AI 용어집
인공지능 완전 사전
Secure Aggregation Protocol
Mechanism that allows aggregating contributions from multiple participants in a way that only the aggregated result is revealed, not the individual contributions. These protocols are essential in federated learning systems to protect the confidentiality of local models.
Privacy-Preserving Data Mining
Set of techniques that allows extracting useful knowledge from data while protecting the confidentiality of sensitive information contained in this data. These methods balance analytical needs with privacy protection requirements.
Secure Sum Protocol
Specific cryptographic protocol that allows multiple parties to compute the sum of their private values without revealing these individual values. This technique is fundamental in secure voting systems and privacy-preserving aggregated statistics.
Trusted Execution Environments (TEE)
Secure isolated zone within a processor ensuring that the code and data loaded inside are protected in confidentiality and integrity. TEEs allow performing computations on sensitive data even on potentially compromised platforms.
Secure Scalar Product Protocol
Mechanism that allows two parties to compute the scalar product of their private vectors without revealing these vectors. This operation is fundamental in many privacy-preserving machine learning algorithms.
Secure Matrix Multiplication
Protocol that allows multiple parties to compute the product of their private matrices without revealing these individual matrices. This operation is crucial for many privacy-preserving distributed machine learning algorithms.
Privacy-Preserving Machine Learning (PPML)
Set of techniques and protocols that enable the training and inference of machine learning models on sensitive data while guaranteeing their confidentiality. These methods combine cryptography, information theory, and distributed algorithms.
Secure Aggregation Trees
Tree structure used to organize the secure aggregation of data from many participants in an efficient and scalable manner. This approach minimizes communication overhead while maintaining confidentiality guarantees.
Threshold Cryptography
Schéma cryptographique où une clé privée est distribuée entre plusieurs participants et où un certain seuil d'entre eux doit collaborer pour effectuer des opérations cryptographiques. Cette technique élimine les points uniques de défaillance dans les systèmes sécurisés.
Secure Inner Product Computation
Protocole spécialisé pour calculer le produit interne entre vecteurs privés appartenant à différentes parties sans révéler ces vecteurs. Cette opération est fondamentale pour les algorithmes de classification et de similarité préservant la confidentialité.