AI-ordlista
Den kompletta ordlistan över AI
Data Heterogeneity
Significant variation in the distribution, quality, and quantity of data across different clients in a federated learning system, directly affecting the convergence of the global model.
System Heterogeneity
Divergences in computational capabilities, available memory, and energy resources of participating devices in a distributed federated learning network.
Non-IID
Non-independent and identically distributed data among federated clients, where each local dataset exhibits distinct statistical characteristics.
Weighted Federated Aggregation
Algorithm that combines local model updates using adaptive weights based on dataset size and the quality of contributions from each client.
Federated Personalization
Approach adapting a global model to the specificities of each client while benefiting from collaborative learning to maintain overall performance.
Availability Bias Compensation
Technique correcting imbalances in federated client participation due to variations in connectivity and resource availability.
Federated Stratification
Method organizing clients into homogeneous groups based on their data or system characteristics to optimize collaborative learning.
Federated Domain Adaptation
Process allowing the federated model to adjust to specific data distributions of each client domain while preserving shared knowledge.
Federated Meta-Learning
Learning paradigm where the federated model learns to learn quickly on new client data distributions with minimal local adaptation.
Robustness to Computational Disparities
Ability of the federated system to maintain optimal performance despite extreme variations in computing power among participants.
Federated Representation Alignment
Technique synchronizing feature spaces between heterogeneous clients to facilitate efficient aggregation of local model updates.
Federated Multi-Task Learning
Framework enabling simultaneous optimization of multiple client-specific tasks while sharing a common representation in the federated environment.
Adaptive Federated Quantization
Method dynamically adjusting the numerical precision of model updates according to each client's computational and communication capabilities.
Active Client Selection
Strategy optimizing the choice of participants in each federated learning round based on their heterogeneity and potential contribution.
Resilient Aggregation
Robust algorithm capable of effectively combining model updates from clients with highly heterogeneous distributions and performances.