Query Set

The small set of labeled examples provided to the model during a few-shot task. It serves as the model's only reference for defining new classes. In an N-way K-shot task, the support set contains K examples for each of the N classes. The model conditions its understanding on this set before making predictions on the query set.

A meta-learning training strategy where each training iteration simulates a complete few-shot task. An episode consists of sampling a support set and a query set from the overall dataset. The model learns to generalize across thousands of these simulated tasks, developing the ability to rapidly adapt to new classes at inference time.

A metric-based few-shot learning architecture that computes a prototype for each class by averaging the embeddings of its support set examples. Query samples are then classified by finding the nearest prototype in embedding space. The query set loss is computed as the negative log-probability of the correct class via softmax over distances.

A measure of similarity between two non-zero vectors calculated as the cosine of the angle between them. In few-shot learning, cosine similarity is often used to compare query embeddings against support set prototypes. It ranges from -1 (completely dissimilar) to 1 (identical), and is particularly effective when the magnitude of embeddings is less informative than their direction.

A probability value indicating the model's certainty in its prediction for a query sample. In prototypical networks, this is typically derived from the softmax over distances to all class prototypes. Low confidence scores on query set examples can signal out-of-distribution samples or ambiguous cases that require human review.

A classification paradigm where the model must correctly classify known classes while also identifying and rejecting unknown classes not present in the support set. Query set examples from unseen emitter types should trigger rejection rather than forced misclassification into a known class, a critical capability for spectrum surveillance applications.