Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Zero-Shot Classification
Classification technique where the model infers labels for unknown classes by mapping data features to semantic descriptions of classes.
Semantic Transfer
Process of transferring knowledge between semantic domains to generalize to unobserved classes during training.
Semantic Spaces
Continuous vector representations where semantic concepts are projected to enable comparisons and knowledge transfers.
Semantic Attributes
Binary or continuous descriptive characteristics that describe object properties and serve as a bridge between seen and unseen classes.
Word Embeddings
Dense vector representations of words capturing their semantic relationships, used as prior knowledge in zero-shot learning.
Vision-Language Models
Unified neural architectures capable of understanding and simultaneously linking visual content and textual descriptions.
Extreme Generalization
Ability of a model to perform on data distributions radically different from those used during training.
Learning with Limited Data
Learning scenario where labeled data is scarce or non-existent for certain target classes.
Inference on Unknown Classes
Prediction process where the model must classify inputs into categories never encountered during training.
Knowledge Transfer
Mechanism that allows reusing knowledge acquired on a source task to solve a different target task.
Semantic Mapping
Mathematical function that projects visual features into a semantic space shared with textual descriptions.
Meta-learning
Learning approach where the model learns to learn, optimizing its ability to quickly adapt to new tasks with few examples.
Multi-modal Projection
Technique that aligns different modalities (text, image, audio) in a common vector space for knowledge transfer.