Artificial Intelligence
Generalisation
A model’s ability to use relationships learned during training to produce useful results on new examples under similar conditions.
Simple example
A digit model trained on varied handwriting correctly reads a new image of 7 that was not used in training. This illustrates generalisation.
Why does it matter?
Success on training examples alone does not establish real-world performance. Check examples kept separate from training that represent actual use. Generalisation is not a guarantee of a correct result in every new situation.