AI-ordlista
Den kompletta ordlistan över AI
Concept Drift
Non-stationary change in the underlying data distribution or in the relationship between features and the target variable over time.
Adaptive Meta-Learning
Approach where the model learns to adapt by optimizing its own adaptation mechanisms, enabling a more effective response to detected changes.
Gradual Adaptation
Mechanism for progressively adjusting model parameters in response to slow and continuous changes in data distribution.
Drift Detector
Algorithm monitoring input data statistics or model performance to identify when adaptation is necessary.
Weighted Base Model
Ensemble of expert models whose contributions are dynamically weighted according to their relevance to the current data context.
Reinforcement Adaptation
Use of reinforcement learning principles to optimize online adaptation decisions based on environment feedback.
Selective Update
Strategy that modifies only critical model parameters upon change detection, thus preserving stability while ensuring adaptability.
Real-Time Adaptation
Ability of a system to modify its parameters instantly or with minimal latency upon arrival of new data or detection of change.
Forgetting Strategy
Mechanism controlling the decay of influence of older observations to allow the model to focus on recent patterns.
Hybrid Adaptation
Combination of multiple adaptation mechanisms (gradual and abrupt) to effectively manage different types and speeds of change in the data.
Continual Learning
Paradigm enabling the system to sequentially acquire new skills without compromising performance on previously learned tasks.
Adaptive Learning Rate
Mechanism dynamically adjusting the magnitude of parameter updates based on the extent and nature of detected change.
Adaptive Episodic Memory
Selective storage system of representative examples retained to facilitate future adaptation while efficiently managing memory space.