Advances in AI Security and Anomaly Detection

The field of AI security and anomaly detection is rapidly evolving, with a focus on developing innovative methods to protect against malicious attacks and identify unusual patterns in data. Recent research has explored the use of generative models, diffusion-based approaches, and explainable AI techniques to improve the accuracy and efficiency of anomaly detection systems. Additionally, there is a growing interest in developing training-free and zero-shot methods that can adapt to new datasets and scenarios without requiring extensive retraining. These advances have significant implications for real-world applications, including computer vision, medical imaging, and cybersecurity. Noteworthy papers in this area include AEDR, which proposes a novel training-free attribution method for generative models, and OCSVM-Guided Representation Learning, which introduces a custom loss formulation for unsupervised anomaly detection. Other notable works include MaXsive, which presents a high-capacity and robust training-free generative image watermarking technique, and DISTIL, which proposes a data-free trigger-inversion strategy for detecting Trojan attacks.

Sources

AEDR: Training-Free AI-Generated Image Attribution via Autoencoder Double-Reconstruction

Explainable AI guided unsupervised fault diagnostics for high-voltage circuit breakers

On Anti-collusion Codes for Averaging Attack in Multimedia Fingerprinting

Hot-Swap MarkBoard: An Efficient Black-box Watermarking Approach for Large-scale Model Distribution

Singularity Cipher: A Topology-Driven Cryptographic Scheme Based on Visual Paradox and Klein Bottle Illusions

OCSVM-Guided Representation Learning for Unsupervised Anomaly Detection

MaXsive: High-Capacity and Robust Training-Free Generative Image Watermarking in Diffusion Models

Staining and locking computer vision models without retraining

Zero-Shot Image Anomaly Detection Using Generative Foundation Models

DISTIL: Data-Free Inversion of Suspicious Trojan Inputs via Latent Diffusion

Out-of-Distribution Detection in Medical Imaging via Diffusion Trajectories

DiffuMatch: Category-Agnostic Spectral Diffusion Priors for Robust Non-rigid Shape Matching

SUB: Benchmarking CBM Generalization via Synthetic Attribute Substitutions

Built with on top of