AI 용어집
인공지능 완전 사전
TensorFlow Federated
Open-source framework developed by Google for federated learning, integrated with TensorFlow and enabling simulation and deployment of distributed learning algorithms on decentralized data.
PySyft
Open-source Python library for federated learning and privacy preservation, combining PyTorch with cryptography and secure secret sharing techniques.
Flower Framework
ML framework-agnostic federated learning framework, offering a flexible architecture for deploying federated learning systems with heterogeneous clients and customizable aggregation strategies.
FedML
Research and production library for federated learning, MLOps and benchmarking, providing a comprehensive platform for developing and deploying large-scale federated learning applications.
OpenFL
Open-source framework for federated learning in healthcare and medical environments, developed by Intel and specifically designed for HIPAA privacy and compliance requirements.
Substrato
Open-source federated learning platform based on Kubernetes, enabling orchestration of distributed learning workloads with enhanced isolation and security between participants.
LEAF
Open-source benchmark for federated learning, providing realistically partitioned datasets and reference implementations for evaluating algorithms under heterogeneous conditions.
FedAvg
Fundamental aggregation algorithm in federated learning performing weighted averaging of client local models based on their dataset sizes, serving as the basis for many variants.
FedProx
Extension of FedAvg adding a proximal regularization term to the local objective to handle system and statistical heterogeneity between clients, improving convergence in non-IID environments.
SCAFFOLD
Federated learning algorithm correcting client-server gradient drift by using control variables to stabilize training in highly heterogeneous scenarios.
FedBN
Federated learning technique preserving local Batch Normalization statistics for each client, allowing better management of non-IID data distributions among participants.
FedOpt
Family of server optimizers for federated learning applying adaptive optimization methods like Adam or Yogi directly to the global model aggregation process.
FedPer
Federated learning approach separating model parameters into base (globally shared) and personalized (local), enabling effective customization while preserving collaboration.
Aggregation Server
Central component in federated architecture collecting local updates from clients, executing the aggregation algorithm and distributing the updated global model to participants.
Client Library
Set of client-side software tools implementing communication protocols, secure local training and model lifecycle management in a federated learning system.
Secure Aggregation
Cryptographic protocol allowing the server to aggregate model updates without accessing individual client updates, thus preserving the confidentiality of local contributions.
Homomorphic Encryption
Cryptographic technique allowing computations to be performed directly on encrypted data, used in federated learning to aggregate models without ever decrypting client contributions.