The field of machine learning is moving towards improving the robustness and explainability of deep learning models. Researchers are exploring new methods to defend against adversarial attacks, such as universal adversarial perturbations, and to improve the transparency of model decisions. Notable papers in this area include 'Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Models', which proposes a novel data-free method for generating universal adversarial perturbations, and 'FakeScope: Large Multimodal Expert Model for Transparent AI-Generated Image Forensics', which introduces a multimodal model for detecting AI-generated images and providing interpretable forensic insights. Another significant direction is the development of methods for explaining and understanding model decisions, such as 'Enhancing Negation Awareness in Universal Text Embeddings: A Data-efficient and Computational-efficient Approach', which proposes a method for improving the negation awareness of universal text embedding models.
Advances in Adversarial Robustness and Explainability
Sources
Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Models
A Dataset for Semantic Segmentation in the Presence of Unknowns
Negation: A Pink Elephant in the Large Language Models' Room?
Instance-Level Data-Use Auditing of Visual ML Models
Tropical Bisectors and Carlini-Wagner Attacks
Synthetic Art Generation and DeepFake Detection A Study on Jamini Roy Inspired Dataset
A Survey on Unlearnable Data
THEMIS: Towards Practical Intellectual Property Protection for Post-Deployment On-Device Deep Learning Models
Evaluation of (Un-)Supervised Machine Learning Methods for GNSS Interference Classification with Real-World Data Discrepancies
FakeScope: Large Multimodal Expert Model for Transparent AI-Generated Image Forensics
Revisiting the Relationship between Adversarial and Clean Training: Why Clean Training Can Make Adversarial Training Better
Universal Zero-shot Embedding Inversion
Unleashing the Power of Pre-trained Encoders for Universal Adversarial Attack Detection
Exploring the Collaborative Advantage of Low-level Information on Generalizable AI-Generated Image Detection
Enhancing Negation Awareness in Universal Text Embeddings: A Data-efficient and Computational-efficient Approach
TenAd: A Tensor-based Low-rank Black Box Adversarial Attack for Video Classification
Beyond Nearest Neighbor Interpolation in Data Augmentation
Leveraging Generalizability of Image-to-Image Translation for Enhanced Adversarial Defense
All Patches Matter, More Patches Better: Enhance AI-Generated Image Detection via Panoptic Patch Learning
Benchmarking the Spatial Robustness of DNNs via Natural and Adversarial Localized Corruptions
Hessian-aware Training for Enhancing DNNs Resilience to Parameter Corruptions
Fault injection analysis of Real NVP normalising flow model for satellite anomaly detection
Evaluating and Enhancing Segmentation Model Robustness with Metamorphic Testing