SimCSEType
Enums · AiDotNet.Enums
Defines the training paradigms for SimCSE (Simple Contrastive Learning of Sentence Embeddings).
For Beginners
Think of this as the “learning style” of the model.
- Unsupervised is like a student learning by comparing different versions of the same book.
- Supervised is like a student learning from a teacher who provides “true” and “false” examples.
How It Works
SimCSE supports two primary modes:
- Unsupervised: Uses dropout masks on identical sentence pairs as a minimal data augmentation.
- Supervised: Uses labeled entailment and contradiction pairs from datasets like SNLI or MultiNLI.
Fields
| Field | Summary |
|---|---|
SimCSEType Supervised |
Supervised learning using Natural Language Inference (NLI) datasets. |
SimCSEType Unsupervised |
Unsupervised learning using dropout as noise. |
int value__ |