book

Advanced Deep Learning with TensorFlow 2 and Keras - Second Edition

by Rowel Atienza

February 2020

Intermediate to advanced

512 pages

11h 47m

English

Packt Publishing

Read now

Unlock full access

Who this book is forWhat this book coversTo get the most out of this bookDownload the example code filesDownload the color imagesConventions usedGet in touchReviews
1. Why is Keras the perfect deep learning library?Installing Keras and TensorFlow2. MLP, CNN, and RNNThe differences between MLP, CNN, and RNN3. Multilayer Perceptron (MLP)The MNIST datasetThe MNIST digit classifier modelBuilding a model using MLP and KerasRegularizationOutput activation and loss functionOptimizationPerformance evaluationModel summary4. Convolutional Neural Network (CNN)ConvolutionPooling operationsPerformance evaluation and model summary5. Recurrent Neural Network (RNN)6. Conclusion7. References
1. Functional APICreating a two-input and one-output model2. Deep Residual Network (ResNet)3. ResNet v24. Densely Connected Convolutional Network (DenseNet)Building a 100-layer DenseNet-BC for CIFAR105. Conclusion6. References
1. Principles of autoencoders2. Building an autoencoder using Keras3. Denoising autoencoders (DAEs)4. Automatic colorization autoencoder5. Conclusion6. References
1. An Overview of GANsPrinciples of GANs2. Implementing DCGAN in Keras3. Conditional GAN4. Conclusion5. References
1. Wasserstein GANDistance functionsDistance function in GANsUse of Wasserstein lossWGAN implementation using Keras2. Least-squares GAN (LSGAN)3. Auxiliary Classifier GAN (ACGAN)4. Conclusion5. References
1. Disentangled representationsInfoGANImplementation of InfoGAN in KerasGenerator outputs of InfoGAN2. StackedGANImplementation of StackedGAN in KerasGenerator outputs of StackedGAN4. Conclusion5. References
1. Principles of CycleGANThe CycleGAN modelImplementing CycleGAN using KerasGenerator outputs of CycleGANCycleGAN on MNIST and SVHN datasets2. Conclusion3. References
1. Principles of VAEVariational inferenceCore equationOptimizationReparameterization trickDecoder testingVAE in KerasUsing CNN for AE2. Conditional VAE (CVAE)3. 𝛽-VAE – VAE with disentangled latent representations4. Conclusion5. References
1. Principles of Reinforcement Learning (RL)2. The Q value3. Q-learning exampleQ-Learning in Python4. Nondeterministic environment5. Temporal-difference learningQ-learning on OpenAI Gym6. Deep Q-Network (DQN)DQN on KerasDouble Q-learning (DDQN)7. Conclusion8. References

1. Policy gradient theorem2. Monte Carlo policy gradient (REINFORCE) method3. REINFORCE with baseline method4. Actor-Critic method5. Advantage Actor-Critic (A2C) method6. Policy Gradient methods using Keras7. Performance evaluation of policy gradient methods8. Conclusion9. References
1. Object detection2. Anchor boxes3. Ground truth anchor boxes4. Loss functions5. SSD model architecture6. SSD model architecture in Keras7. SSD objects in Keras8. SSD model in Keras9. Data generator model in Keras10. Example dataset11. SSD model training12. Non-Maximum Suppression (NMS) algorithm13. SSD model validation14. Conclusion15. References
1. Segmentation2. Semantic segmentation network3. Semantic segmentation network in Keras4. Example dataset5. Semantic segmentation validation6. Conclusion7. References
1. Mutual Information2. Mutual Information and Entropy3. Unsupervised learning by maximizing the Mutual Information of discrete random variables4. Encoder network for unsupervised clustering5. Unsupervised clustering implementation in Keras6. Validation using MNIST7. Unsupervised learning by maximizing the Mutual Information of continuous random variables8. Estimating the Mutual Information of a bivariate Gaussian9. Unsupervised clustering using continuous random variables in Keras10. Conclusion11. References

Content preview from Advanced Deep Learning with TensorFlow 2 and Keras - Second Edition

10 Policy Gradient Methods

In this chapter, we're going to introduce algorithms that directly optimize the policy network in reinforcement learning. These algorithms are collectively referred to as policy gradient methods. Since the policy network is directly optimized during training, the policy gradient methods belong to the family of on-policy reinforcement learning algorithms. Like value-based methods, which we discussed in Chapter 9, Deep Reinforcement Learning, policy gradient methods can also be implemented as deep reinforcement learning algorithms.

A fundamental motivation in studying the policy gradient methods is addressing the limitations of Q-learning. We'll recall that Q-learning is about selecting the action that maximizes the ...