Source Domain Research Articles

Deep domain-adversarial anomaly detection with robust one-class transfer learning

Establishing an anomaly detection model as quickly as possible with a small number of measurement samples for use with new equipment or working conditions has been a research focus. In recent years, despite the effectiveness of transfer learning techniques in terms of coping with such problems, it is still challenging to map the detection rules of an existing device to a target device to enhance its single-class anomaly detection capabilities. In addition, the robustness requirement of the detection rule transfer process makes the problem further complicated, particularly in cases where limited target device data still have unknown outliers. To achieve good anomaly detection performance in such a complex scenario, we propose a robust deep one-class support vector description-based anomaly detection algorithm built on adversarial transfer learning. First, we designed a new hyperspherical adversarial training mechanism that integrates robust adaptive constraints into an improved domain-adversarial neural network. Additionally, to eliminate outlier interference to the greatest extent possible during the domain-adversarial process, we aimed to compute the probability of a sample being normal based on the utilized anomaly detection module and embedded this probability in a domain discriminator. Finally, an end-to-end optimization strategy was derived to find the optimal network parameters that could separate the anomalous classes as much as possible while promoting the consistency of the normal class data in the source and target domains. To validate the performance of our model, we experimented with three scenarios, i.e., digit image detection (with the MNIST and USPS datasets), object recognition (with the Office-Home dataset), and rolling bearing anomaly detection. The results showed that the proposed algorithm outperformed the state-of-the-art methods in terms of detection accuracy and robustness.

Spectral Decomposition and Transformation for Cross-domain Few-shot Learning

Cross-domain few-shot Learning (CDFSL) is proposed to first pre-train deep models on a source domain dataset where sufficient data is available, and then generalize models to target domains to learn from only limited data. However, the gap between the source and target domains greatly hampers the generalization and target-domain few-shot finetuning. To address this problem, we analyze the domain gap from the aspect of frequency-domain analysis. We find the domain gap could be reflected by the compositions of source-domain spectra, and the lack of compositions in the source datasets limits the generalization. Therefore, we aim to expand the coverage of spectra composition in the source datasets to help the source domain cover a larger range of possible target-domain information, to mitigate the domain gap. To achieve this goal, we propose the Spectral Decomposition and Transformation (SDT) method, which first randomly decomposes the spectrogram of the source datasets into orthogonal bases, and then randomly samples different coordinates in the space formed by these bases. We integrate the above process into a data augmentation module, and further design a two-stream network to handle augmented images and original images respectively. Experimental results show that our method achieves state-of-the-art performance in the CDFSL benchmark dataset.

Domain generalized person reidentification based on skewness regularity of higher-order statistics

The goal of domain-generalized person reidentification (DG-ReID) is to train a model in the source domain and apply it directly to unknown target domains for specific pedestrian retrieval. Existing methods rely primarily on low-order statistics (such as the mean, standard deviation, or variance), thereby ensuring the stability of the source domain data distribution for model training. However, such methods underperform when the data follow a non-Gaussian distribution, thereby reducing the generalization ability of the model on unseen target domains. To address this issue, this study proposes an instance normalization-based skewness regularity (INSR) framework that uses high-order statistics (skewness and high-order moments) to measure the skewness and regularity of the data distribution. Such measures allow further learning of the morphological features (skewness degree, trait of data near the mean, etc.) of complex data distributions for DG-ReID. Specifically, the proposed framework first extracts the skewness and third-order moments from the source domains, which provide more features (high-order moments, variance, etc.) to characterize the data distribution. Subsequently, a batch normalization-like operation was implemented to project the data into a new feature space with zero mean and unit variance, enhancing model adaption and accuracy. Extensive experiments were conducted on small-scale (VIPeR, PRID, GRID, and i-LIDS) and large-scale (Market-1501, DukeMTMC-reID, CUHK03, MSMT17) public datasets using two different protocols, demonstrating that the proposed INSR framework significantly outperforms other state-of-the-art counterparts for DG-ReID.

Transfer learning with cross-domain adaptation of vibro-acoustic signals for electric motor mechanical fault detection and diagnosis.

Contact-based vibration measurement techniques involving accelerometers are typically utilized for machine health monitoring and manufacturing end-of-line quality control. Acoustic signals, encompassing sound pressure and particle velocity, carry crucial information about the operational status of underlying equipment and can be utilized for detecting various machine faults. Utilizing one-dimensional Convolution Neural Networks (1D-CNN) based on deep learning for processing raw acoustic time signals has proven to be a valuable approach in detecting machinery faults, offering an effective alternative to using traditional accelerometer signals. However, the success of deep learning methods is heavily dependent on the quantity and availability of data for robust network training. When transitioning to new non-contact type sensor technology, manufacturers may face challenges due to insufficient data for training deep learning models. In such cases, leveraging existing historical accelerometer-based data becomes crucial. Transfer learning emerges as a promising solution, allowing deep learning models trained on a source domain (SD) to be applied to a separate but related target domain (TD). This paper hence introduces a domain adaptation methodology by implementing transfer learning, where the SD 1D-CNN network is trained on accelerometer signals, then the trained model weights are transferred to the TD 1D-CNN network that can process acoustic signals.

Metaphor and gender: are words associated with source domains perceived in a gendered way?

Abstract Conceptual metaphors are one of many linguistic devices that can potentially encode and reinforce gender stereotypes. However, little is known about how metaphors encode gender stereotypes, and in previous literature the concept of “gendered metaphor” has been mostly assumed rather than attested. We take the first step to tackle this issue by examining the gender typicality of specific metaphorical source domains. In the present paper, we conducted three rating experiments (N total = 1,060 English-speaking participants) to determine the genderedness of 50 keywords associated with five frequently used source domains (building, competition, journey, plant, and war). We found that keywords associated with three source domains (building, competition, and war) were viewed as more masculine, while keywords associated with the source domains of journey and plant were viewed as more feminine. These data offer empirical verification for gendered perceptions of keywords associated with some frequently used source domains. The result also provides the first evidence that metaphors could encode gender stereotypes by selection of source domains.

Recognition of aggressive driving behavior under abnormal weather based on Convolutional Neural Network and transfer learning

Objectives Aggressive driving behavior can lead to potential traffic collision risks, and abnormal weather conditions can exacerbate this behavior. This study aims to develop recognition models for aggressive driving under various climate conditions, addressing the challenge of collecting sufficient data in abnormal weather. Methods Driving data was collected in a virtual environment using a driving simulator under both normal and abnormal weather conditions. A model was trained on data from normal weather (source domain) and then transferred to foggy and rainy weather conditions (target domains) for retraining and fine-tuning. The K-means algorithm clustered driving behavior instances into three styles: aggressive, normal, and cautious. These clusters were used as labels for each instance in training a CNN model. The pre-trained CNN model was then transferred and fine-tuned for abnormal weather conditions. Results The transferred models showed improved recognition performance, achieving an accuracy score of 0.81 in both foggy and rainy weather conditions. This surpassed the non-transferred models’ accuracy scores of 0.72 and 0.69, respectively. Conclusions The study demonstrates the significant application value of transfer learning in recognizing aggressive driving behaviors with limited data. It also highlights the feasibility of using this approach to address the challenges of driving behavior recognition under abnormal weather conditions.

A Comment Aspect-Level User Preference Transfer Model for Cross-Domain Recommendations

Traditional cross-domain recommendation models make it difficult to deeply mine users' aspect-level preferences from comment information due to existing problems such as polysemy of comment text, sparse comment data, and user cold start. A Cross-Domain Recommender (CDR) model that integrates comment knowledge enhancement and aspect-level user preference transfer (C-KE-AUT) was proposed to address the above issues. Firstly, an aspect-level user preference extraction model was constructed by combining the RoBERTa word embedding model, high-level feature representation based on Transformer, and aspect-level attention-learning methods. Then, a user aspect-level preference cross-domain transfer model was constructed based on a two-stage generative adversarial network that can transfer the aspect-level interest preferences of users in the source domain to the target domain with sparse data. The experimental results on the Amazon 2018 comment dataset indicated that the recommendation performance of the proposed C-KE-AUT model was significantly superior to other advanced comparative models.

A human-in-the-loop method for pulmonary nodule detection in CT scans

Automated pulmonary nodule detection using computed tomography scans is vital in the early diagnosis of lung cancer. Although extensive well-performed methods have been proposed for this task, they suffer from the domain shift issue between training and test images. Unsupervised domain adaptation (UDA) methods provide a promising means to mitigate the domain variance; however, their performance is still limited since no target domain supervision is introduced. To make the pulmonary nodule detection algorithm more applicable in clinical practice and further boost the performance across domains, we propose a human-in-the-loop method in a semi-supervised fashion to enhance the model generalization ability when transferred from source domain to target domain. Specifically, we first train a detector model on source domain, and then the pre-trained detector is utilized with our proposed uncertainty-guided sample selection scheme (USSS) to find a few target domain samples worth annotating most and obtain their human annotations. Finally, the annotated and the rest unlabeled target domain samples are used together to refine the pre-trained model via our proposed zoom-in and zoom-out constraint (ZZC) strategy. We evaluate our method on the Nodule Analysis 2016 (LUNA16) and TianChi datasets. Experimental results show that our method surpasses recent competitive methods on source domain and also achieves surprising performance on target domain.

Kernelized Bures metric: A framework for effective domain adaptation in sensor data analysis

Unsupervised Domain Adaptation (UDA) plays a crucial role in enabling the transfer of models trained on labeled source domains to unlabeled target domains. However, this transferability encounters significant challenges when dealing with intricate time series models due to the complex temporal dynamics that differ between domains. These divergent dynamics result in disparities in time and frequency representations, leading to misalignments and gaps. Furthermore, the absence of localized information capture in previous approaches undermines the performance potential of the models. We introduce the KBSDA (Kernel Bures Sub-Domain Adaptation) approach to tackle the issues above. Our methodology employs the fast Fourier transform to extract frequency features while encompassing the time features. Our approach’s key focus is capturing intricate attributes containing local information, a task accomplished by applying the Local Maximum Mean Discrepancy (LMMD) metric. Furthermore, we address the distribution gaps between the source and target domains, elevating the alignment process by incorporating the kernel Bures metric. This metric, which encompasses higher-order moments and effectively manages intricate non-linear relationships, significantly manages complex distribution shifts. To validate the effectiveness of our approach, we conducted comprehensive experiments on well-established time series domain adaptation datasets such as HAR, HHAR, WISDM, and SSC. Our approach shows average accuracy of 95.10%, 78.04%, 78.69%, and 74.25% for HAR, HHAR, WISDM, and SSC, respectively.

Infproto-Powered Adaptive Classifier and Agnostic Feature Learning for Single Domain Generalization in Medical Images

AbstractDesigning a single domain generalization (DG) framework that generalizes from one source domain to arbitrary unseen domains is practical yet challenging in medical image segmentation, mainly due to the domain shift and limited source domain information. To tackle these issues, we reason that domain-adaptive classifier learning and domain-agnostic feature extraction are key components in single DG, and further propose an adaptive infinite prototypes (InfProto) scheme to facilitate the learning of the two components. InfProto harnesses high-order statistics and infinitely samples class-conditional instance-specific prototypes to form the classifier for discriminability enhancement. We then introduce probabilistic modeling and provide a theoretic upper bound to implicitly perform the infinite prototype sampling in the optimization of InfProto. Incorporating InfProto, we design a hierarchical domain-adaptive classifier to elasticize the model for varying domains. This classifier infinitely samples prototypes from the instance and mini-batch data distributions, forming the instance-level and mini-batch-level domain-adaptive classifiers, thereby generalizing to unseen domains. To extract domain-agnostic features, we assume each instance in the source domain is a micro source domain and then devise three complementary strategies, i.e., instance-level infinite prototype exchange, instance-batch infinite prototype interaction, and consistency regularization, to constrain outputs of the hierarchical domain-adaptive classifier. These three complementary strategies minimize distribution shifts among micro source domains, enabling the model to get rid of domain-specific characterizations and, in turn, concentrating on semantically discriminative features. Extensive comparison experiments demonstrate the superiority of our approach compared with state-of-the-art counterparts, and comprehensive ablation studies verify the effect of each proposed component. Notably, our method exhibits average improvements of 15.568% and 17.429% in dice on polyp and surgical instrument segmentation benchmarks.

CentroIDA: Cross-domain class discrepancy minimization based on accumulative class-centroids for Imbalanced Domain Adaptation

Unsupervised Domain Adaptation (UDA) is a critical task in transfer learning, which seeks to apply the knowledge acquired in one domain (the source domain) to a different but related domain (the target domain). In this context, the two domains exhibit Out-Of-Distribution(OOD), and the label information of the target domain is unavailable. Although numerous UDA methods have made significant strides in mitigating the discrepancies in feature distributions across domains, they often neglect a more prevalent real-world scenario. This scenario, referred to as Imbalanced Domain Adaptation (IDA) task, is characterized by the coexistence of differences in both feature and label distributions. Our empirical analysis suggests that overlooking the disparities in label distributions can severely compromise the performance of the model, thereby limiting the generalizability of domain adaptation models. Consequently, our objective is to introduce a model capable of concurrently addressing differences in both feature and label distributions, thereby fostering a broader application of domain adaptation methodologies. In this paper, we propose a cross-domain class discrepancy minimization method based on accumulative class-centroids for IDA, termed as centroIDA. The strategies of accumulative class-centroids alignment and class-wise feature alignment pay heed to the relationship between cross-domain intra-class features and inter-class features, thereby facilitating conditional feature alignment to overcome both feature and label distribution differences simultaneously. Our experimental evaluation on OfficeHome and DomainNet datasets reveals that, in comparison to TIToK (a novel method also designed to tackle the IDA problem), our method enhances the average accuracy by 5.6% and 2.7%, respectively. Furthermore, as the label shift intensifies, our method consistently outperforms other methods, demonstrating a robust resistance to label shift.

PheoSeg: A 3D transfer learning framework for accurate abdominal CT pheochromocytoma segmentation and surgical grade prediction

Pheochromocytoma is a rare urological disease. Accurate segmentation of pheochromocytoma and its surrounding abdominal organs is crucial for surgical planning and preoperative diagnosis. However, current clinical datasets typically include tumor labels but lack comprehensive multi-organ abdominal labels, and detailed descriptions of the spatial relationships between the tumor and adjacent organs are insufficient. To address the challenge of automatic multi-organ segmentation in the abdomen without comprehensive labels around pheochromocytoma, we propose a novel method for automatic multi-source migration segmentation and preoperative diagnostic grading of pheochromocytoma and its surrounding organs. This method utilizes nnUNet to extract data fingerprint information from the target domain (Clinical Pheochromocytoma dataset of Peking Union Medical College Hospital, PPGL) as prior knowledge, which is then transferred to two source domains (FLARE2022 and AMOS2022). We trained 10 models and incorporated them into a "Model Bank". Additionally, we employed an improved CC-FV method to screen the models in the "Model Bank," identifying the model with the highest mobility (class consistency and feature diversity). This selected model served as a pre-trained model for abdominal multi-organ segmentation and was subsequently adapted for tumor segmentation. To address the issue of limited clinical data, we introduced a data enhancement technique for 3D medical images. The enhanced dataset was used to fine-tune the pre-trained model, resulting in a model capable of segmenting multiple organs and tumors in the abdomen. From the segmentation results, we extracted the spatial relationships between the tumor and surrounding abdominal organs and constructed a radiomic model to predict and analyze the surgical grading. Experimental results demonstrate that our proposed method achieves a Dice coefficient that is, on average, 10.18 % higher on FLARE2022 and AMOS2022 datasets compared to segmentation frameworks such as Swin-Unet, UNETR, CoTr, nnformer, and nnUNetv2. For the PPGL dataset, the Dice coefficient for pheochromocytoma and surrounding organs was 87.27 %. In summary, our method effectively realizes efficient and automatic segmentation of pheochromocytoma and surrounding organs, as well as preoperative diagnostic classification.

Unsupervised deep domain adaptation algorithm for video based human activity recognition via recurrent neural networks

The integration of cyber-physical systems and artificial intelligent human activity recognition (HAR) applications enables intelligent interactions within a physical environment. In real-world HAR applications, the domain shift between training (source) and testing (target) images captured in different scenarios leads to low classification accuracy. Existing unsupervised domain adaptation (UDA) methods often require some labeled target samples for model adaptation, which limits their practicality. This study proposes a novel unsupervised deep domain adaptation algorithm (UDDAA) for HAR using recurrent neural networks. UDDAA introduces a maximum mean class discrepancy (MMCD) metric, accounting for both inter-class and intra-class differences within each domain. MMCD extends the maximum mean discrepancy to measure the class-level distribution discrepancy across source and target domains, aligning these distributions to enhance domain adaptation performance. Without relying on labeled target data, UDDAA predicts pseudo-labels for the unlabeled target dataset, combining these with labeled source data to train the model on domain-invariant representations. This approach makes UDDAA highly practical for scenarios where labeled target data is difficult or expensive to obtain, enabling human-computer interaction (HCI) systems to function effectively across varied environments and user behaviors. Extensive experiments on benchmark datasets demonstrate UDDAA's superior classification accuracy over existing baselines. Notably, UDDAA achieved 92% and 99% accuracy for University of Central Florida database (UCF) to Human Motion Database (HMDB) and HMDB to UCF transfers, respectively. Additionally, on personal recorded videos with complex backgrounds, it achieved high classification accuracies of 95% for basketball and 90% for football activities, underscoring its generalization ability, robustness, and effectiveness.

Unsupervised domain adaptation with hard-sample dividing and processing strategy

In Unsupervised Domain Adaptation(UDA), the hard samples cause serious negative transfer to model performance. However, the existing UDA methods don’t fully pay attention to the fine-grained processing of hard samples in target and source domains. To address these problems, the UDA framework with Hard-sample Dividing and Processing Strategy (HDPS) is proposed in this paper. In HDPS, we define sample division criteria in target domain and source domain respectively and divide the samples into easy samples, hard samples with low confidence and hard samples with high confidence. We further design systematic processing strategy of the three types of samples. Specifically, we firstly define geometric metrics based on the class prototype to divide the hard samples and eliminate them in source domain. We further design two weight allocation strategies based on sample classification entropy in target domain. Finally, we build a teacher-student guidance mechanism to learn the discriminative features of hard samples. The HDPS framework can be easily loaded to other methods as a plug-in, can promote the diversity feature learning of samples, and significantly improve the discriminant performance of the model. Extensive experiments on six benchmark datasets verify that HDPS is effective and superior to most existing UDA methods.

An uncertainty perception metric network for machinery fault diagnosis under limited noisy source domain and scarce noisy unknown domain

Deep learning has made notable advances in intelligent fault diagnosis. However, industrial application of deep learning models faces challenges due to noise interference and scarce labeled samples. Targeting the above problems, this paper proposes a metric network-based diagnostic method. For the problem of noise, a multi-scale cross feature extraction module (MSCM) is constructed to mine key classification information under noise interference to improve fault identifiability. Different from the current approach to metric learning, this paper models the uncertainty of the similarity between ‘query sample-class prototype’, and develops corresponding loss function for more effective perception, thereby better improving the fault recognition ability of the model under limited noisy source domain and scarce noisy unknown domain. Meanwhile, to visualize the decision-making process of the model under uncertainty and improve interpretability, this paper develops a novel colony-based class activation mapping (Colony-CAM) tool, which is more reliable and focused. The proposed method is compared with five baselines across three datasets. It achieved leading diagnostic accuracies of 98.55% and 94.33% with 70 and 40 noisy training samples, respectively.

TSFAN: tensorized spatial-frequency attention network with domain adaptation for cross-session EEG-based biometric recognition

Objective. Electroencephalogram (EEG) signals are promising biometrics owning to their invisibility, adapting to the application scenarios with high-security requirements. However, It is challenging to explore EEG identity features without the interference of device and state differences of the subject across sessions. Existing methods treat training sessions as a single domain, affected by the different data distribution among sessions. Although most multi-source unsupervised domain adaptation (MUDA) methods bridge the domain gap between multiple source and target domains individually, relationships among the domain-invariant features of each distribution alignment are neglected. Approach. In this paper, we propose a MUDA method, Tensorized Spatial-Frequency Attention Network (TSFAN), to assist the performance of the target domain for EEG-based biometric recognition. Specifically, significant relationships of domain-invariant features are modeled via a tensorized attention mechanism. It jointly incorporates appropriate common spatial-frequency representations of pairwise source and target but also cross-source domains, without the effect of distribution discrepancy among source domains. Additionally, considering the curse of dimensionality, our TSFAN is approximately represented in Tucker format. Benefiting the low-rank Tucker Network, the TSFAN can scale linearly in the number of domains, providing us the great flexibility to extend TSFAN to the case associated with an arbitrary number of sessions. Main results. Extensive experiments on the representative benchmarks demonstrate the effectiveness of TSFAN in EEG-based biometric recognition, outperforming state-of-the-art approaches, as verified by cross-session validation. Significance. The proposed TSFAN aims to investigate the presence of consistent EEG identity features across sessions. It is achieved by utilizing a novel tensorized attention mechanism that collaborates intra-source transferable information with inter-source interactions, while remaining unaffected by domain shifts in multiple source domains. Furthermore, the electrode selection shows that EEG-based identity features across sessions are distributed across brain regions, and 20 electrodes based on 10–20 standard system are able to extract stable identity information.

Video domain adaptation for semantic segmentation using perceptual consistency matching

Unsupervised domain adaptation (UDA) aims to transfer knowledge in previous and related labeled datasets (sources) to a new unlabeled dataset (target). Despite the impressive performance, existing approaches have largely focused on image-based UDA only, and video-based UDA has been relatively understudied and received less attention due to the difficulty of adapting diverse modal video features and modeling temporal associations efficiently. To address this, existing studies use optical flow to capture motion cues between in-domain consecutive frames, but is limited by heavy compute requirements and modeling flow patterns across diverse domains is equally challenging. In this work, we propose an adversarial domain adaptation approach for video semantic segmentation that aims to align temporally associated pixels in successive source and target domain frames without relying on optical flow. Specifically, we introduce a Perceptual Consistency Matching (PCM) strategy that leverages perceptual similarity to identify pixels with high correlation across consecutive frames, and infer that such pixels should correspond to the same class. Therefore, we can enhance prediction accuracy for video-UDA by enforcing consistency not only between in-domain frames, but across domains using PCM objectives during model training. Extensive experiments on public datasets show the benefit of our approach over existing state-of-the-art UDA methods. Our approach not only addresses a crucial task in video domain adaptation but also offers notable improvements in performance with faster inference times.

Limited value of EEG source imaging for decoding hand movement and imagery in youth with brain lesions

ABSTRACT Brain–computer interfaces based on electroencephalography (EEG) often exhibit unreliable performance in the classification of motor tasks. Recent research has shown that EEG source imaging (ESI) has the potential to outperform sensor domain approaches in various movement decoding tasks. However, ESI research to date has predominantly focused on the adult population, so its performance in youth with disabilities is unknown. In this study, we compared the offline classification performance of two ESI approaches (with and without modeling white matter conductivity anisotropy) to that of a sensor domain approach in the classification of left- versus right-hand movement execution and imagery tasks. Magnetic resonance images (MRI) were acquired from nine pediatric participants with brain lesions. Subsequently, cortical activity was recorded from 64 channels. MRI data were used to estimate participant-specific EEG sources. Various feature extraction and classification approaches were investigated in both sensor and source domains. Generally, ESI classification performance did not exceed chance levels and was statistically equivalent to sensor approaches except for isolated participants. However, ESI offered +9.61% improvement over the sensor domain (p = 0.031) in decoding motor execution in a participant with unilateral ventriculomegaly. Future research ought to delineate the specific task and participant characteristics which warrant the source domain approach.

Detecting Offensive Language on Malay Social Media: A Zero-Shot, Cross-Language Transfer Approach Using Dual-Branch mBERT

Social media serves as a platform for netizens to stay informed and express their opinions through the Internet. Currently, the social media discourse environment faces a significant security threat—offensive comments. A group of users posts comments that are provocative, discriminatory, and objectionable, intending to disrupt online discussions, provoke others, and incite intergroup conflict. These comments undermine citizens’ legitimate rights, disrupt social order, and may even lead to real-world violent incidents. However, current automatic detection of offensive language primarily focuses on a few high-resource languages, leaving low-resource languages, such as Malay, with insufficient annotated corpora for effective detection. To address this, we propose a zero-shot, cross-language unsupervised offensive language detection (OLD) method using a dual-branch mBERT transfer approach. Firstly, using the multi-language BERT (mBERT) model as the foundational language model, the first network branch automatically extracts features from both source and target domain data. Subsequently, Sinkhorn distance is employed to measure the discrepancy between the source and target language feature representations. By estimating the Sinkhorn distance between the labeled source language (e.g., English) and the unlabeled target language (e.g., Malay) feature representations, the method minimizes the Sinkhorn distance adversarially to provide more stable gradients, thereby extracting effective domain-shared features. Finally, offensive pivot words from the source and target language training sets are identified. These pivot words are then removed from the training data in a second network branch, which employs the same architecture. This process constructs an auxiliary OLD task. By concealing offensive pivot words in the training data, the model reduces overfitting and enhances robustness to the target language. In the end-to-end framework training, the combination of cross-lingual shared features and independent features culminates in unsupervised detection of offensive speech in the target language. The experimental results demonstrate that employing cross-language model transfer learning can achieve unsupervised detection of offensive content in low-resource languages. The number of labeled samples in the source language is positively correlated with transfer performance, and a greater similarity between the source and target languages leads to better transfer effects. The proposed method achieves the best performance in OLD on the Malay dataset, achieving an F1 score of 80.7%. It accurately identifies features of offensive speech, such as sarcasm, mockery, and implicit expressions, and showcases strong generalization and excellent stability across different target languages.

DW-D3A: dynamic weighted dual-driven domain adaptation for cross-scene hyperspectral image classification

ABSTRACT Domain adaptation (DA) offers an effective way to align feature distributions of the source domain (SD) and the target domain (TD) without requiring any target label samples. As a method of DA, representation learning effectively realizes the alignment of feature distributions in different domains by transferring domain knowledge. However, existing representation learning methods often focus on unilateral representation transfer, which potentially results in transfer bias. Additionally, most methods ignore the connection between domain alignment and discrimination during the DA process, which easily causes negative transfer. This paper proposes a dynamic weighted dual-driven domain adaptation (DW-D 3 A) model that effectively addresses the aforementioned issues through bilateral feature transfer between domains and a dynamic weighted scheme. Technically, we first propose a dual-driven domain adaptation (D 3 A) model, which employs symmetrical structures to facilitate the knowledge transfer of bilateral representations between source and target domain samples, learning the subspaces of two domains and reducing distribution discrepancies between subspaces via joint distribution-driven alignment. This process mitigates transfer bias and goes beyond previous unilateral transfer methods. Then, to alleviate strong constraints on projecting SD and TD into the same subspace in existing approaches, we apply a relaxed subspace constraint to bring the projections of SD and TD closer. Furthermore, data reconstruction is incorporated to preserve discriminant information from the original data. Lastly, we expand (D 3 A) to DW-D 3 A using a dynamic weighted scheme, which adjusts the weights assigned to domain alignment and discrimination based on their significance to inhibit negative transfer. Extensive experimentation on three datasets indicates that DW-D 3 A outperforms seven other DA methods, showing its superior performance.