The field of machine learning is moving towards developing more robust and reliable models, with a focus on adversarial robustness and anomaly detection. Recent research has introduced novel methods for detecting and mitigating adversarial attacks, such as the use of prediction inconsistency and dynamic epsilon scheduling. Additionally, there have been significant advancements in anomaly detection, including the development of new frameworks for multi-label classification and video anomaly detection. These innovations have the potential to improve the performance and trustworthiness of machine learning models in a variety of applications. Noteworthy papers include Confidential Guardian, which proposes a framework for preventing the misuse of cautious predictions, and INP-Former++, which achieves state-of-the-art performance in single-class, multi-class, and few-shot anomaly detection tasks. Overall, the field is advancing towards more secure and dependable machine learning models.
Advancements in Adversarial Robustness and Anomaly Detection
Sources
Learning Normal Patterns in Musical Loops
Confidential Guardian: Cryptographically Prohibiting the Abuse of Model Abstention
NeuronTune: Towards Self-Guided Spurious Bias Mitigation
KairosAD: A SAM-Based Model for Industrial Anomaly Detection on Embedded Devices
Anomaly Detection and Improvement of Clusters using Enhanced K-Means Algorithm
Optimal Weighted Convolution for Classification and Denosing
Black-box Adversarial Attacks on CNN-based SLAM Algorithms
PatchDEMUX: A Certifiably Robust Framework for Multi-label Classifiers Against Adversarial Patches
Are classical deep neural networks weakly adversarially robust?
Generalization Performance of Ensemble Clustering: From Theory to Algorithm
Z-Error Loss for Training Neural Networks
Random at First, Fast at Last: NTK-Guided Fourier Pre-Processing for Tabular DL
MemoryOut: Learning Principal Features via Multimodal Sparse Filtering Network for Semi-supervised Video Anomaly Detection
Investigating Mask-aware Prototype Learning for Tabular Anomaly Detection
Semiconductor SEM Image Defect Classification Using Supervised and Semi-Supervised Learning with Vision Transformers
INP-Former++: Advancing Universal Anomaly Detection via Intrinsic Normal Prototypes and Residual Learning
Prediction Inconsistency Helps Achieve Generalizable Detection of Adversarial Examples
Dynamic Epsilon Scheduling: A Multi-Factor Adaptive Perturbation Budget for Adversarial Training
Softlog-Softmax Layers and Divergences Contribute to a Computationally Dependable Ensemble Learning
Fool the Stoplight: Realistic Adversarial Patch Attacks on Traffic Light Detectors
Identifying and Understanding Cross-Class Features in Adversarial Training
Track Any Anomalous Object: A Granular Video Anomaly Detection Pipeline