanding synthesis/generation Jibo Cozmo
- 62. { Conclusions }
- 63. - Deep Learning is here to stay - AI won¡¯t kill humans (for the time being) - AI will revolutionize labor market (already is) - Nothing to do (1) ¡ú Learn Python - Nothing to do (2) ¡ú Learn TensorFlow/Keras/PyTorch - Nothing to do (3) ¡ú ML Andrew Ng (Coursera), Deep Learning (Udacity), Stanford CS231n (Youtube), Stanford CS224d (Youtube), Oxford NLP (Github) - Nothing to do (4) ¡ú Practice @ Kaggle