Tags → #machine learning

Vulnerabilities
Researcher Proposes Watermarking Method for Deep Neural Networks
Nov 22, 2024 Ikuya Morikawa from Fujitsu Limited has published a paper proposing a new defense mechanism called bounding-box watermarking to protect deep neural networks from model extraction attacks.
Tools
New Model MERLOT Developed for Encrypted Traffic Classification
Nov 21, 2024 A recent study presents MERLOT, a scalable mixture-of-expert model designed for the efficient classification of encrypted traffic while reducing computational costs.
Tools
ProSec Introduces New Approach for Secure Code Generation
Nov 21, 2024 A new paper introduces ProSec, a proactive security alignment approach that enhances code-specific large language models by synthesizing error-inducing coding scenarios to improve secure coding practices.
Vulnerabilities
Research Highlights Vulnerabilities in Large Language Models Security
Nov 21, 2024 A new paper examines the vulnerabilities of large language models in enterprise settings, emphasizing the need for a comprehensive understanding of security risks across software and hardware components in compound AI systems.
Tools
New Defense Mechanism Proposed Against Model Poisoning in Federated Learning
Nov 20, 2024 Researchers from the University of Electronic Science and Technology of China have introduced a new defense mechanism called VERT to combat model poisoning attacks in Federated Learning systems.
Vulnerabilities
Study Examines Data Privacy Risks in Graph Neural Networks
Nov 20, 2024 Researchers from The Pennsylvania State University and Hong Kong University of Science and Technology have published a study addressing the risks of data leakage in Graph Neural Networks (GNNs) and introduced a framework called GraphSteal to reconstruct training graphs while safeguarding privacy.
Tools
Study Introduces New Model for Data Breach Prediction
Nov 20, 2024 A recent study by researchers from Moroccan universities has developed a predictive system for data breaches that incorporates social media data, enhancing prediction accuracy significantly.
Tools
Neural Network-Based Cryptography Advances with RANDOM Model
Nov 19, 2024 Neural Network-based cryptography is being developed as a potential alternative to traditional methods, aiming to enhance encryption security through unique, adaptable schemes.
Tools
New Concept of Defendability Against Backdoors in Machine Learning
Nov 19, 2024 Researchers from the Alignment Research Center have proposed a new formal concept of defendability against backdoors in machine learning models, framing it as a strategic game between an attacker and a defender.
Vulnerabilities
Study Analyzes Security Vulnerabilities in Large Language Models
Nov 19, 2024 A recent study highlights the vulnerabilities of large language models, particularly focusing on a novel jailbreak method called SQL Injection Jailbreak, which demonstrates high success rates in inducing harmful content.