Zero‑shot Learning

Zero‑shot learning is a machine‑learning paradigm in which a system can correctly handle inputs belonging to categories that were never presented during its training phase. Instead of relying on direct examples, the model leverages auxiliary knowledge—such as textual descriptions, attribute vectors, or relationships learned from other tasks—to map novel instances into a shared representation space where they can be compared to known concepts.

The value of zero‑shot learning lies in its ability to sidestep the costly data collection and labeling required for every possible class a system might encounter. By generalizing from what it has already seen, a model can rapidly adapt to new vocabularies, product lines, or visual categories without retraining, which is crucial for applications that must stay current in fast‑moving domains or operate under strict compute constraints.

Zero‑shot techniques appear wherever the set of target classes is incomplete or highly dynamic. In computer vision they enable image classifiers to recognize objects described only by their attributes; in natural language processing they let large language models follow instructions describing unseen tasks; and in multimodal retrieval they allow matching between text and images for concepts that lack explicit training pairs. Across these settings, the core idea remains the same: use learned representations to bridge the gap between known data and novel queries.