Security Advancements in Machine Learning

The field of machine learning is moving towards developing more secure and robust models, with a focus on detecting and mitigating various types of attacks. Researchers are exploring innovative approaches, such as using Variational Auto-encoders and cost-sensitive learning to enhance IoT-botnet detection, and developing adapters to protect machine learning models from competitive activity in network services. Additionally, there is a growing interest in investigating the vulnerability of popular models, such as Mixture of Experts, to backdoor attacks and developing methodologies to thwart Trojan attacks. Noteworthy papers include: MergeGuard, which proposes a novel methodology for mitigation of AI Trojan attacks, and Backdoor Attacks Against Patch-based Mixture of Experts, which investigates the vulnerability of patch-based MoE models to backdoor attacks and proposes fine-tuning as a defense.

Sources

Enhancing IoT-Botnet Detection using Variational Auto-encoder and Cost-Sensitive Learning: A Deep Learning Approach for Imbalanced Datasets

Development of an Adapter for Analyzing and Protecting Machine Learning Models from Competitive Activity in the Networks Services

Backdoor Attacks Against Patch-based Mixture of Experts

MergeGuard: Efficient Thwarting of Trojan Attacks in Machine Learning Models

Built with on top of