• Proceedings of the Korea Information Processing Society Conference
• /
• 2024.05a
• /
• pp.749-752
• /
• 2024

연합학습은 클라이언트가 중앙 서버에 원본 데이터를 주지 않고도 학습할 수 있도록 설계된 분산된 머신러닝 방법이다. 그러나 클라이언트와 중앙 서버 사이에 모델 업데이트 정보를 공유한다는 점에서 여전히 추론 공격(Inference Attack)과 오염 공격(Poisoning Attack)의 위험에 노출되어 있다. 이러한 공격을 방어하기 위해 연합학습에 차분프라이버시(Differential Privacy)를 적용하는 방안이 연구되고 있다. 차분 프라이버시는 데이터에 노이즈를 추가하여 민감한 정보를 보호하면서도 유의미한 통계적 정보 쿼리는 공유할 수 있도록 하는 기법으로, 노이즈를 추가하는 위치에 따라 전역적 차분프라이버시(Global Differential Privacy)와 국소적 차분 프라이버시(Local Differential Privacy)로 나뉜다. 이에 본 논문에서는 차분 프라이버시를 적용한 연합학습의 최신 연구 동향을 전역적 차분 프라이버시를 적용한 방향과 국소적 차분 프라이버시를 적용한 방향으로 나누어 검토한다. 또한 이를 세분화하여 차분 프라이버시를 발전시킨 방식인 적응형 차분 프라이버시(Adaptive Differential Privacy)와 개인화된 차분 프라이버시(Personalized Differential Privacy)를 응용하여 연합학습에 적용한 방식들에 대하여 특징과 장점 및 한계점을 분석하고 향후 연구방향을 제안한다.

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.6
• /
• pp.65-72
• /
• 2020

Today, the Internet of Things is used in many places, including homes, industrial sites, and hospitals, to give us convenience. Many services generate new value through real-time data collection, storage and analysis as devices are connected to the network. Many of these fields are creating services and applications that utilize sensors and communication functions within IoT devices. However, since everything can be hacked, it causes a huge privacy threat to users who provide data. For example, a variety of sensitive information, such as personal information, lifestyle patters and the existence of diseases, will be leaked if data generated by smarwatches are abused. Development of IoT must be accompanied by the development of security. Recently, Differential Privacy(DP) was adopted to privacy-preserving data processing. So we propose the method that can aggregate health data safely on smartwatch platform, based on DP.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.2
• /
• pp.405-416
• /
• 2022

Current AI technology is improving the quality of life by using machine learning based on data. When using machine learning, transmitting distributed data and collecting it in one place goes through a de-identification process because there is a risk of privacy infringement. De-identification data causes information damage and omission, which degrades the performance of the machine learning process and complicates the preprocessing process. Accordingly, Google announced joint learning in 2016, a method of de-identifying data and learning without the process of collecting data into one server. This paper analyzed the effectiveness by comparing the difference between the learning performance of data that went through the de-identification process of K anonymity and differential privacy reproduction data using actual financial data. As a result of the experiment, the accuracy of original data learning was 79% for k=2, 76% for k=5, 52% for k=7, 50% for 𝜖=1, and 82% for 𝜖=0.1, and 86% for Federated learning.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.34 no.1
• /
• pp.143-156
• /
• 2024

Federated Learning (FL) has emerged as a potent methodology for decentralized model training across multiple collaborators, eliminating the need for data sharing. Although FL is lauded for its capacity to preserve data privacy, it is not impervious to various types of privacy attacks. Differential Privacy (DP), recognized as the golden standard in privacy-preservation techniques, is widely employed to counteract these vulnerabilities. This paper makes a specific contribution by applying an existing, task-specific adaptive DP mechanism to the FL environment. Our comprehensive analysis evaluates the impact of this mechanism on the performance of a shared global model, with particular attention to varying data distribution and partitioning schemes. This study deepens the understanding of the complex interplay between privacy and utility in FL, providing a validated methodology for securing data without compromising performance.