2022.2.1 第五次 arxiv论文速览
Debiased-CAM to mitigate systematic error with faithful visual explanations of machine learning
论文链接：https://arxiv.org/pdf/2201.12835.pdf
机构：National University of Singapore，CSIC-UPC
背景：Despite the fidelity of CAMs on clean images, real-world images are typically subjected to systematic error, such as image blurring, color-distortion or lighting changes, and these can affect what CAMs highlight.
ADVERSARIAL EXAMPLES FOR GOOD: ADVERSARIAL EXAMPLES GUIDED IMBALANCED LEARNING
论文链接：https://arxiv.org/pdf/2201.12356.pdf
机构：Zhejiang University
On the Robustness of Quality Measures for GANs
论文链接：https://arxiv.org/pdf/2201.13019.pdf
机构：KAUST, University of Oxford
This work evaluates the robustness of quality measures of generative models such as Inception Score (IS) and Fre ?chet Inception Distance (FID)
Adversarial Robustness in Deep Learning: Attacks on Fragile Neurons
论文链接：https://arxiv.org/pdf/2201.12347.pdf
机构：University of Reading, KTH Royal Institute of Technology, University of Catania, University of Cambridge
We identified fragile neurons (kernels) S and null neurons (kernels) S′ by dropping the kernels out systematically one-by-one and measuring the variance in model performance