Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Deep Learning without Annotations - Xavier Giro - UPC Barcelona 2018

183 views

Published on

https://telecombcn-dl.github.io/2018-dlai/

Deep learning technologies are at the core of the current revolution in artificial intelligence for multimedia data analysis. The convergence of large-scale annotated datasets and affordable GPU hardware has allowed the training of neural networks for data analysis tasks which were previously addressed with hand-crafted features. Architectures such as convolutional neural networks, recurrent neural networks or Q-nets for reinforcement learning have shaped a brand new scenario in signal processing. This course will cover the basic principles of deep learning from both an algorithmic and computational perspectives.

Published in: Data & Analytics
  • Be the first to comment

Deep Learning without Annotations - Xavier Giro - UPC Barcelona 2018

  1. 1. Xavier Giro-i-Nieto xavier.giro@upc.edu Associate Professor Universitat Politecnica de Catalunya Technical University of Catalonia Learning without Annotations Day 4 Lecture 2 #DLUPC http://bit.ly/dlai2018
  2. 2. 2 Outline 1. Unsupervised Learning 2. Semi-supervised Learning 3. Self-supervised Learning
  3. 3. 3 Outline 1. Unsupervised Learning 2. Semi-supervised Learning 3. Self-supervised Learning
  4. 4. 4 Acknowledgements Kevin McGuinness kevin.mcguinness@dcu.ie Research Fellow Insight Centre for Data Analytics Dublin City University
  5. 5. 5 Video lectures Kevin McGuinness, DLCV 2016 Xavier Giró, DLAI 2017
  6. 6. Types of machine learning Yann Lecun’s Black Forest cake 6
  7. 7. 7 Types of machine learning We can categorize three types of learning procedures: 1. Supervised Learning: 𝐲 = ƒ(𝐱) 2. Unsupervised Learning: ƒ(𝐱) 3. Reinforcement Learning (RL): 𝐲 = ƒ(𝐱) 𝐳 Predict label y corresponding to observation x Estimate the distribution of observation x Predict action y based on observation x, to maximize a future reward z
  8. 8. 8 Types of machine learning We can categorize three types of learning procedures: 1. Supervised Learning: 𝐲 = ƒ(𝐱) 2. Unsupervised Learning: ƒ(𝐱) 3. Reinforcement Learning (RL): 𝐲 = ƒ(𝐱) 𝐳
  9. 9. 9 Why Unsupervised Learning ? ● It is the nature of how intelligent beings percept the world. ● It can save us tons of efforts to build a human-alike intelligent agent compared to a totally supervised fashion. ● Vast amounts of unlabelled data.
  10. 10. 10 eg. Bayesian Classifier ● P(Y|X) depends on P(X|Y) and P(X) ● Knowledge of distribution P(X) can help to predict P(Y|X) ● Good model of P(X) must have Y as an implicit latent variable Bayes rule X: Data Y: Labels Slide: Kevin McGuinness (DLCV UPC 2017) Why Unsupervised Learning ?
  11. 11. 11 eg. Clustering before linear classification Slide: Kevin McGuinness (DLCV UPC 2017) x1 x2 Cluster 1 Cluster 2 Cluster 3 Cluster 4 1 2 3 4 1 2 3 4 4D Bag of Words representation Separable with a linear classifiers in a 4D space ! Why Unsupervised Learning ?
  12. 12. 12 Example from Kevin McGuinness Juyter notebook. Why Unsupervised Learning ? eg. Clustering before linear classification
  13. 13. 13 Assumptions for Unsupervised Learning Slide: Kevin McGuinness (DLCV UPC 2017) To model P(X) given data, it is necessary to make some assumptions “You can’t do inference without making assumptions” -- David MacKay, Information Theory, Inference, and Learning Algorithms
  14. 14. 14 Assumptions for Unsupervised Learning Slide: Kevin McGuinness (DLCV UPC 2017) To model P(X) given data, it is necessary to make some assumptions “You can’t do inference without making assumptions” -- David MacKay, Information Theory, Inference, and Learning Algorithms Typical assumptions: ● Smoothness assumption ○ Points which are close to each other are more likely to share a label. ● Cluster assumption ○ The data form discrete clusters; points in the same cluster are likely to share a label ● Manifold assumption ○ The data lie approximately on a manifold of much lower dimension than the input space.
  15. 15. 15 Assumptions for Unsupervised Learning Slide: Kevin McGuinness (DLCV UPC 2017) Smoothness assumption ● Label propagation ○ Recursively propagate labels to nearby points ○ Problem: in high-D, your nearest neighbour may be very far away! Cluster assumption ● Bag of words models ○ K-means, etc. ○ Represent points by cluster centers ○ Soft assignment ○ VLAD ● Gaussian mixture models ○ Fisher vectors Manifold assumption ● Linear manifolds ○ PCA ○ Linear autoencoders ○ Random projections ○ ICA ● Non-linear manifolds: ○ Non-linear autoencoders ○ Deep autoencoders ○ Restricted Boltzmann machines ○ Deep belief nets
  16. 16. 16 Assumptions for Unsupervised Learning Slide: Kevin McGuinness (DLCV UPC 2017) Smoothness assumption ● Label propagation ○ Recursively propagate labels to nearby points ○ Problem: in high-D, your nearest neighbour may be very far away! Cluster assumption ● Bag of words models ○ K-means, etc. ○ Represent points by cluster centers ○ Soft assignment ○ VLAD ● Gaussian mixture models ○ Fisher vectors Manifold assumption ● Linear manifolds ○ PCA ○ Linear autoencoders ○ Random projections ○ ICA ● Non-linear manifolds: ○ Non-linear autoencoders ○ Deep autoencoders ○ Restricted Boltzmann machines ○ Deep belief nets
  17. 17. 17 The manifold hypothesis Slide: Kevin McGuinness (DLCV UPC 2017) The data distribution lie close to a low-dimensional manifold Example: consider image data ● Very high dimensional (1,000,000D) ● A randomly generated image will almost certainly not look like any real world scene ○ The space of images that occur in nature is almost completely empty ● Hypothesis: real world images lie on a smooth, low-dimensional manifold ○ Manifold distance is a good measure of similarity Similar for audio and text
  18. 18. 18 The manifold hypothesis Slide: Kevin McGuinness (DLCV UPC 2017) x1 x2 Linear manifold wT x + b x1 x2 Non-linear manifold
  19. 19. 19 Assumptions for Unsupervised Learning Slide: Kevin McGuinness (DLCV UPC 2017) Smoothness assumption ● Label propagation ○ Recursively propagate labels to nearby points ○ Problem: in high-D, your nearest neighbour may be very far away! Cluster assumption ● Bag of words models ○ K-means, etc. ○ Represent points by cluster centers ○ Soft assignment ○ VLAD ● Gaussian mixture models ○ Fisher vectors Manifold assumption ● Linear manifolds ○ PCA ○ Linear autoencoders ○ Random projections ○ ICA ● Non-linear manifolds: ○ Non-linear autoencoders ○ Deep autoencoders ○ Restricted Boltzmann machines ○ Deep belief nets
  20. 20. 20 F. Van Veen, “The Neural Network Zoo” (2016) Autoencoder (AE)
  21. 21. 21 Autoencoder (AE) Fig: “Deep Learning Tutorial” Stanford Autoencoders: ● Predict at the output the same input data. ● Do not need labels.
  22. 22. 22 Autoencoder (AE) Fig: “Deep Learning Tutorial” Stanford Dimensionality reduction: Use the hidden layer as a feature extractor of any desired size.
  23. 23. 23 Autoencoder (AE) Slide: Kevin McGuinness (DLCV UPC 2017) Encoder W1 Decoder W2 hdata reconstruction Loss (reconstruction error) Latent variables (representation/features) Pretraining: 1. Initialize a NN by solving an autoencoding problem.
  24. 24. 24 Autoencoder (AE) Slide: Kevin McGuinness (DLCV UPC 2017) Encoder W1 hdata Classifier WC Latent variables (representation/features) prediction y Loss (cross entropy) Pretraining: 1. Initialize a NN solving an autoencoding problem. 2. Train for final task with “few” labels.
  25. 25. 25 Autoencoder (AE) Deep Learning TV, “Autoencoders - Ep. 10”
  26. 26. 26 F. Van Veen, “The Neural Network Zoo” (2016) Restricted Boltzmann Machine (RBM) RBMs are a specific type of autoencoder.
  27. 27. 27 Restricted Boltzmann Machine (RBM) Figure: Geoffrey Hinton (2013) Salakhutdinov, Ruslan, Andriy Mnih, and Geoffrey Hinton. "Restricted Boltzmann machines for collaborative filtering." ICML 2007.
  28. 28. 28 Restricted Boltzmann Machine (RBM) DeepLearning4j, “A Beginner’s Tutorial for Restricted Boltzmann Machines”. Forward pass
  29. 29. 29 Restricted Boltzmann Machine (RBM) DeepLearning4j, “A Beginner’s Tutorial for Restricted Boltzmann Machines”. Backward pass
  30. 30. 30 Restricted Boltzmann Machine (RBM) DeepLearning4j, “A Beginner’s Tutorial for Restricted Boltzmann Machines”. Backward pass
  31. 31. 31 Restricted Boltzmann Machine (RBM) DeepLearning4j, “A Beginner’s Tutorial for Restricted Boltzmann Machines”. Backward pass The reconstructed values at the visible layer are compared with the actual ones with the KL Divergence.
  32. 32. 32 F. Van Veen, “The Neural Network Zoo” (2016) Deep Belief Networks (DBN)
  33. 33. 33 Deep Belief Networks (DBN) Architecture like an MLP, training as a stack of RBMs. Hinton, Geoffrey E., Simon Osindero, and Yee-Whye Teh. "A fast learning algorithm for deep belief nets." Neural computation 18, no. 7 (2006): 1527-1554.
  34. 34. 34 Deep Belief Networks (DBN) Hinton, Geoffrey E., Simon Osindero, and Yee-Whye Teh. "A fast learning algorithm for deep belief nets." Neural computation 18, no. 7 (2006): 1527-1554.
  35. 35. 35 Deep Belief Networks (DBN) Hinton, Geoffrey E., Simon Osindero, and Yee-Whye Teh. "A fast learning algorithm for deep belief nets." Neural computation 18, no. 7 (2006): 1527-1554.
  36. 36. 36 Deep Belief Networks (DBN) Hinton, Geoffrey E., Simon Osindero, and Yee-Whye Teh. "A fast learning algorithm for deep belief nets." Neural computation 18, no. 7 (2006): 1527-1554.
  37. 37. 37 Deep Belief Networks (DBN) Hinton, Geoffrey E., Simon Osindero, and Yee-Whye Teh. "A fast learning algorithm for deep belief nets." Neural computation 18, no. 7 (2006): 1527-1554.
  38. 38. 38 Deep Belief Networks (DBN) + Softmax Hinton, Geoffrey E., Simon Osindero, and Yee-Whye Teh. "A fast learning algorithm for deep belief nets." Neural computation 18, no. 7 (2006): 1527-1554. Softmax After the DBN is trained, it can be fine-tuned with a reduced amount of labels to solve a supervised task with superior performance. Supervised learning
  39. 39. 39 Deep Belief Networks (DBN) Deep Learning TV, “Deep Belief Nets”
  40. 40. 40 Deep Belief Networks (DBN) + Softmax Geoffrey Hinton, "Introduction to Deep Learning & Deep Belief Nets” (2012) Geoffrey Hinton, “Tutorial on Deep Belief Networks”. NIPS 2007.
  41. 41. 41 Geoffrey Hinton
  42. 42. 42 Outline 1. Unsupervised Learning 2. Semi-supervised Learning 3. Self-supervised Learning
  43. 43. 43 Slide: Kevin McGuinness (DLCV UPC 2017) Why Semi-supervised Learning ? Max margin decision boundary
  44. 44. 44 Slide: Kevin McGuinness (DLCV UPC 2017) Why Semi-supervised Learning ? Semi supervised decision boundary
  45. 45. 45 Slide: Kevin McGuinness (DLCV UPC 2017) Why Semi-supervised Learning ?
  46. 46. 46 Slide: Kevin McGuinness (DLCV UPC 2017) Why Semi-supervised Learning ?
  47. 47. 47 Outline 1. Unsupervised Learning 2. Semi-supervised Learning 3. Self-supervised Learning a. Predictive Learning b. Learning by verification c. Cross-modal Learning
  48. 48. 48 Outline 1. Unsupervised Learning 2. Semi-supervised Learning 3. Self-supervised Learning a. Predictive Learning b. Learning by verification c. Cross-modal Learning
  49. 49. 49 Acknowledgements Víctor Campos Junting Pan Xunyu Lin
  50. 50. 50 Predictive Learning Slide credit: Yann LeCun
  51. 51. 51 Predictive Learning Srivastava, Nitish, Elman Mansimov, and Ruslan Salakhutdinov. "Unsupervised Learning of Video Representations using LSTMs." In ICML 2015. [Github] Learning video representations (features) by...
  52. 52. 52 Predictive Learning Srivastava, Nitish, Elman Mansimov, and Ruslan Salakhutdinov. "Unsupervised Learning of Video Representations using LSTMs." In ICML 2015. [Github] (1) frame reconstruction (AE): Learning video representations (features) by...
  53. 53. 53 Predictive Learning Srivastava, Nitish, Elman Mansimov, and Ruslan Salakhutdinov. "Unsupervised Learning of Video Representations using LSTMs." In ICML 2015. [Github] Learning video representations (features) by... (2) frame prediction
  54. 54. 54 Predictive Learning Srivastava, Nitish, Elman Mansimov, and Ruslan Salakhutdinov. "Unsupervised Learning of Video Representations using LSTMs." In ICML 2015. [Github] Unsupervised learned features (lots of data) are fine-tuned for activity recognition (small data).
  55. 55. 55 Predictive Learning Mathieu, Michael, Camille Couprie, and Yann LeCun. "Deep multi-scale video prediction beyond mean square error." ICLR 2016 [project] [code] Video frame prediction with a ConvNet.
  56. 56. 56 Predictive Learning Mathieu, Michael, Camille Couprie, and Yann LeCun. "Deep multi-scale video prediction beyond mean square error." ICLR 2016 [project] [code] The blurry predictions from MSE are improved with multi-scale architecture, adversarial learning and an image gradient difference loss function.
  57. 57. 57 Predictive Learning Mathieu, Michael, Camille Couprie, and Yann LeCun. "Deep multi-scale video prediction beyond mean square error." ICLR 2016 [project] [code] The blurry predictions from MSE (l1) are improved with multi-scale architecture, adversarial training and an image gradient difference loss (GDL) function.
  58. 58. 58 Predictive Learning Mathieu, Michael, Camille Couprie, and Yann LeCun. "Deep multi-scale video prediction beyond mean square error." ICLR 2016 [project] [code]
  59. 59. 59 Outline 1. Unsupervised Learning 2. Semi-supervised Learning 3. Self-supervised Learning a. Predictive Learning b. Learning by verification c. Cross-modal Learning
  60. 60. 60 Spatiotemporal coherence Le, Quoc V., Will Y. Zou, Serena Y. Yeung, and Andrew Y. Ng. "Learning hierarchical invariant spatio-temporal features for action recognition with independent subspace analysis." CVPR 2011
  61. 61. 61 Assumption: adjacent video frames contain semantically similar information. Autoencoder trained with regularizations by slowliness and sparisty. Goroshin, Ross, Joan Bruna, Jonathan Tompson, David Eigen, and Yann LeCun. "Unsupervised learning of spatiotemporally coherent metrics." ICCV 2015. Temporal coherence
  62. 62. 62 Temporal coherence Jayaraman, Dinesh, and Kristen Grauman. "Slow and steady feature analysis: higher order temporal coherence in video." CVPR 2016. [video] Slow feature analysis ● Temporal coherence assumption: features should change slowly over time in video Steady feature analysis ● Second order changes also small: changes in the past should resemble changes in the future Train on triplets of frames from video Loss encourages nearby frames to have slow and steady features, and far frames to have different features
  63. 63. 63 Temporal coherence (Slides by Xunyu Lin): Misra, Ishan, C. Lawrence Zitnick, and Martial Hebert. "Shuffle and learn: unsupervised learning using temporal order verification." ECCV 2016. [code] Temporal order of frames is exploited as the supervisory signal for learning.
  64. 64. 64 Temporal coherence (Slides by Xunyu Lin): Misra, Ishan, C. Lawrence Zitnick, and Martial Hebert. "Shuffle and learn: unsupervised learning using temporal order verification." ECCV 2016. [code] Take temporal order as the supervisory signals for learning Shuffled sequences Binary classification In order Not in order
  65. 65. 65 Temporal coherence Fernando, Basura, Hakan Bilen, Efstratios Gavves, and Stephen Gould. "Self-supervised video representation learning with odd-one-out networks." CVPR 2017 Train a network to detect which of the video sequences contains frames in the wrong order.
  66. 66. 66 Outline 1. Unsupervised Learning 2. Semi-supervised Learning 3. Self-supervised Learning a. Predictive Learning b. Learning by verification c. Cross-modal Learning
  67. 67. 67 Cross-modal Learning Representation or Embedding Encoder Decoder
  68. 68. 68 Cross-modal Learning Ephrat et al. Vid2speech: Speech Reconstruction from Silent Video. ICASSP 2017 CNN (VGG) Frame from a silent video Audio feature Post-hoc synthesis
  69. 69. bit.ly/DLCV2018 #DLUPC 69 Ephrat, Ariel, Tavi Halperin, and Shmuel Peleg. "Improved speech reconstruction from silent video." In ICCV 2017 Workshop on Computer Vision for Audio-Visual Media. 2017.
  70. 70. 70 Cross-modal Learning Representation or Embedding Encoder Decoder
  71. 71. 71 Cross-modal Learning Chung, Joon Son, Amir Jamaludin, and Andrew Zisserman. "You said that?." BMVC 2017.
  72. 72. bit.ly/DLCV2018 #DLUPC 72Chung, Joon Son, Amir Jamaludin, and Andrew Zisserman. "You said that?." BMVC 2017.
  73. 73. 73 Cross-modal Learning Suwajanakorn, Supasorn, Steven M. Seitz, and Ira Kemelmacher-Shlizerman. "Synthesizing Obama: learning lip sync from audio." SIGGRAPH 2017.
  74. 74. bit.ly/DLCV2018 #DLUPC 74 Karras, Tero, Timo Aila, Samuli Laine, Antti Herva, and Jaakko Lehtinen. "Audio-driven facial animation by joint end-to-end learning of pose and emotion." SIGGRAPH 2017
  75. 75. 75 Cross-modal Learning Karras, Tero, Timo Aila, Samuli Laine, Antti Herva, and Jaakko Lehtinen. "Audio-driven facial animation by joint end-to-end learning of pose and emotion." SIGGRAPH 2017
  76. 76. bit.ly/DLCV2018 #DLUPC 76 Karras, Tero, Timo Aila, Samuli Laine, Antti Herva, and Jaakko Lehtinen. "Audio-driven facial animation by joint end-to-end learning of pose and emotion." SIGGRAPH 2017
  77. 77. 77 Cross-modal Learning Label Annotator Encoder
  78. 78. 78 Aytar, Yusuf, Carl Vondrick, and Antonio Torralba. "Soundnet: Learning sound representations from unlabeled video." NIPS 2016.
  79. 79. 79 Cross-modal Learning Aytar, Yusuf, Carl Vondrick, and Antonio Torralba. "Soundnet: Learning sound representations from unlabeled video." NIPS 2016. Object & Scenes recognition in videos by analysing the audio track (only).
  80. 80. 80 Cross-modal Learning Representation or EmbeddingEncoder Encoder
  81. 81. 81 Cross-modal Learning Arandjelović, Relja, and Andrew Zisserman. "Look, Listen and Learn." ICCV 2017. Audio and visual features learned by assessing alignment.
  82. 82. 82 Cross-modal Learning Audio and visual features learned by assessing alignment. Arandjelović, Relja, and Andrew Zisserman. "Objects that Sound." ECCV 2018. Audio-visual Object localization Network
  83. 83. 83 Cross-modal Learning Harwath, David, Antonio Torralba, and James Glass. "Unsupervised learning of spoken language with visual context." NIPS 2016. [talk] Train a visual & speech networks with pairs of (non-)corresponding images & speech.
  84. 84. 84 Cross-modal Learning Harwath, David, Antonio Torralba, and James Glass. "Unsupervised learning of spoken language with visual context." NIPS 2016. [talk] Similarity curve show which regions of the spectrogram are relevant for the image. Important: no text transcriptions used during the training !!
  85. 85. 85 Cross-modal Learning Harwath, David, Adrià Recasens, Dídac Surís, Galen Chuang, Antonio Torralba, and James Glass. "Jointly Discovering Visual Objects and Spoken Words from Raw Sensory Input." ECCV 2018. Regions matching the spoken word “WOMAN”:
  86. 86. 86 Outline 1. Unsupervised Learning 2. Semi-supervised Learning 3. Self-supervised Learning a. Predictive Learning b. Learning by verification c. Cross-modal Learning
  87. 87. 87 Final Questions

×