Chuxiong Sun , Jie Hu , Hongming Gu , Jinpeng Chen , Wei Liang , Mingchuan Yang
{"title":"Scalable and Adaptive Graph Neural Networks with Self-Label-Enhanced Training","authors":"Chuxiong Sun , Jie Hu , Hongming Gu , Jinpeng Chen , Wei Liang , Mingchuan Yang","doi":"10.1016/j.patcog.2024.111210","DOIUrl":"10.1016/j.patcog.2024.111210","url":null,"abstract":"<div><div>Although GNNs have achieved success in semi-supervised graph learning tasks, common GNNs suffer from expensive message passing during each epoch and the exponentially growing receptive field occupying too much memory, especially on large graphs. Neighbor sampling techniques can reduce GNNs’ memory footprints, but they encounter either redundant computation or incomplete edges. Some simplified GNNs decouple graph convolutions and feature transformations to reduce computation in training. However, only a part of them can scale to large graphs without neighbor sampling techniques, which can be concluded as decoupled GNNs. Nevertheless, they either only utilize the last convolution output or simply add multi-hop features with uniform weights, which limits their expressiveness. In this paper, we refine the pipeline of decoupled GNNs and propose Scalable and Adaptive Graph Neural Networks (SAGN), which effectively leverages multi-hop information with a scalable attention mechanism. Moreover, we generalize the input of decoupled GNNs to view another classical technique, label propagation, as a special case of decoupled GNNs and propose decoupled label trick (DecLT) to incorporate label information into decoupled GNNs. Furthermore, by incorporating self-training technique, we further propose the Self-Label-Enhanced (SLE) training framework, leveraging pseudo labels to simultaneously augment the training set and improve label propagation. Extensive experiments show that SAGN outperforms other baselines, and that DecLT and SLE can consistently and significantly improve all types of models on semi-supervised node classification tasks. Many top-ranked models on Open Graph Benchmark (OGB) leaderboard adopt our methods as the main backbone.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"160 ","pages":"Article 111210"},"PeriodicalIF":7.5,"publicationDate":"2024-11-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142722839","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}
Zhen Zhang , Lu Yang , Kuikui Wang , Xiaoming Xi , Xiushan Nie , Gongping Yang , Yilong Yin
{"title":"Consistency and label constrained transfer low-rank representation for cross-light finger vein recognition","authors":"Zhen Zhang , Lu Yang , Kuikui Wang , Xiaoming Xi , Xiushan Nie , Gongping Yang , Yilong Yin","doi":"10.1016/j.patcog.2024.111208","DOIUrl":"10.1016/j.patcog.2024.111208","url":null,"abstract":"<div><div>Finger vein sensors are embedded into all kinds of electronic devices for personal identification, and the upgrading of sensors is unavoidable. Therefore, the concern about cross-sensor finger vein recognition is raised recently. However, little attention is paid to cross-sensor finger vein recognition. The imaging light variation is one main difference between different sensors, and it brings large image differences, seriously degrading finger vein recognition performance. This paper focuses on cross-light finger vein recognition problem, in which we assume that the training and testing finger vein images are captured by different near-infrared lights, and proposes a consistency and label constrained transfer low-rank representation (CLTLRR) method for dealing with cross-light finger vein recognition. In the proposed method, we first transfer cross-light finger vein images into a common feature space to narrow the gap between training images and testing images, and achieve the low-rank linear representations of images. Then, we develop a consistency constraint between the low-rank coefficients in the common feature space and the sparse coefficients in the original feature space to enhance the discrimination of linear representation. In addition, we design a class label constraint for the projection matrix to guide image transfer. Finally, the low-rank coefficients and the projected features in the common feature space are integrated for recognition. Experiments are performed on single-light and cross-light finger palmar vein databases and finger dorsal vein databases, and the experimental results prove the effectiveness of our CLTLRR.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"161 ","pages":"Article 111208"},"PeriodicalIF":7.5,"publicationDate":"2024-11-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759371","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}
Miguel de Carvalho , Gabriel Martos , Andrej Svetlošák
{"title":"A game-inspired algorithm for marginal and global clustering","authors":"Miguel de Carvalho , Gabriel Martos , Andrej Svetlošák","doi":"10.1016/j.patcog.2024.111158","DOIUrl":"10.1016/j.patcog.2024.111158","url":null,"abstract":"<div><div>An often overlooked pitfall of model-based clustering is that it typically results in the same number of clusters per margin, an assumption that may not be natural in practice. We develop a clustering method that takes advantage of the sturdiness of model-based clustering, while attempting to mitigate this issue. The proposed approach allows each margin to have a varying number of clusters and employs a strategy game-inspired algorithm, named ‘Reign-and-Conquer’, to cluster the data. Since the proposed clustering approach only specifies a model for the margins, but leaves the joint unspecified, it has the advantage of being partially parallelizable; hence, the proposed approach is computationally appealing as well as more tractable for moderate to high dimensions than a ‘full’ (joint) model-based clustering approach. A battery of numerical experiments on simulated data indicates an overall good performance of the proposed methods in a variety of scenarios, and real datasets are used to showcase their usefulness in practice.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"160 ","pages":"Article 111158"},"PeriodicalIF":7.5,"publicationDate":"2024-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142722940","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":"Reviving undersampling for long-tailed learning","authors":"Hao Yu, Yingxiao Du , Jianxin Wu","doi":"10.1016/j.patcog.2024.111200","DOIUrl":"10.1016/j.patcog.2024.111200","url":null,"abstract":"<div><div>The training datasets used in long-tailed recognition are extremely unbalanced, resulting in significant variation in per-class accuracy across categories. Prior works mostly used average accuracy to evaluate their algorithms, which easily ignores those worst-performing categories. In this paper, we aim to enhance the accuracy of the worst-performing categories and utilize the harmonic mean and geometric mean to assess the model’s performance. We revive the balanced undersampling idea to achieve this goal. In few-shot learning, balanced subsets are few-shot and will surely under-fit, hence it is not used in modern long-tailed learning. But, we find that it produces a more equitable distribution of accuracy across categories with much higher harmonic and geometric mean accuracy, but with lower average accuracy. Moreover, we devise a straightforward model ensemble strategy, which does not result in any additional overhead and achieves improved harmonic and geometric mean while keeping the average accuracy almost intact when compared to state-of-the-art long-tailed learning methods. We validate the effectiveness of our approach on widely utilized benchmark datasets for long-tailed learning. Our code is at <span><span>https://github.com/yuhao318/BTM/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"161 ","pages":"Article 111200"},"PeriodicalIF":7.5,"publicationDate":"2024-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142759373","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}
Jingyu Song , Haiyong Xu , Gangyi Jiang , Mei Yu , Yeyao Chen , Ting Luo , Yang Song
{"title":"Frequency domain-based latent diffusion model for underwater image enhancement","authors":"Jingyu Song , Haiyong Xu , Gangyi Jiang , Mei Yu , Yeyao Chen , Ting Luo , Yang Song","doi":"10.1016/j.patcog.2024.111198","DOIUrl":"10.1016/j.patcog.2024.111198","url":null,"abstract":"<div><div>The degradation of underwater images, due to complex factors, negatively impacts the performance of underwater visual tasks. However, most underwater image enhancement methods (UIE) have been confined to the spatial domain, disregarding the frequency domain. This limitation hampers the full exploitation of the model’s learning and representational capabilities. To address this, a two-stage frequency domain-based latent diffusion model (FD-LDM) is introduced for UIE. Firstly, the model employs a lightweight parameter estimation network (L-PEN) to estimate the degradation parameters of underwater images, thereby mitigating the impact of color bias on the diffusion model. Subsequently, considering the varying degrees of recovery between high and low-frequency images, high and low-frequency priors are extracted in the second stage and integrated with the refined latent diffusion model to enhance the images further. Extensive experiments have confirmed the method’s effectiveness and robustness, particularly under color bias scenes.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"160 ","pages":"Article 111198"},"PeriodicalIF":7.5,"publicationDate":"2024-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142722945","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}
Junbo Yang , Borui Hu , Hanyu Li , Yang Liu , Xinbo Gao , Jungong Han , Fanglin Chen , Xuangou Wu
{"title":"Dynamic VAEs via semantic-aligned matching for continual zero-shot learning","authors":"Junbo Yang , Borui Hu , Hanyu Li , Yang Liu , Xinbo Gao , Jungong Han , Fanglin Chen , Xuangou Wu","doi":"10.1016/j.patcog.2024.111199","DOIUrl":"10.1016/j.patcog.2024.111199","url":null,"abstract":"<div><div>Continual Zero-shot Learning (CZSL) is capable of classifying unseen categories across a sequence of tasks. However, CZSL is often plagued by the challenge of catastrophic forgetting. While recent studies have shown that preserving past data for experience replay can effectively address this issue, it may be limited to specific scenarios and pose a risk of data leakage. Additionally, many existing CZSL models fail to adequately highlight the correlation between semantic and visual features. To tackle these shortcomings, we introduce dynamic Variational Autoencoders (VAEs) via semantic-aligned matching for CZSL. The proposed model utilizes both semantic and visual VAEs to enhance the transfer capability of knowledge from past tasks. Leveraging generative experience replay, our model effectively combats catastrophic forgetting. Our approach was assessed on five datasets: aPY, AWA1, AWA2, CUB, and SUN, yielding superior performance to baseline models.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"160 ","pages":"Article 111199"},"PeriodicalIF":7.5,"publicationDate":"2024-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142722841","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":"Multi-branch feature transformation cross-domain few-shot learning for hyperspectral image classification","authors":"Meilin Shi , Jiansi Ren","doi":"10.1016/j.patcog.2024.111197","DOIUrl":"10.1016/j.patcog.2024.111197","url":null,"abstract":"<div><div>In the field of hyperspectral image (HSI) classification, a source dataset with ample labeled samples is commonly utilized to enhance the classification performance of a target dataset with few labeled samples. Existing few-shot learning (FSL) methods typically assume identical feature distribution in the source and target domains. However, since the classes of samples collected from different regions may vary considerably, it leads to a disparity in the feature distribution. To address the domain distribution shift between the source and target domains, a cross-domain FSL method based on multi-branch feature transformation (MBFT-CFSL) is proposed for HSI classification. First, the spectral–spatial features of the image are extracted by the multi-branch feature fusion module, and the feature diversity is increased using the featurewise transformation layers to boost the generalization performance of the model. Then, the conditional adversarial domain adaptation technique is employed for model training to lessen the impact of domain shift. Finally, the model is optimized by minimizing the maximum mean difference loss function to further diminish the distribution difference between the source and target domains. Experimental results on three distinct hyperspectral datasets validate the effectiveness of MBFT-CFSL, with the overall classification accuracy improved by 1.73%–5.45% compared to the suboptimal method. The source code is available at <span><span>https://github.com/Ziyin2/MBFT-CFSL</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"160 ","pages":"Article 111197"},"PeriodicalIF":7.5,"publicationDate":"2024-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142723042","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}
In-mo Lee , Yoojeung Kim , Taehoon Kim , Hayoung Choi , Seung Yeop Yang , Yunho Kim
{"title":"Recursive reservoir concatenation for salt-and-pepper denoising","authors":"In-mo Lee , Yoojeung Kim , Taehoon Kim , Hayoung Choi , Seung Yeop Yang , Yunho Kim","doi":"10.1016/j.patcog.2024.111196","DOIUrl":"10.1016/j.patcog.2024.111196","url":null,"abstract":"<div><div>We propose a recursive reservoir concatenation architecture in reservoir computing for salt-and-pepper noise removal. The recursive algorithm consists of two components. One is the initial network training for the recursion. Since the standard reservoir computing does not appreciate images as input data, we designed a nonlinear image-specific forward operator that can extract image features from noisy input images, which are to be mapped into a reservoir for training. The other is the recursive reservoir concatenation to further improve the reconstruction quality. Training errors decrease as more reservoirs are concatenated due to the hierarchical structure of the recursive reservoir concatenation. The proposed method outperformed most analytic or machine-learning based denoising models for salt-and-pepper noise with a training cost much lower than other neural network-based models. Reconstruction is completely parallel, in that noise in different pixels can be removed in parallel.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"160 ","pages":"Article 111196"},"PeriodicalIF":7.5,"publicationDate":"2024-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142722838","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}
Kang Wang , Feiyang Zheng , Dayan Guan , Jia Liu , Jing Qin
{"title":"Distilling heterogeneous knowledge with aligned biological entities for histological image classification","authors":"Kang Wang , Feiyang Zheng , Dayan Guan , Jia Liu , Jing Qin","doi":"10.1016/j.patcog.2024.111173","DOIUrl":"10.1016/j.patcog.2024.111173","url":null,"abstract":"<div><div>In the task of classifying histological images, prior works widely leverage Graph neural network (GNN) to aggregate histological knowledge from multi-level biological entities (<em>e.g.,</em> cell and tissue). However, current GNN-based methods suffer from either inadequate entity representation or intolerable computation burden. To the end, we propose a heterogeneous knowledge distillation (HKD) model to capture and amalgamate the spatial-hierarchical feature of multi-level biological entities. We first design multiple message-passing GNNs with different hidden layers as the teachers for extracting adjacent regions of cells, and leverage a transformer-based GNN as the student to model the global interaction of tissues. Such multi-teacher student architecture enables our HKD to simultaneously obtain topological knowledge at different scales from heterogeneous biological entities. We further propose a biological affiliation recognition module to adaptively align the cell knowledge learned from multi-teacher models with cell-corresponding tissue in the student model, encouraging the student model to attentively amalgamate the semantics of multi-level biological entities for highly accurate classification. Extensive experiments show that our method outperforms the state-of-the-art on three public datasets of histological image classification.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"160 ","pages":"Article 111173"},"PeriodicalIF":7.5,"publicationDate":"2024-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142722011","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}
Zhengshan Wang , Long Chen , Juan He , Linyao Yang , Fei-Yue Wang
{"title":"Exploring Latent Transferability of feature components","authors":"Zhengshan Wang , Long Chen , Juan He , Linyao Yang , Fei-Yue Wang","doi":"10.1016/j.patcog.2024.111184","DOIUrl":"10.1016/j.patcog.2024.111184","url":null,"abstract":"<div><div>Feature disentanglement techniques have been widely employed to extract transferable (domain-invariant) features from non-transferable (domain-specific) features in Unsupervised Domain Adaptation (UDA). However, due to the complex interplay among high-dimensional features, the separated “non-transferable” features may still be partially informative. Suppressing or disregarding them, as commonly employed in previous methods, can overlook the inherent transferability. In this work, we introduce two concepts: Partially Transferable Class Features and Partially Transferable Domain Features (PTCF and PTDF), and propose a succinct feature disentanglement technique. Different with prior works, we do not seek to thoroughly peel off the non-transferable features, as it is challenging practically. Instead, we take the two-stage strategy consisting of rough feature disentanglement and dynamic adjustment. We name our model as ELT because it can systematically Explore Latent Transferability of feature components. ELT can automatically evaluate the transferability of internal feature components, dynamically giving more attention to features with high transferability and less to features with low transferability, effectively solving the problem of negative transfer. Extensive experimental results have proved its efficiency. The code and supplementary file will be available at <span><span>https://github.com/njtjmc/ELT</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"160 ","pages":"Article 111184"},"PeriodicalIF":7.5,"publicationDate":"2024-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142743130","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}