3,396 Hits in 4.8 sec

Attention, Please! Adversarial Defense via Attention Rectification and Preservation [article]

Shangxi Wu, Jitao Sang, Kaiyuan Xu, Jiaming Zhang, Yanfeng Sun, Liping Jing, Jian Yu
2021 arXiv   pre-print
Accordingly, an attention-based adversarial defense framework is designed to simultaneously rectify the attention map for prediction and preserve the attention area between adversarial and original images  ...  In particular, we observed that: (1) images with incomplete attention regions are more vulnerable to adversarial attacks; and (2) successful adversarial attacks lead to deviated and scattered attention  ...  of original images focusing on the actual object of interest; (2) attention preservation component, to align the visual attention area between adversarial and original images to alleviate the feature  ... 
arXiv:1811.09831v3 fatcat:nbek5wg2y5hh5h4r3lcirg43py

Customizing First Person Image Through Desired Actions [article]

Shan Su, Jianbo Shi, Hyun Soo Park
2017 arXiv   pre-print
The image is created by combining present and future ActionTunnels in 3D where the missing pixels in adjoining area are computed by a generative adversarial network.  ...  This connects two distinctive first person images through similar walking paths.  ...  The following optimization trains these two models: R ∆ R Rectified scene 1 Rectified scene 2 1  2  1 ν 2 ν R ∆ R (a) ActionTunnel PROJ f c  (b) Connection g  (c) GAN min w G max w D L GAN + λL REC  ... 
arXiv:1704.00098v1 fatcat:3drjmibcond4jpn77mrd2yopza

MLAN: Multi-Level Adversarial Network for Domain Adaptive Semantic Segmentation [article]

Jiaxing Huang, Dayan Guan, Shijian Lu, Aoran Xiao
2021 arXiv   pre-print
MLAN has two novel designs, namely, region-level adversarial learning (RL-AL) and co-regularized adversarial learning (CR-AL).  ...  ., image-to-image translation) and output space (i.e., self-training) effectively.  ...  ) and Nanyang Technological University (NTU) that is funded by the Singapore Government through the Industry Alignment Fund -Industry Collaboration Projects Grant.  ... 
arXiv:2103.12991v1 fatcat:2sgqwqr5ovbpbit2pvfgka3tpu

PureGaze: Purifying Gaze Feature for Generalizable Gaze Estimation [article]

Yihua Cheng, Yiwei Bao, Feng Lu
2021 arXiv   pre-print
We design a plug-and-play self-adversarial framework for the gaze feature purification.  ...  and even facial expression may affect the learning in an unexpected way.  ...  Pseudo Code We also provide the pseudo code of the self-adversarial framework (PureGaze) in Fig. 12 .  ... 
arXiv:2103.13173v2 fatcat:i6sw7keyj5cw3cin2znqexjsxq

Cross-Domain Fault Diagnosis of Rotating Machinery using Discriminative Feature Attention Network

Gye-Bong Jang, Jin-Young Kim, Sung-Bae Cho
2021 IEEE Access  
ADVERSARIAL LEARNING FOR DOMAIN ADAPTATION The learning model consists of two steps. First, we trained the entire system using the source domain data and labels.  ...  T-SNE visualization of feature alignment between models with and without attention networks in datasets of the CWRU and a real machine.  ... 
doi:10.1109/access.2021.3096145 fatcat:vgugcw5qjbgqxnyl2h7ioht5sa

Undoing the Damage of Label Shift for Cross-domain Semantic Segmentation [article]

Yahao Liu, Jinhong Deng, Jiale Tao, Tong Chu, Lixin Duan, Wen Li
2022 arXiv   pre-print
In implementation, we adopt class-level feature alignment for conditional distribution alignment, as well as two simple yet effective methods to rectify the classifier bias from source to target by remolding  ...  In this paper, we give an in-depth analysis and show that the damage of label shift can be overcome by aligning the data conditional distribution and correcting the posterior probability.  ...  Acknowledgement: This work is supported by the Major Project for New Generation of AI under Grant No. 2018AAA0100400, the National Natural Science Foundation of China (Grant No. 62176047) and Beijing Natural  ... 
arXiv:2204.05546v1 fatcat:q3fudtuak5e23bei2ga5eb6m6a

SALIENCE: An Unsupervised User Adaptation Model for Multiple Wearable Sensors Based Human Activity Recognition [article]

Ling Chen, Yi Zhang, Sirou Zhu, Shenghuan Miao, Liangying Peng, Rong Hu, Mingqi Lv
2021 arXiv   pre-print
Unsupervised user adaptation aligns the feature distributions of the data from training users and the new user, so a well-trained wearable human activity recognition (WHAR) model can be well adapted to  ...  In addition, an attention mechanism is proposed to focus the activity classifier of SALIENCE on the sensors with strong feature discrimination and well distribution alignment.  ...  In steps 2 and 3, an adversarial training strategy is used to align the feature distributions of the data from training users and the new user.  ... 
arXiv:2108.10213v1 fatcat:yzhdiqdp25crzl3rufuaykjxxm

Low-Resource Text Classification using Domain-Adversarial Learning [article]

Daniel Grießhaber, Ngoc Thang Vu, Johannes Maucher
2018 arXiv   pre-print
This paper explores the use of domain-adversarial learning as a regularizer to avoid overfitting when training domain invariant features for deep, complex neural network in low-resource and zero-resource  ...  In the case of new languages, we show that monolingual word-vectors can be directly used for training without pre-alignment.  ...  Q, the model can be trained in a non-adversarial way.  ... 
arXiv:1807.05195v1 fatcat:mwww4sjohja6lhklajbi7fb6qi

MeGA-CDA: Memory Guided Attention for Category-Aware Unsupervised Domain Adaptive Object Detection [article]

Vibashan VS, Vikram Gupta, Poojan Oza, Vishwanath A. Sindagi, Vishal M. Patel
2021 arXiv   pre-print
Existing approaches for unsupervised domain adaptive object detection perform feature alignment via adversarial training.  ...  The proposed method consists of employing category-wise discriminators to ensure category-aware feature alignment for learning domain-invariant discriminative features.  ...  This adversarial training between feature extractor and discriminator helps to reduce the domain gap between source and target image features.  ... 
arXiv:2103.04224v2 fatcat:a5wlfjv6zfgqhevhrwy2qergxi

Cyberbullying detection across social media platforms via platform-aware adversarial encoding [article]

Peiling Yi, Arkaitz Zubiaga
2022 arXiv   pre-print
We propose XP-CB, a novel cross-platform framework based on Transformers and adversarial learning.  ...  have received less attention.  ...  These methods are motivated by Generative Adversarial Networks (Creswell et al. 2018) , which consist of two parts: (1) the Generator is trained to generate synthetic instances in a way that confuses  ... 
arXiv:2204.00334v1 fatcat:c2lk5ary3rgrnpyild72dv2idm

Boosting Noise Robustness of Acoustic Model via Deep Adversarial Training [article]

Bin Liu, Shuai Nie, Yaping Zhang, Dengfeng Ke, Shan Liang, Wenju Liu1
2018 arXiv   pre-print
Specifically, a jointly compositional scheme of generative adversarial net (GAN) and neural network-based acoustic model (AM) is used in the training phase.  ...  In this paper, we propose an adversarial training method to directly boost noise robustness of acoustic model.  ...  Setup In following experiments, we take the 80-dimensional filter-banks as the input features, and each dimension of features is normalized to have zero mean and unit variance over the training set.  ... 
arXiv:1805.01357v1 fatcat:o3wpy2tn55h3plhzdjk2fritxy

Exploiting Negative Learning for Implicit Pseudo Label Rectification in Source-Free Domain Adaptive Semantic Segmentation [article]

Xin Luo, Wei Chen, Yusong Tan, Chen Li, Yulin He, Xiaogang Jia
2021 arXiv   pre-print
clean during self-training, making critical tasks relying on semantic segmentation unreliable.  ...  to noisy pseudo labels in training, meanwhile positive learning achieves fast convergence.  ...  Our approach is modelindependent, which enables even black-box model's adaptation. methods utilize adversarial learning to match and align the distribution of features.  ... 
arXiv:2106.12123v1 fatcat:tgxeuol4vfbbvdbaorrej4nqsm

Consistency Regularization with High-dimensional Non-adversarial Source-guided Perturbation for Unsupervised Domain Adaptation in Segmentation [article]

Kaihong Wang, Chenhongyi Yang, Margrit Betke
2020 arXiv   pre-print
However, such methods highly rely on an image translator or feature extractor trained in an elaborated mechanism including adversarial training, which brings in extra complexity and instability in the  ...  Extensive experiments show that our BiSIDA achieves new state-of-the-art on two commonly-used synthetic-to-real domain adaptation benchmarks: GTA5-to-CityScapes and SYNTHIA-to-CityScapes.  ...  "FCN in the wild"" [12] was the first to perform a segmentation task under UDA settings and align both global and local features between domains through adversarial training.  ... 
arXiv:2009.08610v1 fatcat:ftslqpxe6bf45gpisrq6cbv5hu

Neural Alignment for Face De-pixelization [article]

Maayan Shuvi, Noa Fish, Kfir Aberman, Ariel Shamir, Daniel Cohen-Or
2020 arXiv   pre-print
Reconstruction and perceptual losses promote adherence to the ground-truth, and an adversarial loss assists in maintaining domain faithfulness.  ...  Our system exploits the simultaneous similarity and small disparity between close-by video frames depicting a human face, and employs a spatial transformation component that learns the alignment between  ...  Wavelet-SRNet EDVR [42] is a general video SR technique combining spatial and temporal attention with feature-level alignment.  ... 
arXiv:2009.13856v1 fatcat:reritceiczellbkhbwsvzms37m

Semantic Text-to-Face GAN -ST^2FG [article]

Manan Oza, Sukalpa Chanda, David Doermann
2022 arXiv   pre-print
This avoids the loss of features due to inadequate "attention", which may happen if text embedding and latent vector are simply concatenated.  ...  Our models are trained using an Affine Combination Module (ACM) mechanism to combine the text embedding from BERT and the GAN latent space using a self-attention matrix.  ...  In addition, we introduce two adversarial discriminators D X and D Y , where D X aims to distinguish between images {x} and translated images {F (y)}; in the same way, D Y aims to discriminate between  ... 
arXiv:2107.10756v2 fatcat:bidckdrvafhhhpgdcvpsaidcea
« Previous Showing results 1 — 15 out of 3,396 results