GEMTrainerOptions<T>
Options & Configuration · AiDotNet.ContinualLearning.Trainers
Configuration options for the GEM trainer.
For Beginners
These options control how the GEM trainer operates. GEM prevents forgetting by ensuring gradients don’t increase loss on previous tasks.
Constructors
| Constructor | Summary |
|---|---|
GEMTrainerOptions() |
Properties
| Property | Summary |
|---|---|
bool? ComputeValidationMetrics { get; set; } |
Whether to compute validation metrics after each epoch. |
bool? UseAveragedGEM { get; set; } |
Whether to use the average gradient (A-GEM) instead of full GEM. |
double? ConstraintTolerance { get; set; } |
Tolerance for constraint satisfaction in gradient projection. |
double? GradientClipThreshold { get; set; } |
Gradient clipping threshold. |
double? MemoryStrength { get; set; } |
Memory strength parameter (gamma) for gradient projection. |
double? ProjectionMargin { get; set; } |
Margin for gradient projection. |
int? ExamplesPerTask { get; set; } |
Number of examples to store per task for gradient computation. |
int? MaxQPIterations { get; set; } |
Maximum number of iterations for quadratic programming solver. |