book

Deep Learning Illustrated: A Visual, Interactive Guide to Artificial Intelligence

by Jon Krohn, Grant Beyleveld, Aglaé Bassens

September 2019

Intermediate to advanced

416 pages

13h 49m

English

Addison-Wesley Professional

Read now

Unlock full access

How to Read This Book
Biological VisionMachine VisionTensorFlow PlaygroundQuick, Draw!Summary
Deep Learning for Natural Language ProcessingComputational Representations of LanguageElements of Natural Human LanguageGoogle DuplexSummary
A Boozy All-NighterArithmetic on Fake Human FacesStyle Transfer: Converting Photos into Monet (and Vice Versa)Make Your Own Sketches PhotorealisticCreating Photorealistic Images from TextImage Processing Using Deep LearningSummary
Deep Learning, AI, and Other BeastsThree Categories of Machine Learning ProblemsDeep Reinforcement LearningVideo GamesBoard GamesManipulation of ObjectsPopular Deep Reinforcement Learning EnvironmentsThree Categories of AISummary
PrerequisitesInstallationA Shallow Network in KerasSummary
Biological Neuroanatomy 101The PerceptronModern Neurons and Activation FunctionsChoosing a NeuronSummaryKey Concepts
The Input LayerDense LayersA Hot Dog-Detecting Dense NetworkThe Softmax Layer of a Fast Food-Classifying NetworkRevisiting Our Shallow NetworkSummaryKey Concepts
Cost FunctionsOptimization: Learning to Minimize CostBackpropagationTuning Hidden-Layer Count and Neuron CountAn Intermediate Net in KerasSummaryKey Concepts
Weight InitializationUnstable GradientsModel Generalization (Avoiding Overfitting)Fancy OptimizersA Deep Neural Network in KerasRegressionTensorBoardSummaryKey Concepts
Convolutional Neural NetworksPooling LayersLeNet-5 in KerasAlexNet and VGGNet in KerasResidual NetworksApplications of Machine VisionSummaryKey Concepts
Preprocessing Natural Language DataCreating Word Embeddings with word2vecThe Area under the ROC CurveNatural Language Classification with Familiar NetworksNetworks Designed for Sequential DataNon-sequential Architectures: The Keras Functional APISummaryKey Concepts
Essential GAN TheoryThe Quick, Draw! DatasetThe Discriminator NetworkThe Generator NetworkThe Adversarial NetworkGAN TrainingSummaryKey Concepts
Essential Theory of Reinforcement LearningEssential Theory of Deep Q-Learning NetworksDefining a DQN AgentInteracting with an OpenAI Gym EnvironmentHyperparameter Optimization with SLM LabAgents Beyond DQNSummaryKey Concepts
Ideas for Deep Learning ProjectsResources for Further ProjectsThe Modeling Process, Including Hyperparameter TuningDeep Learning LibrariesSoftware 2.0Approaching Artificial General IntelligenceSummary
PyTorch FeaturesPyTorch in Practice

Content preview from Deep Learning Illustrated: A Visual, Interactive Guide to Artificial Intelligence

7. Artificial Neural Networks

In Chapter 6, we examined the intricacies of artificial neurons. The theme of the current chapter is the natural extension of that: We cover how individual neural units are linked together to form artificial neural networks, including deep learning networks.

The Input Layer

In our Shallow Net in Keras Jupyter notebook (a schematic of which is available in Figure 5.4), we crafted an artificial neural network with the following layers:

An input layer consisting of 784 neurons, one for each of the 784 pixels in an MNIST image
A hidden layer composed of 64 sigmoid neurons
An output layer consisting of 10 softmax neurons, one for each of the 10 classes of digits

Of these three, the input layer is the most straightforward ...