KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Client Bias
Systematic distortion in model performance due to uneven distribution of data or resources among different clients participating in the federated system.
Non-IID Heterogeneity
Assessment of the degree of statistical variation between data distributions of different clients, directly impacting the generalization performance of the federated model.
Distribution Drift
Measurement of temporal changes in the statistical characteristics of client data during federated training, requiring model adaptations.
Federated Algorithmic Fairness
Set of indicators evaluating whether the federated model provides fair and non-discriminatory performance across different client groups.
Aggregation Robustness
Ability of the federated system to maintain stable performance despite the presence of malicious or failing clients in the aggregation process.
Federated Convergence Score
Composite indicator combining convergence speed, stability, and energy efficiency to comprehensively evaluate the performance of the federated process.
Disparity Metric
Quantification of performance gaps between clients, measuring inequality in model prediction quality across the federated network.
Communication Round Time
Measured duration for a complete cycle of information exchange between the server and all participating clients, including network latency and processing time.
Distributed Computing Cost
Aggregate measure of computational resources consumed by all clients during federated training, normalized by achieved performance.
Inter-Client Fairness
Evaluation of the balance in contribution and benefit received by each client participating in the federated learning system.
Aggregation Stability
Measure of the consistency of aggregated model updates across multiple training rounds, indicating the reliability of the federated process.
Federated Generalization Performance
Ability of the federated model to maintain high performance on new and unseen data from clients different from those used for training.
Communication Efficiency
Ratio between model performance improvement and volume of data transmitted, optimizing bandwidth usage in federated systems.
Client Participation Rate
Percentage of active clients participating in each training round, directly impacting the robustness and representativeness of the federated model.
Federated Fairness Score
Composite indicator simultaneously measuring algorithmic fairness, equitable resource distribution, and equitable access to model benefits.
Aggregation Latency
Delay between receiving local updates and publishing the aggregated model, affecting the responsiveness of the federated system.
Model Coherence Metric
Measure of similarity between client local models and the aggregated global model, indicating the degree of uniformity in distributed learning.