AI 용어집
인공지능 완전 사전
Zero-Shot Learning (ZSL)
A model's ability to recognize classes never seen during training by using semantic descriptions.
Few-Shot Learning (FSL)
Learning with a very small number of examples per class (typically 1-5) to generalize to new tasks.
Meta-Learning
Learning to learn where the model acquires rapid generalization capabilities to adapt to new tasks.
Prompt Engineering
Optimized design of text prompts to guide models towards ZSL/FSL performance without retraining.
Prototype Networks
Architectures creating class representations (prototypes) from few examples for classification.
Metric Learning
Learning embedding spaces where similarity between examples is optimized for few-shot learning.
Cross-Lingual Zero-Shot
Knowledge transfer between languages to perform tasks in a language without training examples.
Attribute-Based Learning
Using decomposable semantic attributes to recognize unseen classes in ZSL.
Memory-Augmented Networks
Architectures with external memory to quickly store and retrieve information for few-shot learning.
Self-Supervised Pre-training
Self-supervised pre-training on unlabeled data to enhance the model's ZSL/FSL capabilities.
Knowledge Graph Integration
Incorporation of structured relationships between entities to enhance ZSL reasoning.
Relation Networks
Models learning to compare pairs of examples for few-shot classification.
Continual Learning for ZSL/FSL
Progressive adaptation to new classes/tasks without forgetting previous knowledge.
Data Augmentation for Few-Shot
Generative techniques to artificially augment small datasets in few-shot learning.
Transfer Learning Adaptation
Fine-tuning pre-trained models for optimal performance in ZSL/FSL.