One-Shot Learning

Published:

One-shot learning is used when you have exactly one labeled example of something new and still want the model to recognize it again. This comes up when examples are rare or expensive to label, such as identity checks or specialized inspection tasks. The model relies on what it learned earlier, then uses that single example as a reference for what the new category should look like.

This is different from zero-shot learning, which relies on a description and uses no examples at all. In one-shot learning, the example is the anchor. Many approaches treat the task as matching: the model compares a new input to the one reference example and decides whether they belong together. The quality of that single example matters a lot, since it sets the standard for everything the model will recognize next.

Follow us on Facebook and LinkedIn to keep abreast of our latest news and articles