Synthetic Data
Training examples generated rather than collected.
Обзор
Synthetic data is training material produced by a model or a generator instead of gathered from the world. It solves the problem that high-quality labelled data is the binding constraint on further training, and that some categories are rare or expensive to collect. It exists because a model can generate far more examples than people can label, with the caveat that the examples inherit whatever the generator gets wrong.
Граф технологий
Как Synthetic Data связана с остальным стеком.
Используется в
- Distillation technique Techniques Synthetic Data used by Distillation High ✓ source
Применяется к
- Fine-tuning technique Techniques Synthetic Data used for Fine-tuning High ✓ source