1 / 45

Greedy Layer-Wise Training of Deep Networks

Greedy Layer-Wise Training of Deep Networks. Yoshua Bengio , Pascal Lamblin , Dan Popovici , Hugo Larochelle NIPS 2007 Presented by Ahmed Hefny. Story so far ….

afi
Download Presentation

Greedy Layer-Wise Training of Deep Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Greedy Layer-Wise Training of Deep Networks YoshuaBengio, Pascal Lamblin, Dan Popovici, Hugo Larochelle NIPS 2007 Presented by Ahmed Hefny

  2. Story so far … • Deep neural nets are more expressive: Can learn wider classes of functions with less hidden units (parameters) and training examples. • Unfortunately they are not easy to train with randomly initialized gradient-based methods.

  3. Story so far … • Hinton et. al. (2006) proposed greedy unsupervised layer-wise training: • Greedy layer-wise: Train layers sequentially starting from bottom (input) layer. • Unsupervised: Each layer learns a higher-level representation of the layer below. The training criterion does not depend on the labels. • Each layer is trained as a Restricted Boltzman Machine. (RBM is the building block of Deep Belief Networks). • The trained model can be fine tuned using a supervised method. RBM 2 RBM 1 RBM 0

  4. This paper • Extends the concept to: • Continuous variables • Uncooperative input distributions • Simultaneous Layer Training • Explores variations to better understand the training method: • What if we use greedy supervised layer-wise training ? • What if we replace RBMs with auto-encoders ? RBM 2 RBM 1 RBM 0

  5. Outline • Review • Restricted Boltzman Machines • Deep Belief Networks • Greedy layer-wise Training • Supervised Fine-tuning • Extensions • Continuous Inputs • Uncooperative Input Distributions • Simultaneous Training • Analysis Experiments

  6. Outline • Review • Restricted Boltzman Machines • Deep Belief Networks • Greedy layer-wise Training • Supervised Fine-tuning • Extensions • Continuous Inputs • Uncooperative Input Distributions • Simultaneous Training • Analysis Experiments

  7. Restricted Boltzman Machine Undirected bipartite graphical model with connections between visible nodes and hidden nodes. Corresponds to joint probability distribution

  8. Restricted Boltzman Machine Undirected bipartite graphical model with connections between visible nodes and hidden nodes. Corresponds to joint probability distribution Factorized Conditionals

  9. Restricted Boltzman Machine (Training) • Given input vectors , adjust to increase

  10. Restricted Boltzman Machine (Training) • Given input vectors , adjust to increase

  11. Restricted Boltzman Machine (Training) • Given input vectors , adjust to increase Sample given Sample and using Gibbs sampling

  12. Restricted Boltzman Machine (Training) • Now we can perform stochastic gradient descent on data log-likelihood • Stop based on some criterion (e.g. reconstruction error )

  13. Deep Belief Network • A DBN is a model of the form denotes input variables denotes hidden layers of causal variables

  14. Deep Belief Network • A DBN is a model of the form denotes input variables denotes hidden layers of causal variables

  15. Deep Belief Network • A DBN is a model of the form denotes input variables denotes hidden layers of causal variables is an RBM RBM = Infinitely Deep network with tied weights

  16. Greedy layer-wise training • is intractable • Approximate with • Treat bottom two layers as an RBM • Fit parameters using contrastive divergence

  17. Greedy layer-wise training • is intractable • Approximate with • Treat bottom two layers as an RBM • Fit parameters using contrastive divergence • That gives an approximate • We need to match it with

  18. Greedy layer-wise training • Approximate • Treat layers as an RBM • Fit parameters using contrastive divergence • Sample recursively using starting from

  19. Outline • Review • Restricted Boltzman Machines • Deep Belief Networks • Greedy layer-wise Training • Supervised Fine-tuning • Extensions • Continuous Inputs • Uncooperative Input Distributions • Simultaneous Training • Analysis Experiments

  20. Supervised Fine Tuning (In this paper) • Use greedy layer-wise training to initialize weights of all layers except output layer. • For fine-tuning, use stochastic gradient descent of a cost function on the outputs where the conditional expected values of hidden nodes are approximated using mean-field.

  21. Supervised Fine Tuning (In this paper) • Use greedy layer-wise training to initialize weights of all layers except output layer. • Use backpropagation

  22. Outline • Review • Restricted Boltzman Machines • Deep Belief Networks • Greedy layer-wise Training • Supervised Fine-tuning • Extensions • Continuous Inputs • Uncooperative Input Distributions • Simultaneous Training • Analysis Experiments

  23. Continuous Inputs • Recall RBMs: • If we restrict then normalization gives us binomial with given by sigmoid. • Instead, if we get exponential density • If is closed interval then we get truncated exponential

  24. Continuous Inputs (Case for truncated exponential [0,1]) • Sampling For truncated exponential, inverse CDF can be used where is sampled uniformly from [0,1] • Conditional Expectation

  25. Continuous Inputs • To handle Gaussian inputs, we need to augment the energy function with a term quadratic in . • For a diagonal covariance matrix Giving

  26. Continuous Hidden Nodes ?

  27. Continuous Hidden Nodes ? • Truncated Exponential • Gaussian

  28. Uncooperative Input Distributions • Setting • No particular relation between p and f, (e.g. Gaussian and sinus)

  29. Uncooperative Input Distributions • Setting • No particular relation between p and f, (e.g. Gaussian and sinus) • Problem: Unsupvervised pre-training may not help prediction

  30. Outline • Review • Restricted Boltzman Machines • Deep Belief Networks • Greedy layer-wise Training • Supervised Fine-tuning • Extensions • Analysis Experiments

  31. Uncooperative Input Distributions • Proposal: Mix unsupervised and supervised training for each layer Temp. Ouptut Layer Stochastic Gradient of input log likelihood by Contrastive Divergence Stochastic Gradient of prediction error Combined Update

  32. Simultaneous Layer Training • Greedy Layer-wise Training • For each layer • Repeat Until Criterion Met • Sample layer input (by recursively applying trained layers to data) • Update parameters using contrastive divergence

  33. Simultaneous Layer Training • Simultaneous Training • Repeat Until Criterion Met • Sample input to all layers • Update parameters of all layers using contrastive divergence • Simpler: One criterion for the entire network • Takes more time

  34. Outline • Review • Restricted Boltzman Machines • Deep Belief Networks • Greedy layer-wise Training • Supervised Fine-tuning • Extensions • Continuous Inputs • Uncooperative Input Distributions • Simultaneous Training • Analysis Experiments

  35. Experiments • Does greedy unsupervised pre-training help ? • What if we replace RBM with auto-encoders ? • What if we do greedy supervised pre-training ? • Does continuous variable modeling help ? • Does partially supervised pre-training help ?

  36. Experiment 1 • Does greedy unsupervised pre-training help ? • What if we replace RBM with auto-encoders ? • What if we do greedy supervised pre-training ? • Does continuous variable modeling help ? • Does partially supervised pre-training help ?

  37. Experiment 1

  38. Experiment 1

  39. Experiment 1(MSE and Training Errors) Partially Supervised < Unsupervised Pre-training < No Pre-training Gaussian < Binomial

  40. Experiment 2 • Does greedy unsupervised pre-training help ? • What if we replace RBM with auto-encoders ? • What if we do greedy supervised pre-training ? • Does continuous variable modeling help ? • Does partially supervised pre-training help ?

  41. Experiment 2 • Auto Encoders • Learn a compact representation to reconstruct X • Trained to minimize reconstruction cross-entropy X X

  42. Experiment 2 20 nodes in last two layers (500~1000) layer width

  43. Experiment 2 • Auto-encoder pre-training outperforms supervised pre-training but is still outperformed by RBM. • Without pre-training, deep nets do not generalize well, but they can still fit the data if the output layers are wide enough.

  44. Conclusions • Unsupervised pre-training is important for deep networks. • Partial supervision further enhances results, especially when input distribution and the function to be estimated are not closely related. • Explicitly modeling conditional inputs is better than using binomial models.

  45. Thanks

More Related