SelfSupervisedLearningPretrainingResult<T>

Models & Types · AiDotNet.SelfSupervisedLearning

The outcome of a self-supervised pretraining pass: the loss trajectory, a representation-collapse check, and an optional linear-probe quality estimate.

How It Works

Beyond a plain pretraining loop, this reports whether the encoder collapsed — mapping every input to nearly the same vector, the classic BYOL/SimSiam failure that a falling loss can hide — and, when labels are available, a quick linear-probe score of how useful the learned representations are.

Constructors

Constructor Summary
SelfSupervisedLearningPretrainingResult()

Properties

Property Summary
IReadOnlyList<double> EpochLosses { get; set; } Gets the average loss per epoch, in order.
bool CollapseDetected { get; set; } Gets whether representation collapse was detected (the encoder’s outputs have near-zero spread).
double FinalLoss { get; set; } Gets the average loss of the final epoch.
double RepresentationStdDev { get; set; } Gets the mean per-dimension standard deviation of the encoder’s representations across a batch.
double? LinearProbeR2 { get; set; } Gets the in-sample linear-probe R² of the learned representations against the targets, or null when a probe was not run (no targets available).
int EpochsRun { get; set; } Gets the number of pretraining epochs run.
int StepsRun { get; set; } Gets the total number of training steps (batches) run.
string MethodName { get; set; } Gets the configured SSL method’s name.