Zero-shot learning is a machine learning paradigm in which a trained model is evaluated on classes it never encountered during training. Instead of relying on labeled examples of target classes, the model bridges the training-test gap through auxiliary semantic information—such as attribute descriptions, word embeddings, or ontological relationships—that maps known classes to unknown ones in a shared semantic space. The mechanism of persistence is the structured semantic representation itself: once a vocabulary of attributes or embeddings exists, it can be reused across arbitrary zero-shot transfers without retraining. The approach emerged as deep learning models gained access to rich pre-trained embeddings (e.g., Word2Vec, GloVe), enabling semantic generalization beyond visual features. [formal: cognitio nullis shotis | substrate: mind | horizon: hours | explicit: yes | epoch: 0.05]
Accepted ontology entry
zero-shot-learning
Zero-shot learning is a machine learning paradigm in which a trained model is evaluated on classes it never encountered during training. Instead of relying on labeled examples of target classes, the model bridges the training-test gap thro…
Definition
Why it is in scope
A human-made machine learning paradigm wherein a model performs tasks or classifies instances it has never encountered during training, without any task-specific examples. The concept relies on transfer of knowledge from related source domains and rich feature representations learned during pretraining. It emerged as a formal concept within representation learning and natural language processing, building on the broader paradigm of machine learning.
Names and aliases
- zero-shot-learningen · CANONICAL
Relations from this entry
- cmrg0scos00ef2a1nklfvbk7xINSTANCE_OF →
zero-shot-learning IS a specific kind of machine learning. The ML model generalizes to unseen classes during evaluation — this is a specific learning paradigm within the broader machine learning field.
- cmsdcks2103r73vv3v8qlmtjiDERIVED_FROM →
Zero-shot learning emerged after supervised learning was established. Supervised learning provided the foundational paradigm of learning from labeled examples; zero-shot learning extends this by learning tasks without any labeled examples for target classes, building on transferred concepts from supervised-pretrained models.
Relations to this entry
No accepted relations in this direction.
Record identity
- Created
- Aug 10, 2026, 6:02 AM UTC
- Content hash
- c509812b4b65ebba95ec4f3bbe4e60b309f01931559e042a4e7d6589a09577e8