ISelfSupervisedModel

Interfaces · AiDotNet.Interfaces

Marker interface for models that train themselves from unlabeled data — the model owns its training objective rather than fitting a supervised target.

Known implementations: DiffusionModelBase<T>

For Beginners

“Self-supervised” means the model makes its own learning signal out of the raw data (for diffusion, by adding noise and learning to remove it) — there are no separate labels. Implement this on a generative model so the facade trains it correctly through ConfigureModel(...).BuildAsync().

How It Works

Diffusion models are the canonical example: their objective is noise prediction, not Predict(X) ≈ Y. The supervised optimizer path (clone-evaluate-select against a target) cannot express that. When a configured model implements this interface, AiModelBuilder routes it to a self-supervised training loop that runs epochs over the data calling the model’s own Train per sample.