Generative adversarial network

As explained in Evolutionary Deep Learning: Genetic algorithms and neural networks

A generative adversarial network is a model consisting of a generator and discriminator trained in opposition to produce realistic synthetic data.

A generative adversarial network (GAN) is a generative model built from two competing networks: a generator creates synthetic data, while a discriminator classifies samples as real or fake. Their opposition can produce realistic representations, but training is notoriously difficult because both networks must improve at roughly the same rate. section 9.1.1 section 9.6

The generator receives a randomized latent space with added noise and produces samples. The discriminator sees real samples alongside generated samples, learns from its prediction feedback, and supplies feedback that helps the generator produce more convincing outputs. If either network learns substantially faster, the adversarial relationship can break down. section 9.1.1 section 9.2

Inspect these recognizable training artifacts when diagnosing a GAN:

• Vanishing gradients occur when an overly strong discriminator provides too little loss feedback for useful generator updates. section 9.2.1
• Mode collapse occurs when the generator repeatedly produces nearly the same output with little variation. section 9.2.1
• Failure to converge occurs when the generator improves too quickly and overwhelms the discriminator, reducing its predictions to random guesses. section 9.2.1

Training and experimenting with a basic GAN

Implement a basic GAN with Python and Keras, train it to generate realistic digits from a chosen class, then vary batch size, generator and discriminator learning rates, or dataset class selection. Extend the experiment by evolving training hyperparameters, reusing an evolved model on a similarly sized dataset, or testing evolutionary strategies and differential evolution. section 9.1.3 section 9.5.1 section 9.6

Figure 9.1 Classic explanation of a GAN

Use the art-forger and art-detective diagram to visualize how generated fakes challenge the discriminator while its feedback trains the generator. section 9.1.1

How Evolutionary Deep Learning: Genetic algorithms and neural networks covers it

How deep it goes

Explains network roles, feedback, implementation, artifacts, and experiments, but stops short of extensive mathematical derivation.

Examples

Demonstrates art forgery, class-specific handwritten digits, face classification, and experiments with GAN hyperparameters and datasets.

What it assumes

Introduces the basic GAN from an art-forger analogy and reviews its operation before implementation; assumes no further prerequisite explicitly.

What you can do after

Build it from scratch with Python and Keras, inspect training failures, and experiment with evolutionary hyperparameter optimization.

How it connects

Connects generative modeling with latent space, classifiers, Wasserstein loss, genetic algorithms, and evolutionary optimization.

liveBookmanning.com

Read Evolutionary Deep Learning: Genetic algorithms and neural networks on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Inside Deep Learning: Math, Algorithms, ModelsA generative deep-learning model consisting of a generator and discriminator trained adversarially to produce realistic samples.
• 

  liveBookmanning.com
  Learn Generative AI with PyTorchA generative adversarial network, a model family trained with a generator and discriminator to synthesize realistic data or transform data between domains.
• 

  liveBookmanning.com
  Outlier Detection in PythonA generative adversarial network is a model consisting of a generator and discriminator trained competitively so the generator produces data resembling a reference distribution.
• 

  liveBookmanning.com
  Grokking Machine LearningA generative adversarial network is a generative model composed of competing generator and discriminator networks that learn to produce realistic data.
• 

  liveBookmanning.com
  Sutskever's ListA generative adversarial network is a model architecture in which a generator creates candidates and a discriminator learns to distinguish them from genuine data.
• 

  liveBookmanning.com
  Generative AI in ActionA generative adversarial network is a generative model composed of a generator and discriminator trained adversarially to produce realistic synthetic data.
• 

  liveBookmanning.com
  Data Without LabelsA generative adversarial network is a generative model consisting of a generator and discriminator trained adversarially to produce realistic samples.
• 

  liveBookmanning.com
  Machine Learning for Tabular Data: XGBoost, Deep Learning, and AIA generative adversarial network is a deep learning architecture consisting of a generator and discriminator trained adversarially to produce samples resembling real data.
• 

  liveBookmanning.com
  Introduction to Generative AI, Second EditionA generative adversarial network is a model composed of a generator and discriminator trained through competition between synthetic-data creation and authenticity evaluation.
• 

  liveBookmanning.com
  Grokking AI Algorithms, Second EditionA generative adversarial network is a machine-learning model consisting of a generator and a discriminator trained through competition to produce increasingly realistic synthetic data.
• 

  liveBookmanning.com
  Build a Text-to-Image Generator (from Scratch)A generative adversarial network is a generative model trained through competition between a generator that creates samples and a discriminator that distinguishes generated samples from real data.

Compare how 12 books explain generative adversarial network →

Related concepts

• generatorhas part · A generator is a computational component or construct that produces data or values from inputs, representations, context, or specified constraints.
• discriminatorhas part · A discriminator is the model component of a generative adversarial network that distinguishes real data from data generated by its generator.
• VQGANincludes · VQGAN is a generative model that combines vector-quantized discrete latent representations with adversarial and perceptual objectives to synthesize images.
• generative modelingkind of · Generative modeling is a machine-learning approach that learns patterns or a data distribution to generate new samples resembling the data.
• DCGAN / deep convolutional ganincludes · A deep convolutional generative adversarial network is a GAN whose generator and discriminator use deep convolutional neural-network architectures.
• generationkind of · Generation is the computational production of outputs from supplied inputs or context, often by applying a model to produce predictions or new content.

This explainer was generated by Manning's knowledge graph from the text of Evolutionary Deep Learning: Genetic algorithms and neural networks. The section links point to the parts of the book it draws on.