GenerativeReplay<T>

Models & Types · AiDotNet.ContinualLearning

Implements Generative Replay (also known as Deep Generative Replay) for continual learning.

Implements: IContinualLearningStrategy<T>

For Beginners

Generative Replay uses a generative model (like a VAE or GAN) to create pseudo-examples from previous tasks instead of storing real examples. This enables rehearsal without storing actual data, which is useful for privacy-sensitive applications or when memory is limited.

How It Works

How it works:

Key Components:

Advantages:

Reference: Shin, H., Lee, J.K., Kim, J., and Kim, J. “Continual Learning with Deep Generative Replay” (2017). NeurIPS.

Constructors

Constructor Summary
GenerativeReplay(int replayBatchSize, double replayRatio, double lambda, int? seed) Initializes a new instance of the GenerativeReplay class.

Properties

Property Summary
bool AccumulatesAcrossTasks { get; }
double Lambda { get; set; }
double ReplayRatio { get; } Gets the replay ratio.
int ReplayBatchSize { get; } Gets the replay batch size.
int TaskCount { get; } Gets the number of tasks processed.

Methods

Method Summary
T ComputeLoss(INeuralNetwork<T> network)
ValueTuple<Tensor<T>, Tensor<T>> CreateMixedBatch(Tensor<T> currentInputs, Tensor<T> currentTargets, int batchSize) Creates a mixed training batch combining new data with generated replay data.
ValueTuple<Tensor<T>, Tensor<T>> GenerateReplaySamples() Generates pseudo-examples for replay using the generator.
Vector<T> ModifyGradients(INeuralNetwork<T> network, Vector<T> gradients)
void AfterTask(INeuralNetwork<T> network, ValueTuple<Tensor<T>, Tensor<T>> taskData, int taskId)
void BeforeTask(INeuralNetwork<T> network, int taskId)
void Reset()
void SetGenerator(IGenerativeModel<T> generator) Sets the generative model used for replay.