相关参考文献
注意:仅列出部分参考文献,下载原文获取全部文献信息。
Article
Computer Science, Artificial Intelligence
Jing Liu et al.
Summary: Driving style detection is crucial for various real-world applications, but existing methods have limitations in extracting deep spatial-temporal features from multi-modal sensing signals. In this paper, we propose DSDCLA, an attention-based hybrid CNN and LSTM framework, which effectively identifies driving style by utilizing short- and long-term spatial-temporal features. The proposed approach outperforms state-of-the-art methods on two public datasets, with F1-scores of 97.03% and 97.65%, demonstrating the effectiveness and importance of multi-level attention fusion.
APPLIED INTELLIGENCE
(2023)
Review
Computer Science, Artificial Intelligence
Sicheng Zhao et al.
Summary: This article reviews the latest single-source deep unsupervised domain adaptation (DA) methods for visual tasks and discusses new perspectives for future research. The article starts with the definitions of different DA strategies and descriptions of existing benchmark datasets, then summarizes and compares different categories of methods, and finally discusses future research directions.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Engineering, Electrical & Electronic
Yang Liu et al.
Summary: Video anomaly detection under weak supervision is achieved by introducing a collaborative normality learning framework that combines unsupervised learning with regression modeling. An unsupervised deep auto-encoder is used to learn the prototypical spatial-temporal patterns of normal videos, and a regression module is trained using both normal and abnormal videos. The proposed framework achieves comparable performance to state-of-the-art methods on benchmark datasets.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2022)
Article
Engineering, Electrical & Electronic
Yang Liu et al.
Summary: The key to video anomaly detection is understanding the appearance and motion differences between normal and abnormal events. In this paper, an appearance-motion united auto-encoder (AMAE) framework is proposed to jointly learn the prototypical spatial and temporal patterns of normal events. Experimental results demonstrate the effectiveness of the proposed framework.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2022)
Article
Computer Science, Artificial Intelligence
Guoliang Kang et al.
Summary: This paper proposes a contrastive adaptation network (CAN) for unsupervised domain adaptation. CAN explicitly models the intra-class and inter-class domain discrepancy to improve the generalization performance of the model. The paper also introduces an alternating update strategy and class-aware sampling to optimize CAN. Experimental results show that CAN outperforms state-of-the-art methods in both single-source and multi-source domain adaptation scenarios.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Proceedings Paper
Acoustics
Yang Liu et al.
Summary: This paper proposes a spatial-temporal attention mechanism for detecting abnormal events in surveillance videos. The features are made task-specific using the mutual cosine embedding loss, and the method achieves superior performance on standard benchmarks.
2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Zhaoyu Chen et al.
Summary: This article introduces a certifiable patch defense method that combines Vision Transformer and Derandomized Smoothing frameworks, using a progressive smoothed image modeling task to train Vision Transformer, and reconstructs the global self-attention structure into isolated band unit self-attention. The method achieves higher certified accuracy and close to normal accuracy on clean data under patch attacks.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022)
(2022)
Article
Engineering, Electrical & Electronic
Donglai Wei et al.
Summary: This paper proposes a novel fusion method and a Multimodal Supervise-Attention enhanced Fusion (MSAF) framework under weak supervision to address the lack of exploration of multimodal data and the ignore of implicit alignment of multimodal features in video anomaly detection. Extensive experiments on four challenging datasets demonstrate the effectiveness of the framework.
IEEE SIGNAL PROCESSING LETTERS
(2022)
Article
Engineering, Electrical & Electronic
Yuheng Wei et al.
Summary: A multiple-domain adaptation framework named CentriForce is proposed to extract domain-invariant speaker representations, which learns multiple domain-related speaker representation spaces. By reducing the Wasserstein distance between source and target domains in their domain-related representation space and using the target domain as an anchor point, CentriForce effectively mitigates the multiple-domain mismatch problem. Experimental results show that CentriForce outperforms other competing adaptation methods in challenging adaptation tasks.
IEEE SIGNAL PROCESSING LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Chuan-Xian Ren et al.
Summary: The proposed PTMDA method shows excellent performance in MDA tasks, effectively utilizing information correlation among multiple source domains to map the source domains into the target domain. By using pseudo target domains, PTMDA improves the performance on the real target domain.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Yongchun Zhu et al.
Summary: This study introduces a deep subdomain adaptation network (DSAN) that aligns relevant subdomain distributions across different domains based on the local maximum mean discrepancy (LMMD). DSAN is simple but effective, does not require adversarial training, and converges quickly. It can be easily integrated into feedforward network models to achieve efficient adaptation via backpropagation.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Engineering, Electrical & Electronic
Letian Qi et al.
Summary: This study introduces a CKCG algorithm that improves the performance of CKL by transforming it into a globally convex function, addressing the issue of poor performance of CKL in the presence of non-Gaussian noise. Simulations on nonlinear system identification confirm the superiorities of CKCG in terms of robustness and filtering accuracy.
IEEE SIGNAL PROCESSING LETTERS
(2021)
Article
Engineering, Electrical & Electronic
Xiaofeng Zhang et al.
Summary: The proposed deep adversarial data augmentation (DADA) technique defines data augmentation as a problem of training a class-conditional and supervised generative adversarial network (GAN) to address the challenge of training deep networks with insufficient data. A new discriminator loss is proposed to enforce both real and augmented samples to contribute to and be consistent in finding decision boundaries, leading to tailored training techniques. Extensive simulations and experiments on real-world datasets show that DADA outperforms traditional data augmentation and GAN-based options, proving its effectiveness.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Computer Science, Artificial Intelligence
Jindong Wang et al.
PATTERN RECOGNITION LETTERS
(2019)
Article
Engineering, Electrical & Electronic
Hui Yu et al.
IEEE SIGNAL PROCESSING LETTERS
(2019)