Related references
Note: Only part of the references are listed.
Article
Computer Science, Artificial Intelligence
Shaik John Saida et al.
Summary: This paper proposes a deep convolutional neural network model for automated semantic segmentation and identification of ocean eddies at the pixel level. The proposed architecture utilizes an attention mechanism to capture contextual information and a self-attention process to define semantic relationships between pixels. Experimental results show that the proposed model outperforms existing techniques in terms of mean pixel accuracy and intersection of union score.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Yeongmin Yoo et al.
Summary: In the field of virtual product development, a novel engineering design process called DADDA algorithm is proposed to perform virtual data-based generative optimization without requiring many simulations. The DADDA algorithm significantly reduces the time required to derive optimal design variables and can achieve better performance for new systems compared to existing methods. The algorithm was applied to electric vehicle design, showing promising results in reducing time and improving performance.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Pinhao Song et al.
Summary: This work proposes a two-stage underwater object detector called boosting R-CNN, which addresses the challenges of complicated underwater environments. It includes uncertainty modeling and hard example mining, and consists of three key components: RetinaRPN, probabilistic inference pipeline, and boosting reweighting. Experimental results demonstrate the effectiveness and robustness of this method.
Article
Computer Science, Artificial Intelligence
Jun Chen et al.
Summary: In this work, a Multi-Gradient Hierarchical Domain Adaptation Network (MGHDAN) is proposed to improve the transfer diagnosis performance of bearing faults by simultaneously learning transferable domain-invariant and class-discriminative knowledge.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Xing Wei et al.
Summary: Unsupervised domain adaptation aims to transfer knowledge from labeled source domain to unlabeled target domain for improving target domain's classification performance. However, existing methods only focus on domain alignment in contrastive learning, neglecting the classification task, leading to suboptimal solutions. In this paper, we propose Task-oriented Contrastive Learning for Unsupervised Domain Adaptation (TOCL) to address the inconsistency problem. Our method performs feature weighting and extracts task-related features from source domain images after data augmentation for contrastive learning. Additionally, a delimitation discriminator is introduced to detect and minimize the output difference between two types of data-augmented samples in the target domain. Extensive experiments on public datasets demonstrate the effectiveness and adaptability of TOCL in improving classification accuracy.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Jichao Zhuang et al.
Summary: This article proposes a two-stage transfer alignment (TSTA) method for fault diagnosis of bearings under transfer tasks. It extracts common and domain-specific fault features using a temporal transformer and a semantic consistency module, while eliminating feature confusion in aligning domain-specific decision boundaries of different classifiers using an entropy loss.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Engineering, Biomedical
Roghayyeh Arvanaghi et al.
Summary: The study proposed a method using ABP signal instead of ECG signal to automatically classify heartbeat types, achieving good classification results. The ABP signal provides valuable information about heart performance and has potential clinical applications.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2022)
Article
Computer Science, Artificial Intelligence
Xinhui Liu et al.
Summary: Unsupervised domain adaptation (UDA) algorithms aim to transfer knowledge from labeled source domains to unlabeled target domains. This paper proposes a Mutual Constraint Network for multi-level UDA (M2N) to mitigate the impact of noisy pseudo labels and learn better feature representations.
Article
Computer Science, Artificial Intelligence
Jingjing Li et al.
Summary: This paper discusses the generalization problem of conventional machine learning algorithms when dealing with data from different distributions and proposes an unsupervised domain adaptation method. The authors consider the scenario where either the source domain data or the target domain data is unknown, and investigate how to handle the divergence-agnostic adaptive learning problem from the perspective of adversarial attacks. Experimental results show that the proposed method outperforms previous ones.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Minwoo Kim et al.
Summary: The paper proposes a deep learning-based algorithm for time-to-collision (TTC) estimation, which reduces the reality-gap by converting simulated images into real world-like synthetic images. The aleatoric loss function and Monte Carlo dropout method are employed to consider the uncertainties from using the synthetic dataset. The performance of the proposed framework is validated through real flight experiments, achieving a decrease in the average TTC estimation error compared to the baseline approach.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Daqiang Dong et al.
Summary: Computed tomography (CT) is the primary diagnostic tool for brain diseases. This study proposes an unsupervised domain adaptation segmentation model for hemorrhagic lesions, which can be trained without using ground-truth labels. The model demonstrates better performance and has potential application in clinical settings.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Fan Wang et al.
Summary: Source free domain adaptation (SFDA) is a critical method to transfer knowledge from a well-trained source model to an unlabeled target domain. Most existing methods focus on learning target-specific domain-invariant representations, which cannot fully address the distribution shift problem. In contrast, exploring the domain-invariant parameters of the source model is a more effective approach.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Lin Chen et al.
Summary: This work proposes a discriminator-free adversarial learning network (DALN) that leverages the predicted discriminative information for feature alignment. The category classifier is reused as a discriminator to achieve domain alignment and category distinguishment. The introduction of Nuclear-norm Wasserstein discrepancy (NWD) provides guidance for discrimination without extra weight clipping or gradient penalty strategy.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Ning Ding et al.
Summary: This work introduces a new framework for source-free domain adaptation, addressing the domain adaptation task through source distribution estimation. By generating robust pseudo-labels for target data with spherical k-means clustering, estimating the class-conditioned feature distribution of the source domain, and sampling surrogate features to align the two domains, the proposed method achieves state-of-the-art performance on multiple DA benchmarks.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Yuanyuan Xu et al.
Summary: Multi-Source Domain Adaptation (MSDA) aims to transfer knowledge from multiple labeled source domains to an unlabeled target domain. The challenges lie in mitigating domain gaps and combining diverse source domain information. This work proposes a mutual learning based alignment network (MLAN) that combines the advantages of joint and separate domain alignments. The MLAN integrates different domain alignment mechanisms, allowing for better performance by mining rich knowledge from multiple domains.
2022 IEEE WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV 2022)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Haifeng Xia et al.
Summary: The study proposes an Adaptive Adversarial Network (A2Net) with three components to address the challenge of source data unavailability in the Source-free Domain Adaptation scenario. The components include Adaptive Adversarial Inference, Contrastive Category-wise Matching module, and Self-Supervised Rotation, which collectively help improve the adaptability of the model without needing access to the source domain data.
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Shiqi Yang et al.
Summary: G-SFDA is a new domain adaptation paradigm that aims to perform well on both target and source domains, with only access to current unlabeled target data during adaptation. The proposed LSC and SDA methods successfully adapt the model to the target domain while keeping source performance.
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021)
(2021)
Article
Computer Science, Information Systems
Abdollah Amirkhani et al.
Summary: Recent studies have utilized knowledge distillation (KD) technique to address time-consuming annotation task in semantic segmentation. This study aims to train a robust, lightweight student under the supervision of several expert teachers, leading to better performance compared to a single-teacher learning framework. The proposed multi-teacher KD paradigm allows the student to capture a variety of knowledge illustrations from different sources, resulting in significant improvement in both clean and corrupted data in the semantic segmentation task.