Applied Intelligence最新文献

In today's information age, table images play a crucial role in storing structured information, making table image recognition technology an essential component in many fields. However, accurately recognizing the structure and text content of various complex table images has remained a challenge. Recently, large language models (LLMs) have demonstrated exceptional capabilities in various natural language processing tasks. Therefore, applying LLMs to the correction tasks of structure and text content after table image recognition presents a novel solution. This paper introduces a new method, TableGPT, which combines table recognition with LLMs and develops a specialized multimodal agent to enhance the effectiveness of table image recognition. Our approach is divided into four stages. In the first stage, TableGPT_agent initially evaluates whether the input is a table image and, upon confirmation, uses algorithms such as the transformer for preliminary recognition. In the second stage, the agent converts the recognition results into HTML format and autonomously assesses whether corrections are needed. If corrections are needed, the data are input into a trained LLM to achieve more accurate table recognition and optimization. In the third stage, the agent evaluates user satisfaction through feedback and applies superresolution algorithms to low-quality images, as this is often the main reason for user dissatisfaction. Finally, the agent inputs both the enhanced and original images into the trained model, integrating the information to obtain the optimal table text representation. Our research shows that trained LLMs can effectively interpret table images, improving the Tree Edit Distance Similarity (TEDS) score by an average of 4% even when based on the best current table recognition methods, across both public and private datasets. They also demonstrate better performance in correcting structural and textual errors. We also explore the impact of image superresolution technology on low-quality table images. Combined with the LLMs, our TEDS score significantly increased by 54%, greatly enhancing the recognition performance. Finally, by leveraging agent technology, our multimodal model improved table recognition performance, with the TEDS score of TableGPT_agent surpassing that of GPT-4 by 34%.

Federated Knowledge Graph Embedding (FKGE) has recently garnered considerable interest due to its capacity to extract expressive representations from distributed knowledge graphs, while concurrently safeguarding the privacy of individual clients. Existing FKGE methods typically harness the arithmetic mean of entity embeddings from all clients as the global supplementary knowledge, and learn a replica of global consensus entities embeddings for each client. However, these methods usually neglect the inherent semantic disparities among distinct clients. This oversight not only results in the globally shared complementary knowledge being inundated with too much noise when tailored to a specific client, but also instigates a discrepancy between local and global optimization objectives. Consequently, the quality of the learned embeddings is compromised. To address this, we propose Personalized Federated knowledge graph Embedding with client-wise relation Graph (PFedEG), a novel approach that employs a client-wise relation graph to learn personalized embeddings by discerning the semantic relevance of embeddings from other clients. Specifically, PFedEG learns personalized supplementary knowledge for each client by amalgamating entity embedding from its neighboring clients based on their “affinity” on the client-wise relation graph. Each client then conducts personalized embedding learning based on its local triples and personalized supplementary knowledge. We conduct extensive experiments on four benchmark datasets to evaluate our method against state-of-the-art models and results demonstrate the superiority of our method.

The class-conditional image generation based on diffusion models is renowned for generating high-quality and diverse images. However, most prior efforts focus on generating images for general categories, e.g., 1000 classes in ImageNet-1k. A more challenging task, large-scale fine-grained image generation, remains the boundary to explore. In this work, we present a parameter-efficient strategy, called FineDiffusion, to fine-tune large pre-trained diffusion models scaling to large-scale fine-grained image generation with 10,000 categories. FineDiffusion significantly accelerates training and reduces storage overhead by only fine-tuning tiered class embedder, bias terms, and normalization layers’ parameters. To further improve the image generation quality of fine-grained categories, we propose a novel sampling method for fine-grained image generation, which utilizes superclass-conditioned guidance, specifically tailored for fine-grained categories, to replace the conventional classifier-free guidance sampling. Compared to full fine-tuning, FineDiffusion achieves a remarkable 1.56(times ) training speed-up and requires storing merely 1.77% of the total model parameters, while achieving state-of-the-art FID of 9.776 on image generation of 10,000 classes. Extensive qualitative and quantitative experiments demonstrate the superiority of our method compared to other parameter-efficient fine-tuning methods. The code and more generated results are available at our project website: https://finediffusion.github.io/.

Conflicts serve as an important focus of uncertainty analysis, and their reductions facilitate the issue identification and conflict solving to become valuable but rare. At present, conflict analysis reductions mainly embrace relation matrices, and they never concern uncertainty measures with highly concentrated information. In this paper, three-way reductions of conflict analysis are transferred from relation matrices to integration measures, and corresponding heuristic reduction algorithms are constructed for information systems. At first, three-way membership degrees and three-way similarity degrees are proposed for conflict analysis, and their measurement boundedness, issue monotonicity, calculation algorithm, and transformation interrelationship are researched. Then, alliance, conflict, and neutrality reductions are proposed based on similarity degrees to acquire heuristic reduction algorithms, and they can be equivalently characterized by both membership degrees and relation matrices. Finally by table examples and data experiments, similarity degrees and relevant measurement properties are validated, and two groups of three-way reduction algorithms related to relation matrices and similarity degrees are comparatively analyzed; as a result, three-way reduction algorithms based on similarity degrees become novel and effective for conflict analysis. This study provides an in-depth insight into three-way reductions of conflict analysis from algebraic measurement.

Trust is a crucial factor affecting the adoption of machine learning (ML) models. Qualitative studies have revealed that end-users, particularly in the medical domain, need models that can express their uncertainty in decision-making allowing users to know when to ignore the model’s recommendations. However, existing approaches for quantifying decision-making uncertainty are not model-agnostic, or they rely on complex mathematical derivations that are not easily understood by laypersons or end-users, making them less useful for explaining the model’s decision-making process. This work proposes a set of class-independent meta-heuristics that can characterise the complexity of an instance in terms of factors that are mutually relevant to both human and ML decision-making. The measures are integrated into a meta-learning framework that estimates the risk of misclassification. The proposed framework outperformed predicted probabilities and entropy-based methods of identifying instances at risk of being misclassified. Furthermore, the proposed approach resulted in uncertainty estimates that proves more independent of model accuracy and calibration than existing approaches. The proposed measures and framework demonstrate promise for improving model development for more complex instances and provides a new means of model abstention and explanation.

In today’s technology-driven era, the proliferation of data is inevitable across various domains. Within engineering, sciences, and business domains, particularly in the context of big data, it can extract actionable insights that can revolutionize the field. Amid data management and analysis, patterns or groups of interconnected data points, commonly referred to as clusters, frequently emerge. These clusters represent distinct subsets containing closely related data points, showcasing unique characteristics compared to other clusters within the same dataset. Spanning across disciplines such as physics, biology, business, and sales, clustering is important in understanding these novel datasets’ essential characteristics, developing complex statistical models, and testing various hypotheses. However, interpreting the characteristics and physical implications of generated clusters by different clustering algorithms is challenging for researchers unfamiliar with these algorithms’ inner workings. This research addresses the intricacies of comprehending data clustering, cluster attributes, and evaluation metrics, especially for individuals lacking proficiency in clustering or related disciplines like statistics. The primary objective of this study is to simplify cluster analysis by furnishing users or analysts from diverse domains with succinct linguistic synopses of clustering results, circumventing the necessity for intricate numerical or mathematical terms. Deep learning techniques based on large language models, such as encoder-decoders (for example, the T5 model) and generative pre-trained transformers (GPTs), are employed to achieve this. This study aims to construct a summarization model capable of ingesting data clusters, producing a condensed overview of the contained insights in a simplified, easily understandable linguistic format. The evaluation process revealed a clear preference among evaluators for the summaries generated by GPT, with T5 summaries following closely behind. GPT and T5 summaries were good at fluency, demonstrating their ability to capture the original content in a human-like manner. In contrast, while providing a structured framework for summarization, the linguistic protoform-based approach is needed to match the quality and coherence of the GPT and T5 summaries.

Face detection serves as the core foundation for applications such as face analysis, recognition and reconstruction. In dense scenarios, the target scale difference is significant, and the instance pixels are too small as well as the mutual occlusion is serious leading to inconspicuous feature representation. However, existing detection methods rely on convolutional and pooling layers for feature extraction, with insufficient deep feature extraction and limited inference capability, leading to inaccurate recognition and high leakage rate. Therefore, we propose a small-scale face detection model YOLO-SXS based on the extended Transformer structure, which makes full use of contextual information and feature fusion networks to significantly improve the detection performance for small-scale and occluded faces. Specifically, the fusion of Swin Transformer and Convolutional Neural Networks (CNN) for feature extraction enhances the network’s ability to perceive global features; the Space to Depth (SPD-Conv) mapping is used to improve the network’s feature extraction in low-resolution and small-target detection tasks; furthermore, by adding fine-grained features, YOLO-SXS can significantly improve its performance for small-scale and occluded face detection capability; in addition, by adding a fine-grained feature fusion layer, feature information is retained to the maximum extent, which effectively reduces the loss of target information. The performance evaluation was performed on WIDER FACE, SCUT-HEAD and FDDB datasets, and the experimental results show that our proposed method significantly improves the performance of recognizing small-sized faces and achieves high detection rate and low error rate.

The driving conditions in mining areas are complex, and developing a suitable automatic shifting strategy for mining trucks is crucial. However, the development of automatic shifting strategies faces challenges, as it relies on experience and historical experimental data, which are the highest commercial secrets of manufacturers. In recent years, some shifting strategies based on artificial intelligence technologies have been implemented. However, many people shift gears based on the current state of the vehicle, ignoring the influence of historical data. There is a potential risk of mis-shift when unexpected sensor data is received, and continuously shifting gears in a short period of time can increase the likelihood of transmission damage, affecting the driving experience. To this end, this study proposes a novel gear shifting prediction method based on a multi-parameter Bi-directional Long Short-Term Memory(Bi-LSTM) network operating over continuous time periods. Real-time vehicle state data is collected via the CAN bus and 9 parameters that are positively correlated with gear shifting are selected through R/S analysis. By inputting values of those 9 parameters within continuous time periods into the machine learning model, gear shifting prediction is conducted. The experimental results show that our model predicts gear shifting with 96.85% accuracy while its average time cost is around 3.86 ms, meeting the real-time processing requirement. The model balances prediction accuracy and time consumption, and it overcomes the impact of transient abnormal sensor data. Hence, it has the potential for wide application in predictive models based on data with temporal characteristics.

Aspect Sentiment Triple Extraction (ASTE) is an emerging task in sentiment analysis that aims to extract triplets consisting of aspect terms, opinion terms, and sentiment polarity from review texts. Previous span-based methods often struggle with accurately identifying the boundaries of aspect and opinion terms, especially when multiple word spans appear in a sentence. This limitation arises from their reliance on a single, simplistic approach to constructing contextual features. To address these challenges, we propose Multi-Perspective Boundary Enhancement Network (MPBE). The network captures rich contextual features by adopting a dual-encoder mechanism and constructs multiple channels to further enhance these features. Specifically, we introduce enhanced semantic and syntactic information in two channels, while the third channel transforms the features using discrete fourier transform. In addition, we design a dual-graph cross fusion module to fuse features from different channels for more efficient information interaction and integration. Finally, by statistically analyzing the length distribution of aspect and opinion terms, a candidate length-based decoding strategy is proposed to achieve more accurate decoding. In experiments, the proposed MPBE model achieved excellent results on four benchmark datasets (14Lap, 14Res, 15Res, 16Res), with F1 scores of 62.32%, 73.78%, 65.32%, and 73.36%, respectively, demonstrating the superiority of the method.