Table of Contents

Class InfoGAN<T>

Namespace
AiDotNet.NeuralNetworks
Assembly
AiDotNet.dll

Represents an Information Maximizing Generative Adversarial Network (InfoGAN), which learns disentangled representations in an unsupervised manner by maximizing mutual information between latent codes and generated observations.

public class InfoGAN<T> : NeuralNetworkBase<T>, INeuralNetworkModel<T>, INeuralNetwork<T>, IFullModel<T, Tensor<T>, Tensor<T>>, IModel<Tensor<T>, Tensor<T>, ModelMetadata<T>>, IModelSerializer, ICheckpointableModel, IParameterizable<T, Tensor<T>, Tensor<T>>, IFeatureAware, IFeatureImportance<T>, ICloneable<IFullModel<T, Tensor<T>, Tensor<T>>>, IGradientComputable<T, Tensor<T>, Tensor<T>>, IJitCompilable<T>, IInterpretableModel<T>, IInputGradientComputable<T>, IDisposable

Type Parameters

T

The numeric type used for calculations, typically float or double.

Inheritance
InfoGAN<T>
Implements
IFullModel<T, Tensor<T>, Tensor<T>>
IModel<Tensor<T>, Tensor<T>, ModelMetadata<T>>
IParameterizable<T, Tensor<T>, Tensor<T>>
ICloneable<IFullModel<T, Tensor<T>, Tensor<T>>>
IGradientComputable<T, Tensor<T>, Tensor<T>>
Inherited Members
Extension Methods

Remarks

InfoGAN extends the GAN framework by: - Decomposing the input noise into incompressible noise (z) and latent codes (c) - Maximizing the mutual information I(c; G(z,c)) between codes and generated images - Learning interpretable and disentangled representations automatically - Using an auxiliary network Q to approximate the posterior P(c|x) - Enabling control over semantic features without labeled data

For Beginners: InfoGAN learns to separate different features automatically.

Key concept:

  • Splits random input into two parts:
    1. Random noise (z): provides variety
    2. Latent codes (c): control specific features
  • Learns what each code controls WITHOUT labels
  • Example: For faces, might learn codes for:
    • Code 1: controls rotation
    • Code 2: controls width
    • Code 3: controls lighting

How it works:

  • Generator uses both z and c to create images
  • Auxiliary network Q tries to predict c from the generated image
  • If Q can predict c accurately, the codes are meaningful
  • This forces codes to represent interpretable features

Use cases:

  • Discover semantic features in datasets
  • Disentangled representation learning
  • Controllable image generation
  • Feature manipulation (change one aspect, keep others)

Reference: Chen et al., "InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets" (2016)

Constructors

InfoGAN(NeuralNetworkArchitecture<T>, NeuralNetworkArchitecture<T>, NeuralNetworkArchitecture<T>, int, InputType, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>?, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>?, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>?, ILossFunction<T>?, double)

Initializes a new instance of the InfoGAN<T> class with the specified architecture and training parameters.

public InfoGAN(NeuralNetworkArchitecture<T> generatorArchitecture, NeuralNetworkArchitecture<T> discriminatorArchitecture, NeuralNetworkArchitecture<T> qNetworkArchitecture, int latentCodeSize, InputType inputType, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>? generatorOptimizer = null, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>? discriminatorOptimizer = null, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>? qNetworkOptimizer = null, ILossFunction<T>? lossFunction = null, double mutualInfoCoefficient = 1)

Parameters

generatorArchitecture NeuralNetworkArchitecture<T>

The architecture for the generator network.

discriminatorArchitecture NeuralNetworkArchitecture<T>

The architecture for the discriminator network.

qNetworkArchitecture NeuralNetworkArchitecture<T>

The architecture for the Q network (should output latentCodeSize values).

latentCodeSize int

The size of the latent code (number of controllable features).

inputType InputType

The type of input data (e.g., ThreeDimensional for images).

generatorOptimizer IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>

Optional optimizer for the generator. If null, an Adam optimizer with default settings is created.

discriminatorOptimizer IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>

Optional optimizer for the discriminator. If null, an Adam optimizer with default settings is created.

qNetworkOptimizer IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>

Optional optimizer for the Q network. If null, an Adam optimizer with default settings is created.

lossFunction ILossFunction<T>

Optional loss function. If null, the default loss function for generative tasks is used.

mutualInfoCoefficient double

The coefficient for mutual information loss. Higher values prioritize feature learning. Default is 1.0.

Remarks

This constructor creates an InfoGAN with three networks: - Generator: Creates images from noise and latent codes - Discriminator: Determines if images are real or fake - Q Network: Predicts latent codes from generated images

The mutual information loss encourages the generator to use the latent codes in meaningful ways that can be recovered by the Q network.

For Beginners: InfoGAN learns controllable features automatically: - The generator creates images using random noise + controllable codes - The Q network tries to guess which codes were used - This forces the codes to represent real, interpretable features - After training, you can manipulate specific features by changing the codes

Exceptions

ArgumentNullException

Thrown when any of the architecture parameters is null.

ArgumentOutOfRangeException

Thrown when latentCodeSize is not positive or mutualInfoCoefficient is negative.

Properties

Discriminator

Gets the discriminator network.

public ConvolutionalNeuralNetwork<T> Discriminator { get; }

Property Value

ConvolutionalNeuralNetwork<T>

Generator

Gets the generator network.

public ConvolutionalNeuralNetwork<T> Generator { get; }

Property Value

ConvolutionalNeuralNetwork<T>

ParameterCount

Gets the total number of trainable parameters in the InfoGAN.

public override int ParameterCount { get; }

Property Value

int

QNetwork

Gets the auxiliary Q network that predicts latent codes from images.

public ConvolutionalNeuralNetwork<T> QNetwork { get; }

Property Value

ConvolutionalNeuralNetwork<T>

Remarks

The Q network shares most of its parameters with the discriminator (up to the last layer). It outputs the predicted latent code distribution given an image. This network is key to maximizing mutual information.

For Beginners: The Q network is the "feature detector".

  • Takes an image as input
  • Outputs: "I think these codes were used to make this"
  • Training makes Q better at guessing codes
  • This forces generator to use codes meaningfully

Methods

CreateNewInstance()

Creates a new instance of the InfoGAN with the same configuration.

protected override IFullModel<T, Tensor<T>, Tensor<T>> CreateNewInstance()

Returns

IFullModel<T, Tensor<T>, Tensor<T>>

A new InfoGAN instance with the same architecture and hyperparameters.

Remarks

This method creates a fresh InfoGAN instance with the same network architectures and hyperparameters. The new instance has freshly initialized optimizers.

For Beginners: This method creates a copy of the InfoGAN structure but with new, untrained networks and fresh optimizers.

DeserializeNetworkSpecificData(BinaryReader)

Deserializes InfoGAN-specific data from a binary reader.

protected override void DeserializeNetworkSpecificData(BinaryReader reader)

Parameters

reader BinaryReader

The binary reader to read from.

Remarks

This method deserializes the InfoGAN-specific configuration and all three networks. After deserialization, the optimizers are reset to their initial state.

For Beginners: This method loads the InfoGAN's settings and all three networks (generator, discriminator, and Q network) from a file.

Generate(Tensor<T>, Tensor<T>)

Generates images with specific latent codes.

public Tensor<T> Generate(Tensor<T> noise, Tensor<T> latentCodes)

Parameters

noise Tensor<T>

Random noise.

latentCodes Tensor<T>

Latent codes to control generation.

Returns

Tensor<T>

Generated images.

GenerateRandomLatentCodes(int)

Generates random latent codes (continuous, uniform in [-1, 1]).

public Tensor<T> GenerateRandomLatentCodes(int batchSize)

Parameters

batchSize int

Returns

Tensor<T>

GenerateRandomNoiseTensor(int, int)

Generates random noise tensor using vectorized Gaussian noise generation with CPU/GPU acceleration.

public Tensor<T> GenerateRandomNoiseTensor(int batchSize, int noiseSize)

Parameters

batchSize int

The number of noise samples in the batch.

noiseSize int

The size of each noise sample.

Returns

Tensor<T>

A tensor of shape [batchSize, noiseSize] filled with Gaussian noise.

GetModelMetadata()

Gets the metadata for this neural network model.

public override ModelMetadata<T> GetModelMetadata()

Returns

ModelMetadata<T>

A ModelMetaData object containing information about the model.

InitializeLayers()

Initializes the layers of the neural network based on the architecture.

protected override void InitializeLayers()

Remarks

For Beginners: This method sets up all the layers in your neural network according to the architecture you've defined. It's like assembling the parts of your network before you can use it.

Predict(Tensor<T>)

Makes a prediction using the neural network.

public override Tensor<T> Predict(Tensor<T> input)

Parameters

input Tensor<T>

The input data to process.

Returns

Tensor<T>

The network's prediction.

Remarks

For Beginners: This is the main method you'll use to get results from your trained neural network. You provide some input data (like an image or text), and the network processes it through all its layers to produce an output (like a classification or prediction).

ResetOptimizerState()

Resets the state of all optimizers to their initial values.

public void ResetOptimizerState()

Remarks

This method resets all three optimizers (generator, discriminator, and Q network) to their initial state. This is useful when restarting training or when you want to clear accumulated momentum and adaptive learning rate information.

For Beginners: Call this method when you want to start fresh with training, as if the model had never been trained before. The network weights remain unchanged, but the optimizer's memory of past gradients is cleared.

SerializeNetworkSpecificData(BinaryWriter)

Serializes InfoGAN-specific data to a binary writer.

protected override void SerializeNetworkSpecificData(BinaryWriter writer)

Parameters

writer BinaryWriter

The binary writer to write to.

Remarks

This method serializes the InfoGAN-specific configuration and all three networks. Optimizer state is managed by the optimizer implementations themselves.

For Beginners: This method saves the InfoGAN's settings and all three networks (generator, discriminator, and Q network) to a file.

Train(Tensor<T>, Tensor<T>)

Trains the neural network on a single input-output pair.

public override void Train(Tensor<T> input, Tensor<T> expectedOutput)

Parameters

input Tensor<T>

The input data.

expectedOutput Tensor<T>

The expected output for the given input.

Remarks

This method performs one training step on the neural network using the provided input and expected output. It updates the network's parameters to reduce the error between the network's prediction and the expected output.

For Beginners: This is how your neural network learns. You provide: - An input (what the network should process) - The expected output (what the correct answer should be)

The network then:

  1. Makes a prediction based on the input
  2. Compares its prediction to the expected output
  3. Calculates how wrong it was (the loss)
  4. Adjusts its internal values to do better next time

After training, you can get the loss value using the GetLastLoss() method to see how well the network is learning.

TrainStep(Tensor<T>, Tensor<T>, Tensor<T>)

Performs one training step for InfoGAN.

public (T discriminatorLoss, T generatorLoss, T mutualInfoLoss) TrainStep(Tensor<T> realImages, Tensor<T> noise, Tensor<T> latentCodes)

Parameters

realImages Tensor<T>

Real images.

noise Tensor<T>

Random noise (z).

latentCodes Tensor<T>

Latent codes (c) to condition generation.

Returns

(T Precision, T Recall, T F1Score)

Tuple of (discriminator loss, generator loss, mutual info loss).

Remarks

InfoGAN training: 1. Train discriminator (standard GAN objective) 2. Train generator with GAN loss + mutual information loss 3. Train Q network to predict latent codes from generated images

For Beginners: One round of InfoGAN training.

Steps:

  1. Generate images using noise + latent codes
  2. Train discriminator to spot fakes (standard GAN)
  3. Train generator to fool discriminator
  4. Train Q network to guess the codes from images
  5. Make generator use codes that Q can predict

UpdateParameters(Vector<T>)

Updates the parameters of all networks in the InfoGAN.

public override void UpdateParameters(Vector<T> parameters)

Parameters

parameters Vector<T>

The new parameters vector containing parameters for all networks.

Edit this page