AI 용어집
인공지능 완전 사전
Continual Meta-Learning
Hybrid approach combining the principles of meta-learning and continual learning to enable models to acquire new skills while retaining previously learned knowledge.
Meta-Replay Buffer
Adaptive buffer that selectively stores past experiences based on their meta-information to optimize future continual learning.
Task-Agnostic Continual Learning
Learning scenario where the model must perform on all learned tasks without knowing the current task identity during inference.
Continual Adaptation
Ability of a meta-learned model to dynamically adapt to new data distributions without complete retraining.
Memory Synapse
Neuro-inspired mechanism allowing neural connections to selectively retain relevant information from previous tasks.
Continual Meta-Optimization
Continuous optimization process of hyperparameters and meta-parameters to maintain optimal balance between stability and plasticity.
Task Distribution Shift
Gradual or abrupt change in task distribution during continual learning, requiring meta-cognitive adaptation.
Meta-Continual Learning
Formal framework where meta-learning itself is performed continually, adapting fast learning capabilities over time.
Lifelong Meta-Learning
Long-term vision of continuous meta-learning aimed at developing systems capable of learning to learn throughout their operational lifecycle.
Continual Transfer Learning
Sequential transfer of knowledge between tasks where the model learns to effectively generalize learned representations to new contexts.
Meta-Knowledge Distillation
Process of compressing meta-knowledge acquired during continuous learning into a more compact model without performance loss.