ContinualLearnerConfig<T>

Options & Configuration · AiDotNet.ContinualLearning.Config

Production-ready configuration for continual learning algorithms.

Implements: IContinualLearnerConfig<T>

For Beginners

This class provides all settings needed for continual learning. All properties have industry-standard defaults set in the constructor.

How It Works

Usage Example:

Industry Standard Defaults:

  • Learning Rate: 0.001 (Adam optimizer standard)
  • Batch Size: 32 (balance of speed and gradient noise)
  • EWC Lambda: 1000 (based on Kirkpatrick et al. 2017)
  • Distillation Temperature: 2.0 (based on Li and Hoiem 2017)

Constructors

Constructor Summary
ContinualLearnerConfig() Initializes a new instance with industry-standard default values.

Properties

Property Summary
MemorySamplingStrategy MemoryStrategy { get; set; }
T AGemMargin { get; set; }
T BiCValidationFraction { get; set; }
T DistillationTemperature { get; set; }
T DistillationWeight { get; set; }
T EwcLambda { get; set; }
T GemMemoryStrength { get; set; }
T GradientClipNorm { get; set; }
T HatSmax { get; set; }
T HatSparsity { get; set; }
T LearningRate { get; set; }
T MasLambda { get; set; }
T OnlineEwcGamma { get; set; }
T PackNetPruneRatio { get; set; }
T PnnLateralScaling { get; set; }
T SiC { get; set; }
T SiXi { get; set; }
T WeightDecay { get; set; }
bool ComputeBackwardTransfer { get; set; }
bool ComputeForwardTransfer { get; set; }
bool ICarlUseHerding { get; set; }
bool NormalizeFisher { get; set; }
bool PnnUseLateralConnections { get; set; }
bool UseEmpiricalFisher { get; set; }
bool UseGradientClipping { get; set; }
bool UsePrioritizedReplay { get; set; }
bool UseSoftTargets { get; set; }
bool UseWeightDecay { get; set; }
int AGemReferenceGradients { get; set; }
int BatchSize { get; set; }
int EpochsPerTask { get; set; }
int EvaluationFrequency { get; set; }
int FisherSamples { get; set; }
int ICarlExemplarsPerClass { get; set; }
int MaxTasks { get; set; }
int MemorySize { get; set; }
int PackNetRetrainEpochs { get; set; }
int SamplesPerTask { get; set; }
int? RandomSeed { get; set; }

Methods

Method Summary
ContinualLearnerConfig<T> ForEwc(T lambda, int? fisherSamples) Creates a configuration optimized for EWC strategy.
ContinualLearnerConfig<T> ForExperienceReplay(int? memorySize, MemorySamplingStrategy? strategy, bool? prioritized) Creates a configuration optimized for experience replay.
ContinualLearnerConfig<T> ForGem(T memoryStrength, int? memorySize) Creates a configuration optimized for GEM strategy.
ContinualLearnerConfig<T> ForLwf(T temperature) Creates a configuration optimized for LwF strategy with default distillation weight.
ContinualLearnerConfig<T> ForLwf(T temperature, T weight) Creates a configuration optimized for LwF strategy.
bool IsValid()