ContinualLearningStrategyBase<T, TInput, TOutput>

Base Classes · AiDotNet.ContinualLearning.Strategies

Abstract base class for continual learning strategies providing common functionality.

Implements: IContinualLearningStrategy<T, TInput, TOutput>

Known derived types: ElasticWeightConsolidation<T, TInput, TOutput>, ExpectedGradientLength<T, TInput, TOutput>, GradientEpisodicMemory<T, TInput, TOutput>, LearningWithoutForgetting<T, TInput, TOutput>, MemoryAwareSynapses<T, TInput, TOutput>, PackNet<T, TInput, TOutput>, ProgressiveNeuralNetworks<T, TInput, TOutput>, SynapticIntelligence<T, TInput, TOutput>

For Beginners

This base class provides common functionality that all continual learning strategies share, such as:

How It Works

Derived classes implement specific algorithms like EWC, LwF, GEM, SI, and MAS.

Properties

Property Summary
bool ModifiesArchitecture { get; }
bool RequiresMemoryBuffer { get; }
long MemoryUsageBytes { get; }
string Name { get; }

Methods

Method Summary
IReadOnlyDictionary<string, object> GetMetrics()
T ComputeRegularizationLoss(IFullModel<T, TInput, TOutput> model)
Vector<T> AdjustGradients(IFullModel<T, TInput, TOutput> model, Vector<T> gradients)
bool IsCompatibleWith(IFullModel<T, TInput, TOutput> model)
string GetIncompatibilityReason(IFullModel<T, TInput, TOutput> model)
void FinalizeTask(IFullModel<T, TInput, TOutput> model)
void Load(string path)
void PrepareForTask(IFullModel<T, TInput, TOutput> model, IDataset<T, TInput, TOutput> taskData)
void Reset()
void Save(string path)