Test-time training – Definition & Meaning

Definition of test-time training

: a method that updates a model at deployment time using a learning signal derived from the unlabeled test input or stream before or while producing predictions
- The vision model used test-time training to adapt to images from a camera it had not encountered during development.
- Test-time training improved robustness under distribution shift by learning from each incoming unlabeled sample.

Origin & history

first known use2020

Yu Sun and collaborators proposed Test-Time Training as a general method in a 2020 ICML paper. Their approach turned each unlabeled test sample into a self-supervised learning problem and updated model parameters before prediction to improve generalization under distribution shift.

Test yourself

Which of these is the meaning of test-time training?

Related terms

training the process of feeding data to a model and repeatedly adjusting its parameters so its outputs… fine-tuning further training of an existing model on a smaller, focused dataset so it gets better at a… test-time compute extra computing power spent while answering a question, for example by letting the model think… inference the act of running a trained model to produce an output, such as answering a prompt or… on-device AI AI that runs directly on a phone, laptop, car, or other device instead of in the cloud, which… reinforcement learning a way of training in which a model tries actions, receives rewards or penalties, and gradually… distribution shift a difference between the statistical distribution of the data used to train or evaluate a…

Cite this entry

"Test-time training." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/test-time-training/. Accessed 6 Oct. 2026.