AI-ordlista
Den kompletta ordlistan över AI
Uncertainty Threshold
Critical parameter determining the minimum level of uncertainty required for a stream sample to be selected for labeling by the human oracle.
Sampling Strategy
Algorithmic methodology used to optimally select instances from the stream deserving manual labeling according to predefined criteria.
Stream Classification
Process of assigning labels to sequentially arriving instances, with real-time and limited memory constraints.
Instance Selection
Discriminating mechanism identifying the most informative samples from the stream to maximize model improvement per unit of labeling.
Labeling Oracle
External source of information, typically human, providing correct labels for instances selected by the active learning system.
Online Evaluation
Method of continuously evaluating model performance measuring accuracy or other metrics as new instances arrive and are processed.
Temporal Adaptation
System's ability to dynamically modify its parameters and strategies to adjust to temporal evolutions of stream characteristics.
Exploration-Exploitation Trade-off
Fundamental dilemma consisting of arbitrating between exploring new regions of the space and exploiting acquired knowledge to optimize learning.
Labeling Latency
Temporal delay between the labeling request for an instance and the reception of the corresponding label, directly impacting streaming performance.
Stream Density
Measure characterizing the arrival frequency of samples in the stream, directly influencing selection strategies and labeling budget.
Sample Prioritization
Ranking algorithm for instances in the stream according to their expected information value to optimize labeling budget allocation.
Incremental Update
Technique for progressively adding new knowledge to the existing model without requiring complete retraining from scratch.
Rejection Strategy
Policy defining the conditions under which the system refuses to classify an instance due to excessive uncertainty or limited resources.
Acquisition Cost
Temporal, financial, or computational resources required to obtain a label from the oracle in an active learning context.