Noisy Student Training
Definition
Noisy-Student Training combines teacher predictions with noisy learning of a student model.
A self-training method where a teacher model generates pseudolabels and a larger or noised student model learns on labeled and unlabeled data.
Simply explained
One model provides preliminary labels; a second learns from them and additional noise.
Why it matters in practice
Helps realistically assess model quality, limits, and technical decisions.
Example from business
An image classifier uses unlabeled product photos with pseudolabels and strong augmentation.
Limits and pitfalls
- Amplifying faulty pseudolabels or failing to control comparison data and data provenance.
Origin & history
Xie and co-authors published Noisy Student in 2019.
Use in mid-sized companies
- Possible application: An image classifier uses unlabeled product photos with pseudolabels and strong augmentation.
Frequently asked questions
What does Noisy-Student Training mean?
A self-training method where a teacher model generates pseudolabels and a larger or noised student model learns on labeled and unlabeled data.
Which mistakes should be avoided?
Amplifying faulty pseudolabels or failing to control comparison data and data provenance.
Sources
Source checked on 10.10.2026. Editorial application examples are hypothetical; no proof of real customer results.
Editorially reviewed on: