Peng Xu

Peng Xu

Auto-generated profile, based on publicly available information

Peng Xu

Department of Computer Science & Technology, Tsinghua University, Beijing, China

Research Focus & Metrics

Academic Career Length

13Years (2014-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
6 August 2014 in NeuroReport
Verified
Journal Article
Closed access
YTYin Tian
Yin Tian
SLShanshan Liang
Shanshan Liang
Zhen YuanZhen Yuan
Zhen Yuan
SCSifan Chen
Sifan Chen
PXPeng Xu
Peng Xu
DYDezhong Yao
Dezhong Yao
A pilot study was carried out to determine individual differences in perceived loneliness using diffusion tensor imaging. To the best of our knowledge, this is the first preliminary diffusion tensor imaging evidence that the ventral attention network, generally activated by attentional reorienting, was also related to loneliness. Image reconstruction results indicated significantly decreased fractional anisotropy of white matter fibers and that associated nodes of the ventral attention network are highly correlated with increased loneliness ratings. By providing evidence on the structural level, our findings suggested that attention-reorienting capabilities play an important role in shaping an individual’s loneliness.

14 February 2022 in Virtual Reality & Intelligent Hardware
Verified
Journal Article
Open access
CLCunbo Li
Cunbo Li
NLNing Li
Ning Li
YQYuan Qiu
Yuan Qiu
YPYueheng Peng
Yueheng Peng
YWYifeng Wang
Yifeng Wang
PXPeng Xu
Peng Xu

28 October 2024
Verified
Conference Paper
Open access
XZXiaopei Zhu
Xiaopei Zhu
PXPeng Xu
Peng Xu
GZGuanning Zeng
Guanning Zeng
YDYinpeng Dong
Yinpeng Dong
XHXiaolin Hu
Xiaolin Hu
Research of adversarial attacks is important for AI security because it shows the vulnerability of deep learning models and helps to build more robust models. Adversarial attacks on images are most widely studied, which include noise-based attacks, image editing-based attacks, and latent space-based attacks. However, the adversarial examples crafted by these methods often lack sufficient semantic information, making it challenging for humans to understand the failure modes of deep learning models under natural conditions. To address this limitation, we propose a natural language induced adversarial image attack method. The core idea is to leverage a text-to-image model to generate adversarial images given input prompts, which are maliciously constructed to lead to misclassification for a target model. To adopt commercial text-to-image models for synthesizing more natural adversarial images, we propose an adaptive genetic algorithm (GA) for optimizing discrete adversarial prompts without requiring gradients and an adaptive word space reduction method for improving query efficiency. We further used CLIP to maintain the semantic consistency of the generated images. In our experiments, we found that some high-frequency semantic information such as ''foggy'', ''humid'', ''stretching'', etc. can easily cause classifier errors. This adversarial semantic information exists not only in generated images but also in photos captured in the real world. We also found that some adversarial semantic information can be transferred to unknown classification tasks. Furthermore, our attack method can transfer to different text-to-image models (e.g., Midjourney, DALL·E 3, etc.) and image classifiers. Our code is available at:https://github.com/zxp555/Natural-Language-Induced-Adversarial-Images.

1 December 2020 in PLOS ONE
Verified
Journal Article
Open access
PDF file
QLQuanshan Long
Quanshan Long
TLTing Luo
Ting Luo
SZsheng zhang
sheng zhang
YJYuanling Jiang
Yuanling Jiang
NHNa Hu
Na Hu
ACAntao Chen
Antao Chen
Information in working memory (WM) can guide visual attention towards matched features. While recent work has suggested that cognitive control can act upon WM guidance of visual attention, little is known about how the state of memorized items retaining in WM contribute to its influence over attention. Here, we disentangle the role of inhibition and maintenance on WM-guided attention with a novel delayed match-to-sample dual-task. The results showed that active inhibition facilitated searching by diminishing sensory processing and deterring attentional guidance, indexed by an attenuated P1 amplitude and unaffected N2pc amplitude, respectively. By contrast, active maintenance impaired searching by attentional guidance while sensory processing remained unimpaired, indexed by an enhanced N2pc amplitude and unchanged P1 amplitude, respectively. Furthermore, multivariate pattern analyses could sucessfully decode maintenance and inhibition, suggesting that two states differed in modulating visual attention. We propose that remembered contents may play an anchoring role for attentional guidance, and the state of those contents retaining in WM may directly influence the shifting of attention. The maintenance could guide attention by accessing input information, while the inhibition could deter the shifting of attention by suppressing sensory processing. These findings provide a possible reinterpretation of the influence of WM on attention.