As Machine Learning (ML) and Artificial Intelligence (AI) progress rapidly, the issue of ML model generalization has emerged as a critical concern for academics and practitioners alike. In practical scenarios, it is essential for models to sustain high performance when encountering varied and novel data distributions. Nevertheless, current domain generalization techniques have their shortcomings in tackling this challenge. The objective of this paper is to introduce a novel Meta-Learning approach, incorporating Fourier transform-based Data augmentation, called MLFD, for the purpose of domain generalization. Utilizing both data augmentation and a meta-learning architecture, this proposed technique empowers models to extend their generalization to multiple unseen target domains using just a single training domain. In contrast to other domain generalization methods, the method presented in this paper achieves comparable accuracy on the Digits-DG datasets, and demonstrates substantial improvements in terms of reducing model training time.

Underwater object detection technology is essential for maintaining marine ecological health and supporting economic development. However, the underwater environment poses significant challenges, including low contrast, small object sizes, and complex backgrounds. Existing generic object detectors often fail to identify these organisms effectively. This paper proposes a Joint Multi-scale channel attention and Multi-perception head Network (JMM-Net), a detection algorithm for underwater organisms. JMM-Net comprises three main components: Multi-Scale Channel Attention (MSCA)-based backbone network, Multi-Perception Parallel detection head (MPPhead), and lightweight GSconv-Path Aggregation Network (GS-PAN). MSCA is embedded into the backbone to enhance feature extraction for blurred and small-sized objects in low-quality environments by integrating local and global channel attention through multi-scale parallel sub-networks and cross-channel learning. MPPhead enhances the model’s classification and localization capabilities by leveraging scale, spatial, and task perception, thereby enhancing the detection of marine organisms in complex backgrounds. The adoption of GS-PAN over the traditional Path Aggregation Network (PAN) structure significantly reduces the model’s parameters and computational load, making it more suitable for deployment on edge devices. Extensive experiments on three public underwater datasets demonstrate that our method achieves excellent performance on underwater object detection at a lightweight cost.

The objective of knowledge graph completion is to comprehend the structure and inherent relationships of domain knowledge, thereby providing a valuable foundation for knowledge reasoning and analysis. However, existing methods for knowledge graph completion face challenges. For instance, rule-based completion methods exhibit high accuracy and interpretability, but encounter difficulties when handling large knowledge graphs. In contrast, embedding-based completion methods demonstrate strong scalability and efficiency, but also have limited utilisation of domain knowledge. In response to the aforementioned issues, we propose a method of pre-training and inference for knowledge graph completion based on integrated rules. The approach combines rule mining and reasoning to generate precise candidate facts. Subsequently, a pre-trained language model is fine-tuned and probabilistic structural loss is incorporated to embed the knowledge graph. This enables the language model to capture more deep semantic information while the loss function reconstructs the structure of the knowledge graph. This enables the language model to capture more deep semantic information while the loss function reconstructs the structure of the knowledge graph. Extensive tests using various publicly accessible datasets have indicated that the suggested model performs better than current techniques in tackling knowledge graph completion problems.

With the advent of the era of big data, an increasing amount of duplicate data are expressed in different forms. In order to reduce redundant data storage and improve data quality, data deduplication technology has never become more significant than nowadays. It is usually necessary to connect multiple data tables and identify different records pointing to the same entity, especially in the case of multi-source data deduplication. Active learning trains the model by selecting the data items with the maximum information divergence and reduces the data to be annotated, which has unique advantages in dealing with big data annotations. However, most of the current active learning methods only employ classical entity matching and are rarely applied to data deduplication tasks. To fill this research gap, we propose a novel graph deep active learning framework for data deduplication, which is based on similarity algorithms combined with the bidirectional encoder representations from transformers (BERT) model to extract the deep similarity features of multi-source data records, and first introduce the graph active learning strategy to build a clean graph to filter the data that needs to be labeled, which is used to delete the duplicate data that retain the most information. Experimental results on real-world datasets demonstrate that the proposed method outperforms state-of-the-art active learning models on data deduplication tasks.

Although the Faster Region-based Convolutional Neural Network (Faster R-CNN) model has obvious advantages in defect recognition, it still cannot overcome challenging problems, such as time-consuming, small targets, irregular shapes, and strong noise interference in bridge defect detection. To deal with these issues, this paper proposes a novel Multi-scale Feature Fusion (MFF) model for bridge appearance disease detection. First, the Faster R-CNN model adopts Region Of Interest (ROI) pooling, which omits the edge information of the target area, resulting in some missed detections and inaccuracies in both detecting and localizing bridge defects. Therefore, this paper proposes an MFF based on regional feature Aggregation (MFF-A), which reduces the missed detection rate of bridge defect detection and improves the positioning accuracy of the target area. Second, the Faster R-CNN model is insensitive to small targets, irregular shapes, and strong noises in bridge defect detection, which results in a long training time and low recognition accuracy. Accordingly, a novel Lightweight MFF (namely MFF-L）model for bridge appearance defect detection using a lightweight network EfficientNetV2 and a feature pyramid network is proposed, which fuses multi-scale features to shorten the training speed and improve recognition accuracy. Finally, the effectiveness of the proposed method is evaluated on the bridge disease dataset and public computational fluid dynamic dataset.

As a class of effective methods for incomplete multi-view clustering, graph-based algorithms have recently drawn wide attention. However, most of them could use further improvement regarding the following aspects. First, in some graph-based models, all views are forced to share a common similarity graph regardless of the severe consistency degeneration due to incomplete views. Next, similarity graph construction and cluster analysis are sometimes performed separately. Finally, the contribution difference of individual views is not always carefully considered. To address these issues simultaneously, this paper proposes an incomplete multi-view clustering algorithm based on auto-weighted fusion in partition space. In our algorithm, the information of cluster structure is introduced into the process of similarity learning to construct a desirable similarity graph, information fusion is performed in partition space to alleviate the negative impact brought about by consistency degradation, and all views are adaptively weighted to reflect their different contributions to clustering tasks. Finally, all the subtasks are collaboratively optimized in a united framework to reach an overall optimal result. Experimental results show that the proposed method compares favorably with the state-of-the-art methods.

The aspect-based sentiment analysis (ABSA) consists of two subtasks—aspect term extraction and aspect sentiment prediction. Most methods conduct the ABSA task by handling the subtasks in a pipeline manner, whereby problems in performance and real application emerge. In this study, we propose an end-to-end ABSA model, namely, SSi-LSi, which fuses the syntactic structure information and the lexical semantic information, to address the limitation that existing end-to-end methods do not fully exploit the text information. Through two network branches, the model extracts syntactic structure information and lexical semantic information, which integrates the part of speech, sememes, and context, respectively. Then, on the basis of an attention mechanism, the model further realizes the fusion of the syntactic structure information and the lexical semantic information to obtain higher quality ABSA results, in which way the text information is fully used. Subsequent experiments demonstrate that the SSi-LSi model has certain advantages in using different text information.

Dialog State Tracking (DST) aims to extract the current state from the conversation and plays an important role in dialog systems. Existing methods usually predict the value of each slot independently and do not consider the correlations among slots, which will exacerbate the data sparsity problem because of the increased number of candidate values. In this paper, we propose a multi-domain DST model that integrates slot-relevant information. In particular, certain connections may exist among slots in different domains, and their corresponding values can be obtained through explicit or implicit reasoning. Therefore, we use the graph adjacency matrix to determine the correlation between slots, so that the slots can incorporate more slot-value transformer information. Experimental results show that our approach has performed well on the Multi-domain Wizard-Of-Oz (MultiWOZ) 2.0 and MultiWOZ2.1 datasets, demonstrating the effectiveness and necessity of incorporating slot-relevant information.

The aspect-based sentiment analysis (ABSA) consists of two subtasks'aspect term extraction and aspect sentiment prediction. Existing methods deal with both subtasks one by one in a pipeline manner, in which there lies some problems in performance and real application. This study investigates the end-to-end ABSA and proposes a novel multitask multiview network (MTMVN) architecture. Specifically, the architecture takes the unified ABSA as the main task with the two subtasks as auxiliary tasks. Meanwhile, the representation obtained from the branch network of the main task is regarded as the global view, whereas the representations of the two subtasks are considered two local views with different emphases. Through multitask learning, the main task can be facilitated by additional accurate aspect boundary information and sentiment polarity information. By enhancing the correlations between the views under the idea of multiview learning, the representation of the global view can be optimized to improve the overall performance of the model. The experimental results on three benchmark datasets show that the proposed method exceeds the existing pipeline methods and end-to-end methods, proving the superiority of our MTMVN architecture.

As an important branch of natural language processing, sentiment analysis has received increasing attention. In teaching evaluation, sentiment analysis can help educators discover the true feelings of students about the course in a timely manner and adjust the teaching plan accurately and timely to improve the quality of education and teaching. Aiming at the inefficiency and heavy workload of college curriculum evaluation methods, a Multi-Attention Fusion Modeling (Multi-AFM) is proposed, which integrates global attention and local attention through gating unit control to generate a reasonable contextual representation and achieve improved classification results. Experimental results show that the Multi-AFM model performs better than the existing methods in the application of education and other fields.