Results 121 to 130 of about 5,380,268 (331)

Securely Fine-tuning Pre-trained Encoders Against Adversarial Examples

With the evolution of self-supervised learning, the pre-training paradigm has emerged as a predominant solution within the deep learning landscape.
Wan, W, Yao, D, Li, M, Jin, H, Liu, W, Zhang, LY, Zhou, Z, Xue, L, Zhang, Y, Hu, S   +9 more
core   +1 more source

Atomic Defects in Layered Transition Metal Dichalcogenides for Sustainable Energy Storage and the Intelligent Trends in Data Analytics

Advanced Science, EarlyView.
This review comprehensively summarizes the atomic defects in TMDs for their applications in sustainable energy storage devices, along with the latest progress in ML methodologies for high‐throughput TEM data analysis, offering insights on how ML‐empowered microscopy facilitates bridging structure–property correlation and inspires knowledge for precise ...
Zheng Luo, Ying Yang, Zizhen Fu, Fengqi Liu, Susu Fang, Kele Xu, Shanshan Wang   +6 more
wiley   +1 more source

A Gradual Adversarial Training Method for Semantic Segmentation

Remote Sensing
Deep neural networks (DNNs) have achieved great success in various computer vision tasks. However, they are susceptible to artificially designed adversarial perturbations, which limit their deployment in security-critical applications.
Yinkai Zan, Pingping Lu, Tingyu Meng
doaj   +1 more source

Generating Natural Adversarial Examples

CoRR, 2017
Due to their complex nature, it is hard to characterize the ways in which machine learning models can misbehave or be exploited when deployed. Recent work on adversarial examples, i.e. inputs with minor perturbations that result in substantially different model predictions, is helpful in evaluating the robustness of these models by exposing the ...
Zhengli Zhao, Dheeru Dua, Sameer Singh 0001   +2 more
openaire   +3 more sources

Trace-Norm Adversarial Examples

CoRR, 2020
White box adversarial perturbations are sought via iterative optimization algorithms most often minimizing an adversarial loss on a $l_p$ neighborhood of the original image, the so-called distortion set. Constraining the adversarial search with different norms results in disparately structured adversarial examples.
Ehsan Kazemi 0003, Thomas Kerdreux, Liqiang Wang 0001   +2 more
openaire   +2 more sources

Sustainable Materials Design With Multi‐Modal Artificial Intelligence

Advanced Science, EarlyView.
Critical mineral scarcity, high embodied carbon, and persistent pollution from materials processing intensify the need for sustainable materials design. This review frames the problem as multi‐objective optimization under heterogeneous, high‐dimensional evidence and highlights multi‐modal AI as an enabling pathway.
Tianyi Xu, Tianshuo Wei, Yan Ge, Bo Peng, Yue Li, Maolin Wang, Peng Wen, Chao Yang, Ye Wei   +8 more
wiley   +1 more source

CommanderUAP: a practical and transferable universal adversarial attacks on speech recognition models

Cybersecurity
Most of the adversarial attacks against speech recognition systems focus on specific adversarial perturbations, which are generated by adversaries for each normal example to achieve the attack.
Zheng Sun, Jinxiao Zhao, Feng Guo, Yuxuan Chen, Lei Ju   +4 more
doaj   +1 more source

SpaMode: A Broadly Applicable Framework for Deciphering Spatial Multi‐Omics Using Multimodal Mixture of Disentangled Experts

Advanced Science, EarlyView.
SpaMode introduces a versatile framework for spatial multi‐omics integration across vertical, horizontal, and mosaic scenarios. By disentangling modality‐invariant and variant features through a mixture‐of‐experts mechanism, it adaptively reconfigures spatially heterogeneous signals.
Xubin Zheng, Xinlei Huang, Xiang Zhou, Dian Meng, Ziyue Qiao, Zhiyuan Yuan, Lixin Cheng   +6 more
wiley   +1 more source

Enhancing Machine Learning Security: The Significance of Realistic Adversarial Examples [PDF]

Adversarial attacks pose a significant security threat in Machine Learning (ML), employing subtle, invisible perturbations on original examples to craft instances that deceive model decisions.
DYRMISHI, Salijona
core  

Diffusion‐Based Generative Model With Scaffold‐Hopping Strategy Yields Highly Potent Bioactive Molecules

Advanced Science, EarlyView.
SMarT‐Diff introduces a multi‐objective generative paradigm that integrates scaffold hopping with structure‐aware scoring to enable controlled exploration beyond the training distribution. The framework consistently balances drug‐likeness, synthesizes accessibility and bioactivity, yielding chemically diverse candidates with enhanced properties.
Yuwei Yang, Xiaoqing Gong, Shukai Gu, Jing Li, Bo Liu, Yanan Tian, Qianqian Zhang, Xiaojun Yao, Huanxiang Liu   +8 more
wiley   +1 more source