Book Image

Hands-On Neural Networks with Keras

By : Niloy Purkait
Book Image

Hands-On Neural Networks with Keras

By: Niloy Purkait

Overview of this book

Neural networks are used to solve a wide range of problems in different areas of AI and deep learning. Hands-On Neural Networks with Keras will start with teaching you about the core concepts of neural networks. You will delve into combining different neural network models and work with real-world use cases, including computer vision, natural language understanding, synthetic data generation, and many more. Moving on, you will become well versed with convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory (LSTM) networks, autoencoders, and generative adversarial networks (GANs) using real-world training datasets. We will examine how to use CNNs for image recognition, how to use reinforcement learning agents, and many more. We will dive into the specific architectures of various networks and then implement each of them in a hands-on manner using industry-grade frameworks. By the end of this book, you will be highly familiar with all prominent deep learning models and frameworks, and the options you have when applying deep learning to real-world scenarios and embedding artificial intelligence as the core fabric of your organization.

Related Content you might be interested in

Current Title:

Small book image
Hands-On Neural Networks with Keras
Niloy Purkait
Mar, 2019 | 462 pages
Small book image
Hands-On Neural Networks
Leonardo De Marchi | Laura Mitchell
May, 2019 | 280 pages
Small book image
Deep Learning with Keras
Sujit Pal | Antonio Gulli
Apr, 2017 | 318 pages
Small book image
Deep Learning for Beginners
Dr Pablo Rivas
Sep, 2020 | 432 pages
Table of Contents (16 chapters)
Preface
Preface
Who this book is for
What this book covers
To get the most out of this book
Get in touch
Free Chapter
1
Section 1: Fundamentals of Neural Networks
Section 1: Fundamentals of Neural Networks
2
Overview of Neural Networks
Overview of Neural Networks
Defining our goal
Knowing our tools
The fundamentals of neural learning
The fundamentals of data science
Summary
Further reading
3
A Deeper Dive into Neural Networks
A Deeper Dive into Neural Networks
From the biological to the artificial neuron – the perceptron
Building a perceptron
Learning through errors
Training a perceptron
Backpropagation
Scaling the perceptron
A single layered network
Summary
4
Signal Processing - Data Analysis with Neural Networks
Signal Processing - Data Analysis with Neural Networks
Processing signals
Images as numbers
Feeding a neural network
Examples of tensors
Building a model
Compiling the model
Evaluating model performance
Implementing weight regularization in Keras
Weight regularization experiments
Implementing dropout regularization in Keras
Language processing
The internet movie reviews dataset
Plotting a single training instance
One-hot encoding
Vectorizing features
Vectorizing labels
Building a network
Callbacks
Accessing model predictions
Probing the predictions
Feature-wise normalization
Cross validation with scikit-learn API
Summary
Exercises
5
Section 2: Advanced Neural Network Architectures
Section 2: Advanced Neural Network Architectures
6
Convolutional Neural Networks
Convolutional Neural Networks
Why CNNs?
The birth of vision
Understanding biological vision
Conceptualizing spatial invariance
Defining receptive fields of neurons
Implementing a hierarchy of neurons
The birth of the modern CNN
Designing a CNN
The convolution operation
Visualizing feature extraction with filters
Looking at complex filters
Summarizing the convolution operation
Understanding pooling layers
Implementing CNNs in Keras
Convolutional layer
Leveraging a fully connected layer for classification
Summarizing our model
Checking model accuracy
The problem with detecting smiles
Introducing Keras's functional API
Verifying the number of channels per layer
Understanding saliency
Visualizing saliency maps with ResNet50
Loading pictures from a local directory
Using Keras's visualization module
Searching through layers
Exercise
Gradient weighted class activation mapping
Visualizing class activations with Keras-vis
Using the pretrained model for prediction
Visualizing maximal activations per output class
Converging a model
Using multiple filter indices to hallucinate
Problems with CNNs
Neural network pareidolia
Summary
7
Recurrent Neural Networks
Recurrent Neural Networks
Modeling sequences
Using RNNs for sequential modeling
Summarizing different types of sequence processing tasks
Predicting an output per time step
Backpropagation through time
Exploding and vanishing gradients
GRUs
Building character-level language models in Keras
Statistics of character modeling
The purpose of controlling stochasticity
Testing different RNN models
Building a SimpleRNN
Building GRUs
On processing reality sequentially
Bi-directional layer in Keras
Visualizing output values
Summary
Further reading
Exercise
8
Long Short-Term Memory Networks
Long Short-Term Memory Networks
On processing complex sequences
The LSTM network
Dissecting the LSTM
LSTM memory block
Visualizing the flow of information
Computing contender memory
Computing activations per timestep
Variations of LSTM and performance
Understanding peephole connections
Importance of timing and counting
Putting our knowledge to use
On modeling stock market data
Denoising the data
Implementing exponential smoothing
The problem with one-step-ahead predictions
Creating sequences of observations
Building LSTMs
Closing comments
Summary
Exercises
9
Reinforcement Learning with Deep Q-Networks
Reinforcement Learning with Deep Q-Networks
On reward and gratification
Conditioning machines with reinforcement learning
The explore-exploit dilemma
Path to artificial general intelligence
Simulating environments
A self-driving taxi cab
Trade-off between immediate and future rewards
Discounting future rewards
Markov decision process
Understanding policy functions
Assessing the value of a state
Assessing the quality of an action
Using the Bellman equation
Updating the Bellman equation iteratively
Why use neural networks?
Performing a forward pass in Q-learning
Performing a backward pass in Q-Learning
Deep Q-learning in Keras
Balancing exploration with exploitation
Initializing the deep Q-learning agent
Double Q-learning
Dueling network architecture
Exercise
Summary
10
Section 3: Hybrid Model Architecture
Section 3: Hybrid Model Architecture
11
Autoencoders
Autoencoders
Why autoencoders?
Automatically encoding information
Understanding the limitations of autoencoders
Breaking down the autoencoder
Training an autoencoder
Overviewing autoencoder archetypes
Network size and representational power
Understanding regularization in autoencoders
Regularization with sparse autoencoders
Probing the data
Building the verification model
Designing a deep autoencoder
Using functional API to design autoencoders
Deep convolutional autoencoder
Compiling and training the model
Testing and visualizing the results
Denoising autoencoders
Training the denoising network
Summary
Exercise
12
Generative Networks
Generative Networks
Replicating versus generating content
Understanding the notion of latent space
Diving deeper into generative networks
Using randomness to augment outputs
Sampling from the latent space
Understanding types of generative networks
Understanding VAEs
Designing a VAE in Keras
Building the encoding module in a VAE
Building the decoder module
Visualizing the latent space
Latent space sampling and output generation
Exploring GANs
Diving deeper into GANs
Designing a GAN in Keras
Designing the generator module
Designing the discriminator module
Putting the GAN together
The training function
Defining the discriminator labels
Training the generator per batch
Executing the training session
Conclusion
Summary
13
Section 4: Road Ahead
Section 4: Road Ahead
14
Contemplating Present and Future Developments
Contemplating Present and Future Developments
Sharing representations with transfer learning
Concluding our experiments
Learning representations
Limits of current neural networks
Encouraging sparse representation learning
Tuning hyperparameters
Automatic optimization and evolutionary algorithms
Multi-network predictions and ensemble models
The future of AI and neural networks
The road ahead
Problems with classical computing
The advent of quantum computing
Quantum neural networks
Technology and society
Contemplating our future
Summary
15
Other Books You May Enjoy
Other Books You May Enjoy
Leave a review - let other readers know what you think

Computing contender memory

We now know how the memory at time (t) is calculated, but how about the contender (c ̴t) itself? After all, it is partially responsible for maintaining a relevant state of memory, characterized by possibly useful representations occurring at each timestep.

This is the same idea that we saw in the GRU unit, where we allow the possibility for memory values to be updated using a contender value at each timestep. Earlier, with the GRU, we used a relevance gate that helped us compute it for the GRU. However, that is not necessary in the case of the LSTM, and we get a much simpler and arguably more elegant formulation as follows:

  • Contender memory value (c ̴t ) = tanh ( Wc [ at-1, t ] + bc)

Here, Wc is a weight matrix that is initialized at the beginning of a training session, and iteratively updated as the network trains. The dot product of...

Previous Section
End of Section 7
Next Section