Glosarium AI
Kamus lengkap Kecerdasan Buatan
Continual Domain Adaptation
Learning process where a model progressively adapts to new domains while retaining its competencies on previously learned domains without access to the initial data.
Weight Replication
Regularization technique that penalizes significant changes to network weights for parameters important to previous tasks.
Continual Knowledge Distillation
Method where the current model learns to mimic the predictions of its previous version to preserve knowledge while adapting to the new domain.
Continual Meta-Learning
Approach where the model learns to learn efficiently on new domains by developing fast and generalizable adaptation strategies.
Dynamic Network Expansion
Architecture that selectively adds new capacities (neurons or layers) for new domains while preserving existing structures.
Generative Replay
Technique using generative models to synthetically recreate data from previous domains and maintain performance during continual adaptation.
Accumulated Domain Bias
Phenomenon where biases from previous domains accumulate and negatively affect performance on new domains in a continual adaptation scenario.
Plasticity-Stability Dilemma
Fundamental constraint between a model's ability to adapt to new information (plasticity) and its ability to retain existing knowledge (stability).
Continual Reinforcement Adaptation
Method where agents continuously learn to adapt to new environments using progressive feedback signals.
Continual Task Selection
Mechanism enabling the model to identify and dynamically specialize in relevant domains among a continuous sequence of new domains.