Neurocomputing最新文献

{"title":"Boosting accuracy of student models via Masked Adaptive Self-Distillation","authors":"Haoran Zhao ,&nbsp;Shuwen Tian ,&nbsp;Jinlong Wang ,&nbsp;Zhaopeng Deng ,&nbsp;Xin Sun ,&nbsp;Junyu Dong","doi":"10.1016/j.neucom.2025.129988","DOIUrl":"10.1016/j.neucom.2025.129988","url":null,"abstract":"<div><div>Knowledge distillation (KD) has achieved impressive success, yet conventional KD approaches are time-consuming and computationally costly. In contrast, self-distillation methods provide an efficient alternative. However, existing self-distillation methods mostly suffer from information redundancy due to the same network architecture from the teacher and student models. Additionally, they simultaneously face the inherent limitation of lacking a high-capacity teacher model. To cope with the above challenges, we propose a novel and efficient method named Masked Adaptive Self-Distillation (MASD). Specifically, we first introduce the Mask Generation Module, which masks random pixels of the feature maps and force it to reconstruct and refine more valuable features on different layers. Moreover, the Adaptive Weighting Mechanism is designed to dynamically adjust and optimize the weights of supervisory signals utilizing the probabilities from the mutual masked supervisory signals, thereby compensating the absence of high-capacity teacher model. We demonstrate the effectiveness of our MASD method on conventional image classification datasets and fine-grained datasets using state-of-the-art CNN architectures, and show that MASD significantly enhances the generalization of various backbone networks. For instance, on the CIFAR-100 classification benchmark, the proposed MASD method achieves an accuracy of 80.40% with the ResNet-18 architecture, surpassing the baseline with a 4.16% margin in Top-1 accuracy.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 129988"},"PeriodicalIF":5.5,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143706306","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":"DASC: Learning discriminative latent space for video clustering","authors":"Jiaxin Lin,&nbsp;Xizhan Gao,&nbsp;Zhihan Zhang,&nbsp;Haotian Deng","doi":"10.1016/j.neucom.2025.130050","DOIUrl":"10.1016/j.neucom.2025.130050","url":null,"abstract":"<div><div>In recent years, significant advancements have been made in video analysis technologies. However, most existing methods are primarily designed for supervised learning, particularly in video classification. Accurately labeling video data is often time-consuming and labor-intensive, making large-scale annotation challenging. As a result, most of the available video data remain in an unsupervised or weakly supervised state. This situation underscores the urgent need to develop efficient methods for unsupervised video data analysis, with a particular emphasis on video clustering techniques, which can effectively alleviate the high cost and labor intensity of video data annotation by automatically grouping similar videos, thus reducing the reliance on manual labeling. This significantly enhances the efficiency and scalability of video analysis. In this paper, we propose a deep aggregation subspace clustering (DASC) network, designed to learn a video-level self-representation matrix in an end-to-end manner, without the need for any labeled data, thus operating in an unsupervised learning environment. Specifically, DASC consists of four main components: auto-encoder backbone, video modeling module (VMM), self-representation module (SrM) and feature recovered module (FRM). A frame-level latent space is first established by utilizing the auto-encoder backbone. Then, a video-level latent space is established by constructing the VMM. Next, the video-level self-representation matrix is learned in the latent space by using the SrM. Finally, the video-level latent feature will be restored to frame-level features using the FRM. Experimental results on multiple benchmark datasets demonstrate the effectiveness of the proposed method.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130050"},"PeriodicalIF":5.5,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143706327","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":"MLC-DTA: Drug-target affinity prediction based on multi-level contrastive learning and equivariant graph neural networks","authors":"Mengxin Zheng,&nbsp;Guicong Sun,&nbsp;Yongxian Fan","doi":"10.1016/j.neucom.2025.130052","DOIUrl":"10.1016/j.neucom.2025.130052","url":null,"abstract":"<div><div>With the development of computer-aided drug design in the field of pharmaceutical research, drug-target affinity (DTA) prediction is of great significance for compound screening and drug development. Recent studies have widely adopted deep learning techniques for DTA prediction, focusing on feature extraction from sequences and graph structures. Despite progress, these methods often overlook interaction relationships at the network level. Moreover, graph-based molecular representation methods relying on graph neural networks (GNNs) fail to incorporate 3D molecular structural information, limiting their potential for DTA prediction. To overcome these issues, we propose MLC-DTA, a computational method specifically designed for drug-target affinity prediction tasks. MLC-DTA uses equivariant graph neural networks to extract the structural features of drugs and targets, retaining the geometric equivariance of proteins while capturing the specific structural information of drug molecules. In addition, it integrates the interaction relationships between drugs and targets to obtain the network-level features, understanding the DTA interactions from multiple perspectives of molecules and networks. Finally, a contrastive learning strategy is introduced to maximize mutual information at both the molecular and network levels, thereby improving the prediction performance. Comparative experiments and case analyses on two datasets show that MLC-DTA has a significant improvement in accuracy.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130052"},"PeriodicalIF":5.5,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739385","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":"PUAL: A classifier on trifurcate positive-unlabelled data","authors":"Xiaoke Wang ,&nbsp;Xiaochen Yang ,&nbsp;Rui Zhu ,&nbsp;Jing-Hao Xue","doi":"10.1016/j.neucom.2025.130080","DOIUrl":"10.1016/j.neucom.2025.130080","url":null,"abstract":"<div><div>Positive-unlabelled (PU) learning aims to train a classifier using the data containing only labelled-positive instances and unlabelled instances. However, existing PU learning methods are generally hard to achieve satisfactory performance on trifurcate data, where the positive instances distribute on both sides of the negative instances. To address this issue, firstly we propose a PU classifier with asymmetric loss (PUAL), by introducing a structure of asymmetric loss on positive instances into the objective function of the global and local learning classifier. Then we develop a kernel-based algorithm to enable PUAL to obtain non-linear decision boundary. We show that, through experiments on both simulated and real-world datasets, PUAL can achieve satisfactory classification on trifurcate data.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130080"},"PeriodicalIF":5.5,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143715508","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":"Robust attack-aware spread spectrum watermarking in real scenes","authors":"Huijuan Guo ,&nbsp;Baoning Niu ,&nbsp;Ying Huang ,&nbsp;Hu Guan ,&nbsp;Peng Zhao","doi":"10.1016/j.neucom.2025.130060","DOIUrl":"10.1016/j.neucom.2025.130060","url":null,"abstract":"<div><div>Digital image watermarking hides copyright information in digital images and allows for its extraction when necessary to confirm ownership. Imperceptibility and robustness, the key indices of watermarking performance, constrain each other and are affected by the embedding location and weight. Existing techniques resolve this constraint by prioritizing imperceptibility while endeavoring to maximize robustness, which still entails a risk of missing higher visual quality or a failure in watermark extraction in real scenes. When the spread spectrum scheme is applied to embed watermarks in the discrete cosine transform domain, the embedding location and weight are not well calibrated. The embedding location is heuristically selected from among the divisions of the frequency domain, and the determination of the embedding weight relies on the predetermined imperceptibility. We address the issue between imperceptibility and robustness with the opposite strategy, prioritizing robustness while maximizing imperceptibility, and propose the attack-aware spread spectrum watermarking (ASSW) algorithm. ASSW takes prior attacks into consideration when determining the embedding location and weight with three goals: ensuring the stability, small modulus and small weight of the feature vector of the embedding location. Our experiments indicate that, with a carefully calibrated embedding location and weight, ASSW achieves greater imperceptibility and robustness than the state-of-the-art methods both on average and individual images.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130060"},"PeriodicalIF":5.5,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143760269","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":"MetaGen: A framework for metaheuristic development and hyperparameter optimization in machine and deep learning","authors":"David Gutiérrez-Avilés ,&nbsp;Manuel Jesús Jiménez-Navarro ,&nbsp;José Francisco Torres ,&nbsp;Francisco Martínez-Álvarez","doi":"10.1016/j.neucom.2025.130046","DOIUrl":"10.1016/j.neucom.2025.130046","url":null,"abstract":"<div><div>Hyperparameter optimization is a pivotal step in enhancing model performance within machine learning. Traditionally, this challenge is addressed through metaheuristics, which efficiently explore large search spaces to uncover optimal solutions. However, implementing these techniques can be complex without adequate development tools, which is the primary focus of this paper. Hence, we introduce <span>MetaGen</span>, a novel Python package designed to provide a comprehensive framework for developing and evaluating metaheuristic algorithms. <span>MetaGen</span> follows best practices in Python design, ensuring minimalistic code implementation, intuitive comprehension, and full flexibility in solution representation. The package defines two distinct user roles: Developers, responsible for algorithm implementation for hyperparameter optimization, and Solvers, who leverage pre-implemented metaheuristics to address optimization problems. Beyond algorithm implementation, <span>MetaGen</span> facilitates benchmarking through built-in test functions, ensuring standardized performance comparisons. It also provides automated reporting and visualization tools to analyze optimization progress and outcomes effectively. Furthermore, its modular design allows distribution and integration into existing machine learning workflows. Several illustrative use cases are presented to demonstrate its adaptability and efficacy. The package, along with code, a user manual, and supplementary materials, is available at: <span><span>https://github.com/Data-Science-Big-Data-Research-Lab/MetaGen</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130046"},"PeriodicalIF":5.5,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143725090","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":"What learns next: Learning intents guided dual contrastive learning model for online course recommendation","authors":"Weiqiang Yao,&nbsp;Xiaohuan Hu","doi":"10.1016/j.neucom.2025.130051","DOIUrl":"10.1016/j.neucom.2025.130051","url":null,"abstract":"<div><div>Course recommender systems play a crucial role in assisting learners from diverse backgrounds in selecting suitable courses from various options to achieve their educational pursuits. However, these systems face challenges in accuracy and robustness due to partial learning preference modeling and unconventional long sequences sparsity. This paper proposes a learning intents guided dual contrastive learning model for online course recommendation to address these issues. Grounded in self-determination theory, the proposed model incorporates both intrinsic motivation (long-term learning intent) and extrinsic motivation (short-term learning intent) to characterize learners’ decision-making processes. Additionally, a dual contrastive learning module, comprising self-contrastive learning and last-course guided supervised contrastive learning, is designed to mitigate the sparsity issue and enhance the robustness of sequence representation learning. Extensive experiments demonstrate the effectiveness of the model compared to several state-of-the-art methods. Furthermore, the model exhibits excellent performance for learners with varying learning sequences.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130051"},"PeriodicalIF":5.5,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739378","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":"One multimodal plugin enhancing all: CLIP-based pre-training framework enhancing multimodal item representations in recommendation systems","authors":"Minghao Mo ,&nbsp;Weihai Lu ,&nbsp;Qixiao Xie ,&nbsp;Zikai Xiao ,&nbsp;Xiang Lv ,&nbsp;Hong Yang ,&nbsp;Yanchun Zhang","doi":"10.1016/j.neucom.2025.130059","DOIUrl":"10.1016/j.neucom.2025.130059","url":null,"abstract":"<div><div>With advances in multimodal pre-training, more efforts focus on integrating it into recommendation models. Current methods mainly focus on utilizing multimodal pre-training models to obtain multimodal representations of items and designing specific model architectures for downstream tasks. However, these methods often neglect the suitability of multimodal representations for recommendation systems since the pre-training is not conducted on recommendation datasets, making the directly obtained representations potentially suboptimal due to semantic biases from domain discrepancy and noise interference. Furthermore, collaborative information, a key element in recommendation systems, significantly impacts the effectiveness of recommendation models, but existing advanced multimodal pre-training models (e.g., CLIP) are unable to capture the collaborative information of items. To bridge the gap between multimodal pre-training models and recommendation systems, we propose a novel multimodal pre-training framework <strong>C</strong>LIP-based <strong>P</strong>re-training <strong>M</strong>ulti<strong>M</strong>odal (CPMM) item representations model for recommendation. First, the representations of images, text, and IDs are mapped to a new low-dimensional contrastive representation space for alignment and semantic enhancement, ensuring the consistency and robustness of the multimodal contrastive representation (MCR). A contrastive learning approach is designed to regulate the inter-modal distances, mitigating the impact of noise on recommendation performance. Finally, modeling of the first-order similarities of the items is conducted, thereby integrating the collaborative information of the items into the multimodal contrastive representations. Extensive experiments on Amazon benchmark datasets (Beauty, Toys, Tools) validate CPMM’s effectiveness across three core recommendation tasks: sequential recommendation, collaborative filtering, and click-through rate prediction.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130059"},"PeriodicalIF":5.5,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739558","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":"RECoT: Relation-enhanced Chains-of-Thoughts for knowledge-intensive multi-hop questions answering","authors":"Ronghan Li ,&nbsp;Dongdong Li ,&nbsp;Haowen Yang ,&nbsp;Xiaoxi Liu ,&nbsp;Haoxiang Jin ,&nbsp;RongCheng Pu ,&nbsp;Qiguang Miao","doi":"10.1016/j.neucom.2025.129903","DOIUrl":"10.1016/j.neucom.2025.129903","url":null,"abstract":"<div><div>Open Domain question answering is designed to enable a computer to understand and answer any question on a wide range of topics. The prevalent retrieval-reading paradigm helps large language models (LLMs) when retrieving relevant text from external knowledge sources using questions, however the multi-hop question answering approach based on Chains-of-Thoughts (CoT) may perform poorly when it comes to complex questions. This is because there can be errors in generating sentences at each hop, and these errors accumulate, leading to significant deviations in the final result. In order to solve the above problems, this paper first extracted the relational triples of complex problems. Next, triples are used to select the most representative sentence at each step during CoT generation as the query for the next-hop retrieval.</div><div>The RECoT with GPT-3 results in significant improvements with F1 score up 5.1 points in downstream QA on 2WikiMultihopQA datasets and up 2.9 points on HotpotQA datasets. In addition, improvements in results can be obtained even with smaller models such as Flan-T5-large without additional training. In conclusion, RECoT reduced model hallucination and accelerated more accurate CoT reasoning to guide retrieval to get improved results. Code is publicly available at <span><span>https://github.com/XD-BDIV-NLP/RECoT</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 129903"},"PeriodicalIF":5.5,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739556","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":"An adversarial contrastive learning based cross-modality zero-watermarking scheme for DIBR 3D video copyright protection","authors":"Xiyao Liu ,&nbsp;Qingyu Dang ,&nbsp;Huiyi Wang ,&nbsp;Xiaoheng Deng ,&nbsp;Xunli Fan ,&nbsp;Cundian Yang ,&nbsp;Zhihong Chen ,&nbsp;Hui Fang","doi":"10.1016/j.neucom.2025.130068","DOIUrl":"10.1016/j.neucom.2025.130068","url":null,"abstract":"<div><div>Copyright protection of depth image-based rendering (DIBR) videos has raised significant concerns due to their increasing popularity. Zero-watermarking, emerging as a powerful tool to protect the copyright of DIBR 3D videos, mainly relies on traditional feature extraction methods, thus necessitating improvements in robustness against complex geometric attacks and its ability to strike a balance between robustness and distinguishability. This paper presents a novel zero-watermarking scheme based on cross-modality feature fusion within a contrastive learning framework. Our approach integrates complementary information from 2D frames and depth maps using a cross-modality attention feature fusion mechanism to obtain discriminative features. Moreover, our features achieve a better trade-off between robustness and distinguishability by leveraging a designed contrastive learning strategy with an adversarial distortion simulator. Experimental results demonstrate our remarkable performance by reducing the false negative rates to around 0.2% when the false positive rate is equal to 0.5%, which is superior to the state-of-the-art zero-watermarking methods.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"637 ","pages":"Article 130068"},"PeriodicalIF":5.5,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143715510","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}