Neurocomputing最新文献

{"title":"Optimal decorrelated score subsampling for Cox regression with massive survival data","authors":"Yujing Shao ,&nbsp;Zhaohan Hou ,&nbsp;Lei Wang ,&nbsp;Heng Lian","doi":"10.1016/j.neucom.2025.130090","DOIUrl":"10.1016/j.neucom.2025.130090","url":null,"abstract":"<div><div>This paper investigates optimal subsampling strategies for the preconceived low-dimensional parameters of main interest in the presence of the nuisance parameters for Cox regression with massive survival data. A general subsampling decorrelated score function based on the log-partial likelihood is constructed to reduce the influence of the less accurate nuisance parameter estimation with a possibly slow convergence rate. The consistency and asymptotic normality of the resultant subsample estimators are established. We derive unified optimal subsampling probabilities based on A- and L-optimality criteria. A two-step algorithm is further proposed to implement practically, and the asymptotic properties of the resultant estimators are also given. The satisfactory performance of our proposed subsample estimators is demonstrated by simulation results and an airline dataset.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130090"},"PeriodicalIF":5.5,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143777433","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MCI-GRU: Stock prediction model based on multi-head cross-attention and improved GRU","authors":"Peng Zhu ,&nbsp;Yuante Li ,&nbsp;Yifan Hu ,&nbsp;Sheng Xiang ,&nbsp;Qinyuan Liu ,&nbsp;Dawei Cheng ,&nbsp;Yuqi Liang","doi":"10.1016/j.neucom.2025.130168","DOIUrl":"10.1016/j.neucom.2025.130168","url":null,"abstract":"<div><div>As financial markets become increasingly complex and the era of big data unfolds, accurate stock prediction has become more critical. Although traditional time series models, such as GRU, have been widely applied to stock prediction, they still exhibit limitations in addressing the intricate nonlinear dynamics of markets, particularly in the flexible selection and effective utilization of key historical information. In recent years, emerging methods like Graph Neural Networks and Reinforcement Learning have shown significant potential in stock prediction. However, these methods often demand high data quality and quantity, and they tend to exhibit instability when dealing with data sparsity and noise. Moreover, the training and inference processes for these models are typically complex and computationally expensive, limiting their broad deployment in practical applications. Existing approaches also generally struggle to capture unobservable latent market states effectively, such as market sentiment and expectations, microstructural factors, and participant behavior patterns, leading to an inadequate understanding of market dynamics and subsequently impact prediction accuracy. To address these challenges, this paper proposes a stock prediction model, MCI-GRU, based on a multi-head cross-attention mechanism and an improved GRU. First, we enhance the GRU model by replacing the reset gate with an attention mechanism, thereby increasing the model’s flexibility in selecting and utilizing historical information. Second, we design a multi-head cross-attention mechanism for learning unobservable latent market state representations, which are further enriched through interactions with both temporal features and cross-sectional features. Finally, extensive experiments conducted on the CSI 300 and CSI 500 datasets from the Chinese stock market, as well as the NASDAQ 100 and S&amp;P 500 datasets from the U.S. stock market, demonstrate that the proposed method outperforms the current state-of-the-art methods across multiple metrics. Furthermore, this approach has been successfully applied in the real-world operations of a fund management company, validating its effectiveness and practicality in actual financial environments.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130168"},"PeriodicalIF":5.5,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143785315","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Can large language models independently complete tasks? A dynamic evaluation framework for multi-turn task planning and completion","authors":"Jun Gao ,&nbsp;Junlin Cui ,&nbsp;Huijia Wu ,&nbsp;Liuyu Xiang ,&nbsp;Han Zhao ,&nbsp;Xiangang Li ,&nbsp;Meng Fang ,&nbsp;Yaodong Yang ,&nbsp;Zhaofeng He","doi":"10.1016/j.neucom.2025.130135","DOIUrl":"10.1016/j.neucom.2025.130135","url":null,"abstract":"<div><div>Large language models (LLMs) are increasingly relied upon for multi-turn dialogue to conduct complex tasks. However, existing benchmarks mainly evaluate LLMs as agents, overlooking their potential as independent systems to accomplish complex tasks. In addition, these benchmarks typically evaluate the planning and completion capabilities of the models individually, rather than simultaneously. To address these issues, we propose a new <strong>Dynamic Evaluation Framework for Multi-Turn task planning and completion (DEF-MT)</strong> to assess the ability of LLM to independently complete complex tasks in multi-turn scenarios. Our approach quantifies the model’s planning capability by guiding it to generate planning and responses sequentially. Simultaneously, we use a dynamic approach to generate data that simulates the complex intents of real users. Finally, experiments conducted on 9 mainstream models using the Multiwoz 2.2 dataset, indicate that the existing models’ sub-task planning capabilities hinder their ability to complete complex tasks, providing a meaningful reference for the future optimization direction of LLM.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130135"},"PeriodicalIF":5.5,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143783889","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Generative Prompt Controlled Diffusion for weakly supervised semantic segmentation","authors":"Wangyu Wu ,&nbsp;Tianhong Dai ,&nbsp;Zhenhong Chen ,&nbsp;Xiaowei Huang ,&nbsp;Fei Ma ,&nbsp;Jimin Xiao","doi":"10.1016/j.neucom.2025.130103","DOIUrl":"10.1016/j.neucom.2025.130103","url":null,"abstract":"<div><div>Weakly supervised semantic segmentation (WSSS), aiming to train segmentation models solely using image-level labels, has received significant attention. Existing approaches mainly concentrate on creating high-quality pseudo labels by utilizing existing images and their corresponding image-level labels. However, <strong><em>a major challenge</em></strong> arises when the available dataset is limited, as the quality of pseudo labels degrades significantly. In this paper, we tackle this challenge from a different perspective by introducing a novel approach called <em>Generative Prompt Controlled Diffusion</em> (GPCD) for data augmentation. This approach enhances the current labeled datasets by augmenting them with a variety of images, achieved through controlled diffusion guided by Generative Pre-trained Transformer (GPT) prompts. In this process, the existing images and image-level labels provide the necessary control information, while GPT enriches the prompts to generate diverse backgrounds. Moreover, we make an <strong><em>original contribution</em></strong> by integrating data source information as tokens into the Vision Transformer (ViT) framework, which improves the ability of downstream WSSS models to recognize the origins of augmented images. Our proposed GPCD approach clearly surpasses existing state-of-the-art methods, with its advantages being more pronounced when the available data is scarce, thereby demonstrating the effectiveness of our method. Our source code will be released.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130103"},"PeriodicalIF":5.5,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143783888","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MMamba: Enhancing image deraining with Morton curve-driven locality learning","authors":"Zongzhi Ouyang ,&nbsp;Wenhui Li","doi":"10.1016/j.neucom.2025.130161","DOIUrl":"10.1016/j.neucom.2025.130161","url":null,"abstract":"<div><div>Rain streaks in images exhibit complex forms and spatial structures. Their removal not only requires restoring the clarity of the image but also aims to preserve as much of the original information as possible. Transformer-based deep learning models have made significant progress in image deraining, effectively capturing the diversity of rain streaks and removing them through deep feature learning. However, when handling long sequence data, the computational complexity rises sharply, severely limiting the efficiency of practical applications. In recent years, the Mamba module, with its unique design, has demonstrated great potential in handling long sequences and enabling fast inference in image deraining tasks. It reduces computational complexity while preserving the underlying features of the image, thus improving efficiency. However, compared to Transformer-based models, the Mamba model still faces certain limitations in preserving local information and dealing with complex rain streaks. To address these issues, this paper proposes an efficient MMamba model, which combines the Morton curve-based state space model (SSM) to further enhance the ability to retain local information. Additionally, a dynamic channel attention mechanism is introduced, which significantly improves image detail restoration by partitioning channels and assigning learnable weights to each subset. Finally, for images with rich and complex rain streaks, a Selective Rain Stripe Compensator is proposed, effectively identifying and removing these intricate rain streaks. Extensive experiments on five commonly used benchmark datasets demonstrate that our method outperforms state-of-the-art techniques in terms of performance.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130161"},"PeriodicalIF":5.5,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143783887","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-resolution Patch-based Fourier Graph Spectral Network for spatiotemporal time series forecasting","authors":"Jiankai Zheng ,&nbsp;Liang Xie ,&nbsp;Haijiao Xu","doi":"10.1016/j.neucom.2025.130132","DOIUrl":"10.1016/j.neucom.2025.130132","url":null,"abstract":"<div><div>Spatiotemporal time series is of great significance in many fields such as finance and meteorology. Accurate forecasting depends on the effective understanding of their spatiotemporal correlations. However, existing methods have two major problems: firstly, the analysis of multi-resolution spatiotemporal correlations is insufficient; secondly, they are difficult to analyze in terms of spatiotemporal correlations efficiently. Some graph-based methods have high computational costs, while efficient methods struggle to learn spatiotemporal relationships. To overcome these problems, we introduce the Multi-resolution Patch-based Fourier Graph Spectral Network (MPFGSN), which leverages a hyperpatch graph and a novel graph spectral convolution to spatiotemporal time series prediction, enhancing temporal dependency learning. We first capture the multi-scale features of the time series through multi-resolution patches, then represent the spatiotemporal relations in the time series with the hyperpatch graph structure. Then, the Fourier Graph Spectral Network (FGSN) is proposed. The FGSN utilizes Fourier transformation and point multiplication operations in the spectral domain to implement efficient graph spectral convolution. And it maintains the capabilities of traditional multi-layer graph convolution while also being faster. Finally, it integrates information from different resolutions through a residual fusion mechanism, capturing both long-term trends and short-term fluctuations, thus enhancing the model’s robustness. Extensive experiments on public datasets confirm MPFGSN’s superiority in accuracy and generalization over existing models, validating its practical applicability in sophisticated forecasting scenarios.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130132"},"PeriodicalIF":5.5,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143785395","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Efficient spatio-temporal modeling and text-enhanced prototype for few-shot action recognition","authors":"Qian Zhang,&nbsp;Shuo Yan,&nbsp;Mingwen Shao,&nbsp;Hong Liang","doi":"10.1016/j.neucom.2025.130119","DOIUrl":"10.1016/j.neucom.2025.130119","url":null,"abstract":"<div><div>Few-Shot Action Recognition (FSAR) aims to classify new action categories accurately with only a limited number of labeled samples. Current methods face challenges in capturing spatiotemporal dynamics and integrating multimodal information effectively. This work presents a new framework that improves FSAR performance by improving spatiotemporal modeling and integrating cross-modal semantics. To capture complex spatiotemporal relationships in videos, we introduce two complementary modules: Temporal Enhancement Adaptation (TEA), which enhances temporal modeling capability, and Spatio-Temporal Fusion Adaptation (STFA), which integrates spatial and temporal features for better representations. Additionally, we propose the Text-Enhanced Prototype Module (TEPM), which strengthens prototype representations by fusing textual and visual features at multiple levels, improving the discriminability and generalization of prototypes. The experiments show that our approach achieves competitive performance on various benchmark datasets, confirming its effectiveness in FSAR.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130119"},"PeriodicalIF":5.5,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143777427","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Information bottleneck with input sampling for attribution","authors":"Bruno Coelho,&nbsp;Jaime S. Cardoso","doi":"10.1016/j.neucom.2025.130112","DOIUrl":"10.1016/j.neucom.2025.130112","url":null,"abstract":"<div><div>In order to facilitate the adoption of deep learning in areas where decisions are of critical importance, understanding the model’s internal workings is paramount. Nevertheless, since most models are considered black boxes, this task is usually not trivial, especially when the user does not have access to the network’s intermediate outputs. In this paper, we propose IBISA, a model-agnostic attribution method that reaches state-of-the-art performance by optimizing sampling masks using the Information Bottleneck Principle. Our method improves on the previously known RISE and IBA techniques by placing the bottleneck right after the image input without complex formulations to estimate the mutual information. The method also requires only twenty forward passes and ten backward passes through the network, which is significantly faster than RISE, which needs at least 4000 forward passes. We evaluated IBISA using a VGG-16 and a ResNET-50 model, showing that our method produces explanations comparable or superior to IBA, RISE, and Grad-CAM but much more efficiently.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130112"},"PeriodicalIF":5.5,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143777430","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Urban traffic flow prediction based on multi-spatio-temporal feature fusion","authors":"Hui Chen ,&nbsp;Jian Huang ,&nbsp;Yong Lu ,&nbsp;Jijie Huang","doi":"10.1016/j.neucom.2025.130117","DOIUrl":"10.1016/j.neucom.2025.130117","url":null,"abstract":"<div><div>Urban traffic flow exhibits non-linear characteristics and highly complex spatio-temporal correlations, posing a huge challenge to traffic flow prediction. To address this challenge, we constructed a novel Multi-Spatio-Temporal Feature Fusion (MSTFF) model. This model accurately depicts the spatio-temporal correlations of traffic flow from multiple spatio-temporal perspectives. It decomposes the non-linear traffic flow into stable trend and random fluctuation, and decomposes the topological graph of the road network into first-order adjacency, second-order in-degree and second-order out-degree relationship graphs which describe the connectivity, input and output relationship of the nodes in road network respectively, thus constructing multiple perspectives that can exhibit different spatio-temporal properties. For each perspective, we use the self-attention mechanism to learn the long-term spatio-temporal correlations in the traffic flow, and use convolutional neural networks to learn the short-term spatio-temporal correlations. By fusing the features of spatio-temporal dependencies from multiple perspectives, we explored the high-dimensional spatio-temporal features of traffic flow and predicted the future traffic flow. We also developed the traffic flow prediction algorithm based on MSTFF, verified its performance using four real-world datasets, and compared it with twelve baseline models in the past five years. Experimental results show that the performance of the MSTFF model far exceeds that of existing related models, with average improvement rates of 15.10%, 19.63%, and 13.59% respectively in terms of the three indicators of MAE, RMSE, and MAPE.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"638 ","pages":"Article 130117"},"PeriodicalIF":5.5,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143777429","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Adaptive blank compensation for few-shot image classification","authors":"Jianfeng Xu ,&nbsp;Baozhe Wang ,&nbsp;Huan Wan ,&nbsp;Pengxiang Su ,&nbsp;Xin Wei","doi":"10.1016/j.neucom.2025.130127","DOIUrl":"10.1016/j.neucom.2025.130127","url":null,"abstract":"<div><div>In recent years, few-shot learning has garnered significant attention from researchers. A notable trend in this domain involves the integration of feature distribution information from approximate base classes and the calibration of distributions stemming from a limited number of samples. However, the approximate classes found by existing methods are often not optimal. Through a limited number of samples in a novel class, approaching their class center is the key to determining the approximate base classes. We assume that the center of a novel class tends to be close to the center of the blank region between multiple approximate base classes. We verified this hypothesis and propose a calibration method based on it. With this method, a limited number of samples are compensated and calibrated in the direction close to the blank center of the approximate base class distribution. Finally, the more approximate base classes and more accurate distribution information can be obtained. Extensive experiments on miniImagenet, CUB, and CIFAR-FS datasets show that the proposed method achieves competitive performance. The code is available at <span><span>https://github.com/DN-KID/ABC</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130127"},"PeriodicalIF":5.5,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143760266","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}