Somanath Tripathy, Harsh Kasyap, Minghong Fang | Federated learning. Security and privacy (2026) [PDF, EPUB] [EN]

Contents
CHAPTER 1. Introduction to Machine Learning
1.1 TYPES OF LEARNING
1.2 LEARNING TASKS
1.3 COST FUNCTION
1.4 OPTIMIZATION
1.4.1 Undertting
1.4.2 Overtting
1.4.3 Regularization
1.5 EVALUATION METRICS
1.5.1 Regression Metrics
1.5.2 Classication Metrics
1.5.3 Clustering Metrics
1.6 ARTIFICIAL NEURAL NETWORK
1.6.1 Convolutional Neural Network
1.7 IMPLEMENTATION
1.7.1 CNN Model
1.7.2 Training
1.7.3 Evaluation

CHAPTER 2. Federated Learning
2.1 DEFINITION OF FL
2.2 IMPORTANCE OF FL
2.3 TYPES OF FL
2.3.1 Cross-Device and Cross-Silo FL
2.3.2 FL Based on Data Partitioning
2.4 APPLICATIONS OF FL
2.5 CHALLENGES IN FL
2.6 SECURITY AND PRIVACY ISSUES IN FL
2.6.1 Poisoning Attacks
2.6.2 Inference Attacks
2.7 DEFENSE TECHNIQUES
2.7.1 Byzantine-Robust FL
2.7.2 Inference-Resistant FL
2.8 IMPLEMENTATION
2.8.1 Federated Average
2.8.2 Model Training and Evaluation
2.8.3 Setup
2.8.4 FL Cycle

CHAPTER 3. Poisoning Attacks on FL
3.1 ATTACKER GOAL
3.2 LABEL FLIPPING ATTACK
3.3 GAUSSIAN ATTACK
3.4 LIE ATTACK
3.5 KRUM ATTACK
3.6 TRIM ATTACK
3.7 SHEJWALKAR ATTACK
3.7.1 Min-Max Attack
3.7.2 Min-Sum Attack
3.8 SINE ATTACK
3.8.1 Attack Procedure
3.9 SCALING ATTACK
3.10 EDGE ATTACK
3.11 VULNERABILITIES IN COSINE SIMILARITY-BASED DEFENSES
3.12 IMPLEMENTATION
3.12.1 Helper Libraries
3.12.2 Label Flipping Attack
3.12.3 Trim Attack
3.12.4 LIE Attack
3.12.5 Min-max Attack
3.12.6 Min-sum Attack
3.12.7 SINE Poisoning Attack
3.12.8 HDC-based Data Poisoning Attack

CHAPTER 4. Inference Attacks on FL
4.1 ATTACKER GOAL
4.2 DATA RECONSTRUCTION ATTACKS
4.2.1 DLG Attack
4.2.2 Hitaj Attack
4.3 MEMBERSHIP INFERENCE ATTACKS
4.3.1 Shokri Attack
4.3.2 Nasr Attack
4.3.3 Zhang Attack
4.4 PROPERTY INFERENCE ATTACK
4.4.1 MILSA: Model Interpretation-Based Label Sni
5.5 FOOLSGOLD
5.5.1 Defense Execution
5.5.2 Adaptive Attack Analysis
5.6 FLTRUST
5.6.1 Trust Score
5.6.2 Normalization
5.6.3 Aggregation
5.7 MOAT
5.7.1 Overview
5.7.2 Detection of Poisoned Clients and Labels
5.8 DeFL
5.8.1 Federated Gradient Norm Vector (FGNV)
5.8.2 Detecting Malicious Clients
5.8.3 Evaluation
5.9 RDFL
5.9.1 Adaptive Clustering
5.9.2 Detecting Malicious Models
5.10 FLTC: FL TRUSTED COORDINATES
5.10.1 Trusted Coordinates
5.10.2 Adaptive Attack
5.10.3 Evaluation
5.11 IMPLEMENTATION
5.11.1 FoolsGold
5.11.2 FLTrust
5.11.3 Krum
5.11.4 Multi-Krum
5.11.5 Median 5.11.6 Trimmed-Mean
5.11.7 DnC

CHAPTER 6. Privacy-Preserving FL
6.1 DIFFERENTIAL PRIVACY
6.1.1 DPFL: A Client-Level Perspective
6.2 HOMOMORPHIC ENCRYPTION
6.2.1 BatchCrypt: HE-Based Scheme
6.2.2 Threshold Multi-Key HE Scheme
6.3 SECURE MULTI-PARTY COMPUTATION
6.3.1 Practical Secure Aggregation
6.4 IMPLEMENTATION
6.4.1 Differential Privacy
6.4.2 Homomorphic Encryption
6.4.3 Practical Secure Aggregation
Bibliography
Index

About the Author

Somanath Tripathy received his PhD from IIT Guwahati in 2007. Currently, he is a professor in the Department of Computer Science and Engineering at the Indian Institute of Technology, Patna, where he has been a faculty member since December 2008. His research interests encompass Cybersecurity, Malware Detection, Secure Machine Learning, Lightweight Cryptography, and Blockchain. Tripathy holds two patents and has published over 130 research papers in reputed journals and conferences.

Harsh Kasyap is an Assistant Professor in the Department of Computer Science and Engineering at the Indian Institute of Technology (BHU), Varanasi, India. He is also an honorary research fellow at WMG, University of Warwick, UK. He obtained his Ph.D. from the IIT Patna, India. His research interests are Federated Learning, Machine Learning Security, Trustworthy AI, Privacy and Data Security.

Minghong Fang is a tenure-track Assistant Professor in the Department of Computer Science and Engineering at the University of Louisville. He has published several high-impact research papers in top-tier security conferences, including the USENIX Security Symposium, the ACM Conference on Computer and Communications Security (CCS), and the Network and Distributed System Security (NDSS) Symposium. His research interests broadly span various aspects of AI safety and security.