AI 용어집
인공지능 완전 사전
Homomorphic Encryption
Allows performing mathematical calculations directly on encrypted data without decrypting them beforehand
Differential Privacy
Technique that adds controlled noise to data to protect individual information while preserving statistical properties
Federated Learning
Distributed training approach where models improve locally without centralizing sensitive data
Adversarial Attacks and Defenses
Study of malicious manipulations of models and development of countermeasures to maintain their robustness
Secure Multi-Party Computation
Cryptographic protocols that allow multiple parties to collaborate on a computation without revealing their private data
Privacy Mechanisms
Set of formal techniques ensuring information protection in learning systems
Model Evasion
Detection and Prevention Techniques for Sensitive Information Leaks from Trained Models
Local Differential Privacy
Variant of differential privacy applied directly at the level of individual data before collection
Attribute Inference
Protection against attacks aimed at inferring sensitive information from model outputs
Model Robustness
Reinforcement of machine learning models to resist manipulations and maintain their performance
Anonymization and K-Anonymity
Identity masking techniques ensuring that each record cannot be distinguished from at least k-1 others
Secure Model Downloading
Protocols for protecting models during their transfer and deployment in unsecured environments
Secure Aggregation
Methods for securely combining results from multiple sources without compromising confidentiality
Private Cross-Validation
Model evaluation techniques without exposing original training or test data
Data Perturbation
Controlled modification of training data to prevent information leakage while preserving utility