keyword
multi-label zero-shot learning
Multi-label zero-shot learning is a machine learning paradigm where a model learns to identify and assign multiple relevant category labels to a single data instance, such as an image or text document, even when some or all of those labels were never present in the training data. Unlike conventional zero-shot learning, which typically assumes each instance belongs to only one unseen class, this task addresses complex scenarios containing multiple co-occurring objects or concepts. To recognize novel categories without direct supervision, models map input features and auxiliary semantic information, such as word embeddings, class attributes, or vision-language representations, into a shared embedding space. This alignment enables the system to transfer knowledge from seen to unseen classes while modeling relationships, dependencies, and feature alignments across multiple labels simultaneously.
1 item
