Neural Networks最新文献

{"title":"Red alarm: Controllable backdoor attack in continual learning","authors":"Rui Gao,&nbsp;Weiwei Liu","doi":"10.1016/j.neunet.2025.107479","DOIUrl":"10.1016/j.neunet.2025.107479","url":null,"abstract":"<div><div>Continual learning (CL) studies the problem of learning a single model from a sequence of disjoint tasks. The main challenge is to learn without catastrophic forgetting, a scenario in which the model’s performance on previous tasks degrades significantly as new tasks are added. However, few works focus on the security challenge in the CL setting. In this paper, we focus on the backdoor attack in the CL setting. Specifically, we provide the threat model and explore what attackers in a CL setting will face. Based on these findings, we propose a controllable backdoor attack mechanism in continual learning (CBACL). Experimental results on the Split Cifar and Tiny Imagenet datasets confirm the advantages of our proposed mechanism.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107479"},"PeriodicalIF":6.0,"publicationDate":"2025-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143877169","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Saccade and purify: Task adapted multi-view feature calibration network for few shot learning","authors":"Jing Zhang,&nbsp;Yunzuo Hu,&nbsp;Xinzhou Zhang,&nbsp;Mingzhe Chen,&nbsp;Zhe Wang","doi":"10.1016/j.neunet.2025.107482","DOIUrl":"10.1016/j.neunet.2025.107482","url":null,"abstract":"<div><div>Current few-shot image classification methods encounter challenges in extracting multi-view features that can complement each other and selecting optimal features for classification in a specific task. To address this problem, we propose a novel Task-adapted Multi-view feature Calibration Network (TMCN) inspired by the different saccade patterns observed in the human visual system. The TMCN is designed to “saccade” for extracting complementary multi-view features and “purify” multi-view features in a task-adapted manner. To capture more representative features, we propose a multi-view feature extraction method that simulates the voluntary saccades and scanning saccades in the human visual system, which generates global, local grid, and randomly sampled multi-view features. To purify and obtain the most appropriate features, we employ a global local feature calibration module to calibrate global and local grid features for achieving more stable non-local image features. Furthermore, a sampling feature fusion method is proposed to fuse the randomly sampled features from classes to obtain better prototypes, and a multi-view feature calibrating module is proposed to adaptively fuse purified multi-view features based on the task information obtained from the task feature extracting module. Extensive experiments conducted on three widely used public datasets prove that our proposed TMCN can achieve excellent performance and surpass state-of-the-art methods. The code is available at the following address: <span><span>https://github.com/huyunzuo/TMCN</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107482"},"PeriodicalIF":6.0,"publicationDate":"2025-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143880919","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Enhancing motor imagery EEG classification with a Riemannian geometry-based spatial filtering (RSF) method","authors":"Lincong Pan ,&nbsp;Kun Wang ,&nbsp;Yongzhi Huang ,&nbsp;Xinwei Sun ,&nbsp;Jiayuan Meng ,&nbsp;Weibo Yi ,&nbsp;Minpeng Xu ,&nbsp;Tzyy-Ping Jung ,&nbsp;Dong Ming","doi":"10.1016/j.neunet.2025.107511","DOIUrl":"10.1016/j.neunet.2025.107511","url":null,"abstract":"<div><div>Motor imagery (MI) refers to the mental simulation of movements without physical execution, and it can be captured using electroencephalography (EEG). This area has garnered significant research interest due to its substantial potential in brain-computer interface (BCI) applications, especially for individuals with physical disabilities. However, accurate classification of MI EEG signals remains a major challenge due to their non-stationary nature, low signal-to-noise ratio, and sensitivity to both external and physiological noise. Traditional classification methods, such as common spatial pattern (CSP), often assume that the data is stationary and Gaussian, which limits their applicability in real-world scenarios where these assumptions do not hold. These challenges highlight the need for more robust methods to improve classification accuracy in MI-BCI systems. To address these issues, this study introduces a Riemannian geometry-based spatial filtering (RSF) method that projects EEG signals into a lower-dimensional subspace, maximizing the Riemannian distance between covariance matrices from different classes. By leveraging the inherent geometric properties of EEG data, RSF enhances the discriminative power of the features while maintaining robustness against noise. The performance of RSF was evaluated in combination with ten commonly used MI decoding algorithms, including CSP with linear discriminant analysis (CSP-LDA), Filter Bank CSP (FBCSP), Minimum Distance to Riemannian Mean (MDM), Tangent Space Mapping (TSM), EEGNet, ShallowConvNet (sCNN), DeepConvNet (dCNN), FBCNet, Graph-CSPNet, and LMDA-Net, using six publicly available MI-BCI datasets. The results demonstrate that RSF significantly improves classification accuracy and reduces computational time, particularly for deep learning models with high computational complexity. These findings underscore the potential of RSF as an effective spatial filtering approach for MI EEG classification, providing new insights and opportunities for the development of robust MI-BCI systems. The code for this research is available at <span><span>https://github.com/PLC-TJU/RSF</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107511"},"PeriodicalIF":6.0,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143877170","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Time series compression using quaternion valued neural networks and quaternion backpropagation","authors":"Johannes Pöppelbaum,&nbsp;Andreas Schwung","doi":"10.1016/j.neunet.2025.107465","DOIUrl":"10.1016/j.neunet.2025.107465","url":null,"abstract":"<div><div>We propose a novel quaternionic time series compression methodology where we divide a long time series into segments of data, extract the min, max, mean and standard deviation of these chunks as representative features and encapsulate them in a quaternion, yielding a quaternion valued time series. This time series is processed using quaternion valued neural network layers, where we aim to preserve the relation between these features through the usage of the Hamilton product. To train this quaternion neural network, we derive quaternion backpropagation employing the GHR calculus, which is required for a valid product and chain rule in quaternion space. Furthermore, we investigate the connection between the derived update rules and automatic differentiation.</div><div>We apply our proposed compression method on the Tennessee Eastman Dataset, where we perform fault classification using the compressed data in two settings: a fully supervised one and in a semi supervised, contrastive learning setting. Both times, we were able to outperform real valued counterparts as well as two baseline models: one with the uncompressed time series as the input and the other with a regular downsampling using the mean. Further, we could improve the classification benchmark set by SimCLR-TS from 81.43% to 83.90%.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107465"},"PeriodicalIF":6.0,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143870676","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Symmetry discovery for different data types","authors":"Lexiang Hu ,&nbsp;Yikang Li ,&nbsp;Zhouchen Lin","doi":"10.1016/j.neunet.2025.107481","DOIUrl":"10.1016/j.neunet.2025.107481","url":null,"abstract":"<div><div>Equivariant neural networks incorporate symmetries into their architecture, achieving higher generalization performance. However, constructing equivariant neural networks typically requires prior knowledge of data types and symmetries, which is difficult to achieve in most tasks. In this paper, we propose LieSD, a method for discovering symmetries via trained neural networks which approximate the input–output mappings of the tasks. It characterizes equivariance and invariance (a special case of equivariance) of continuous groups using Lie algebra and directly solves the Lie algebra space through the inputs, outputs, and gradients of the trained neural network. Then, we extend the method to make it applicable to multi-channel data and tensor data, respectively. We validate the performance of LieSD on tasks with symmetries such as the two-body problem, the moment of inertia matrix prediction, top quark tagging, and rotated MNIST. Compared with the baseline, LieSD can accurately determine the number of Lie algebra bases without the need for expensive group sampling. Furthermore, LieSD can perform well on non-uniform datasets, whereas methods based on GANs fail. Code and data are available at <span><span>https://github.com/hulx2002/LieSD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107481"},"PeriodicalIF":6.0,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143870677","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"On robust learning of memory attractors with noisy deep associative memory networks","authors":"Xuan Rao ,&nbsp;Bo Zhao ,&nbsp;Derong Liu","doi":"10.1016/j.neunet.2025.107474","DOIUrl":"10.1016/j.neunet.2025.107474","url":null,"abstract":"<div><div>Developing the computational mechanism for memory systems is a long-standing focus in machine learning and neuroscience. Recent studies have shown that overparameterized autoencoders (OAEs) implement associative memory (AM) by encoding training data as attractors. However, the learning of memory attractors requires that the norms of all eigenvalues of the input–output Jacobian matrix are strictly less than one. Motivated by the observed strong negative correlation between the attractor robustness and the largest singular value of the Jacobian matrix, we develop the noisy overparameterized autoencoders (NOAEs) for learning robust attractors by injecting random noises into their inputs during the training procedure. Theoretical demonstrations show that the training objective of the NOAE approximately minimizes the upper bound of the weighted sum of the reconstruction error and the square of the largest singular value. Extensive experiments in terms of numerical and image-based datasets show that NOAEs not only increase the success rate of the training samples becoming attractors, but also improve the attractor robustness. Codes are available at <span><span>https://github.com/RaoXuan-1998/neural-netowrk-journal-NOAE</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107474"},"PeriodicalIF":6.0,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143880917","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Fine-tune language models as multi-modal differential equation solvers","authors":"Liu Yang,&nbsp;Siting Liu,&nbsp;Stanley J. Osher","doi":"10.1016/j.neunet.2025.107455","DOIUrl":"10.1016/j.neunet.2025.107455","url":null,"abstract":"<div><div>In the growing domain of scientific machine learning, in-context operator learning has shown notable potential in building foundation models, as in this framework the model is trained to learn operators and solve differential equations using prompted data, during the inference stage without weight updates. However, the current model’s overdependence on function data overlooks the invaluable human insight into the operator. To address this, we present a transformation of in-context operator learning into a multi-modal paradigm. In particular, we take inspiration from the recent success of large language models, and propose using “captions” to integrate human knowledge about the operator, expressed through natural language descriptions and equations. Also, we introduce a novel approach to train a language-model-like architecture, or directly fine-tune existing language models, for in-context operator learning. We beat the baseline on single-modal learning tasks, and also demonstrated the effectiveness of multi-modal learning in enhancing performance and reducing function data requirements. The proposed method not only significantly enhanced the development of the in-context operator learning paradigm, but also created a new path for the application of language models.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107455"},"PeriodicalIF":6.0,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143873154","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"SpikeCLIP: A contrastive language–image pretrained spiking neural network","authors":"Changze Lv ,&nbsp;Tianlong Li ,&nbsp;Wenhao Liu ,&nbsp;Yufei Gu ,&nbsp;Jianhan Xu ,&nbsp;Cenyuan Zhang ,&nbsp;Muling Wu ,&nbsp;Xiaoqing Zheng ,&nbsp;Xuanjing Huang","doi":"10.1016/j.neunet.2025.107475","DOIUrl":"10.1016/j.neunet.2025.107475","url":null,"abstract":"<div><div>Spiking Neural Networks (SNNs) have emerged as a promising alternative to conventional Artificial Neural Networks (ANNs), demonstrating comparable performance in both visual and linguistic tasks while offering the advantage of improved energy efficiency. Despite these advancements, the integration of linguistic and visual features into a unified representation through spike trains poses a significant challenge, and the application of SNNs to multimodal scenarios remains largely unexplored. This paper presents SpikeCLIP, a novel framework designed to bridge the modality gap in spike-based computation. Our approach employs a two-step recipe: an “alignment pre-training” to align features across modalities, followed by a “dual-loss fine-tuning” to refine the model’s performance. Extensive experiments reveal that SNNs achieve results on par with ANNs while substantially reducing energy consumption across various datasets commonly used for multimodal model evaluation. Furthermore, SpikeCLIP maintains robust image classification capabilities, even when dealing with classes that fall outside predefined categories. This study marks a significant advancement in the development of energy-efficient and biologically plausible multimodal learning systems.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107475"},"PeriodicalIF":6.0,"publicationDate":"2025-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143870673","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"DGPrompt: Dual-guidance prompts generation for vision-language models","authors":"Tai Zheng,&nbsp;Zhen-Duo Chen,&nbsp;Zi-Chao Zhang,&nbsp;Zhen-Xiang Ma,&nbsp;Li-Jun Zhao,&nbsp;Chong-Yu Zhang,&nbsp;Xin Luo,&nbsp;Xin-Shun Xu","doi":"10.1016/j.neunet.2025.107472","DOIUrl":"10.1016/j.neunet.2025.107472","url":null,"abstract":"<div><div>Introducing learnable prompts into CLIP and fine-tuning them have demonstrated excellent performance across many downstream tasks. However, existing methods have insufficient interaction between modalities and neglect the importance of hierarchical contextual information, leading to ineffective alignment in both the visual and textual representation spaces. Additionally, CLIP is highly sensitive to prompts, making learnable prompts prone to overfitting on seen classes, which results in the forgetting of general knowledge of CLIP and severely impair generalization ability on unseen classes. To address these issues, we propose an original <span><math><mi>D</mi></math></span>ual-<span><math><mi>G</mi></math></span>uidance <span><math><mi>Prompt</mi></math></span>s Generation (<span><math><mi>DGPrompt</mi></math></span>) method that promotes alignment between visual and textual spaces while ensuring the continuous retention of general knowledge. The main ideas of DGPrompt are as follows: 1) The extraction of image and text embeddings are guided mutually by generating visual and textual prompts, making full use of complementary information from both modalities to align visual and textual spaces. 2) The prompt-tuning process is restrained by a retention module, reducing the forgetting of general knowledge. Extensive experiments conducted in settings of base-to-new class generalization and few-shot learning demonstrate the superiority of the proposed method. Compared with the baseline method CLIP and the state-of-the-art method MaPLe, DGPrompt exhibits favorable performance and achieves an absolute gain of 7.84% and 0.99% on overall harmonic mean, averaged over 11 diverse image recognition datasets.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107472"},"PeriodicalIF":6.0,"publicationDate":"2025-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143870675","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Multiscroll hidden attractor in memristive autapse neuron model and its memristor-based scroll control and application in image encryption","authors":"Zhiqiang Wan ,&nbsp;Yi-Fei Pu ,&nbsp;Qiang Lai","doi":"10.1016/j.neunet.2025.107473","DOIUrl":"10.1016/j.neunet.2025.107473","url":null,"abstract":"<div><div>In current neurodynamic studies, memristor models using polynomial or multiple nested composite functions are primarily employed to generate multiscroll attractors, but their complex mathematical form restricts both research and application. To address this issue, without relying on polynomial and multiple nested composite functions, this study devises a unique memristor model and a memristive autapse HR (MAHR) neuron model featuring multiscroll hidden attractor. Specially, the quantity of scrolls within the multiscroll hidden attractors is regulated by simulation time. Besides, a simple control factor is incorporated into the memristor to improve the MAHR neuron model. Numerical analysis further finds that the quantity of scrolls within the multiscroll hidden attractor from the improved MAHR neuron model can be conveniently adjusted by only changing a single parameter or initial condition of the memristor. Moreover, a microcontroller-based hardware experiment is conducted to confirm that the improved MAHR neuron model is physically feasible. Finally, an elegant image encryption scheme is proposed to explore the real-world applicability of the improved MAHR neuron model.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107473"},"PeriodicalIF":6.0,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143854911","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}