Deep learning models are known to often learn features that spuriously correlate with the class label during training but are irrelevant to the prediction task. Existing methods typically address this issue by annotating potential spurious attributes, or filtering spurious features based on some empirical assumptions (e.g., simplicity of bias). However, these methods may yield unsatisfactory performance due to the intricate and elusive nature of spurious correlations in real-world data. In this paper, we propose a data-oriented approach1to mitigate the spurious correlation in deep learning models. We observe that samples that are influenced by spurious features tend to exhibit a dispersed distribution in the learned feature space. This allows us to identify the presence of spurious features. Subsequently, we obtain a bias-invariant representation by neutralizing the spurious features based on a simple grouping strategy. Then, we learn a feature transformation to eliminate the spurious features by aligning with this bias-invariant representation. Finally, we update the classifier by incorporating the learned feature transformation and obtain an unbiased model. By integrating the aforementioned identifying, neutralizing, eliminating and updating procedures, we build an effective pipeline for mitigating spurious correlation. Experiments on image and NLP debiasing benchmarks show an improvement in worst group accuracy of more than 20% compared to standard empirical risk minimization (ERM).

Cropland Change Detection (CCD) is a critical technological means for ensuring food security and advancing agricultural sustainability, holding significant application value in remote sensing image interpretation. However, prevalent domain shifts between bi-temporal remote sensing images—caused by land cover phenological variations (e.g., crop growth cycles) and multi-source imaging disparities (e.g., sensor differences, illumination changes, atmospheric interference)—severely degrade the detection accuracy and robustness of conventional methods. To address these challenges, this paper proposes a novel end-to-end domain adaptation framework named MDANet for cropland change detection, which innovatively integrates the long-sequence modeling capability of Mamba with domain-invariant feature learning strategies. First, a Mamba-based domain adaptation method is constructed to conduct cross-domain alignment of bi-temporal images, aiming to eliminate domain shifts caused by imaging and geographical differences. Specifically, a source-target weight adjustment mechanism is designed to adaptively balance the migration degree of image source and target features, and two identity losses are proposed to preserve the geospatial structure and semantic consistency of croplands. Then, a cross-domain feature alignment and fusion mechanism is designed to align transformed features via domain-invariant learning, effectively suppressing pseudo-changes and enhancing discriminability in genuine change regions. Experimental results on JLYHCD, CLCD, and Hi-CNA datasets demonstrate that the proposed method achieves F1-scores of 79.87%, 81.94%, and 87.24%, respectively, outperforming existing advanced change detection approaches and exhibiting superior detection accuracy and cross-domain generalization capability. All codes are available at https://github.com/dawu/MDANet.

Cross-domain adaptive segmentation is a practical solution for the scenario that lacks expensive annotations or is inaccessible to ground truth. Prior works have tried to improve cross-domain adaptive segmentation with domain alignment, but most of them ignore the problem of training target deviation of distance-regularizing based domain alignment method. To address this, we propose a novel domain alignment mechanism that unifies the two optimization objectives, domain alignment, and segmentation performance, into one. In addition, existing methods are hard to apply under the source-free setting. We introduce a novel domain adaptive segmentation framework suitable for vanilla Unsupervised Domain Adaptation (UDA) and source-free UDA settings. Experiments show the proposed method outperforms competitive works with much more complicated mechanisms and achieves the state-of-the-art performance on both GTA→Cityscapes and Synthia→Cityscapes benchmarks. Our work can be easily added to existing methods and boost their performance.

Real-world systems from a variety of domains, ranging from physics to medicine, can naturally be modelled as dynamic networks. Dynamic community detection is regarded as a fundamental tool to probe into the mechanisms of networks. Here, we describe a framework for tracking the network evolution over time, where each community is characterized by a series of transition events, which is one of the most influential evolutionary patterns in dynamic networks. The framework is used to motivate a temporal smoothness strategy for efficiently identifying dynamic communities and exploring the transition behavior of networks from community-level and node-level. Evaluations on two synthetic and real-world datasets containing embedded transition events demonstrate that the framework can successfully discover dynamic communities and analyze the transition behavior of networks.

In this paper, we propose a novel Convolutional Neural Network (CNN) based method that extracts the location information (displacement features) of the maximums in the max-pooling operation and fuses it with the pooling features to capture the micro deformations between the genuine signatures and skilled forgeries as a feature extraction procedure. After the feature extraction procedure, we apply support vector machines (SVMs) as writer-dependent classifiers for each user to build the signature verification system. The extensive experimental results on GPDS-150, GPDS-300, GPDS-1000, GPDS-2000, and GPDS-5000 datasets demonstrate that the proposed method can discriminate the genuine signatures and their corresponding skilled forgeries well and achieve state-of-the-art results on these datasets.

Signature verification systems suffer from imbalanced learning, which imposes strict requirements on classifiers. The standard classification approaches, such as SVM, often degrade the performance for imbalanced data or require additional parameters for data balancing. In this study, as a new approach for signature verification, we use RankSVM as the writer-dependent classifiers, which theoretically guarantees the generalization performance for imbalanced data. To investigate the ability of RankSVM for solving imbalanced learning problems in signature verification tasks, the extensive experiments are conducted on bitmaps of GPDS-150, GPDS-300, GPDS-600, and GPDS-1000 datasets and deep features of GPDS-960 dataset. The experimental results demonstrate that the RankSVM-based approach obtains a nearly equivalent performance with the state-of-the-art method on deep features of the GPDS-960 dataset, and achieves significantly better performance than standard-SVM-based approach on bitmaps of GPDS-150, GPDS-300, GPDS-600, and GPDS-1000 datasets.

The traditional max-pooling operation in Convolutional Neural Networks (CNNs) only obtains the maximal value from a pooling window. However, it discards the information about the precise position of the maximal value. In this paper, we extract the location of the maximal value in a pooling window and transform it into "displacement feature". We analyze and discover the class-wise trend of the displacement features in many ways. The experimental results and discussion demonstrate that the displacement features have beneficial behaviors for solving the problems in max-pooling.

In recent years, many deep architectures have been proposed for handwritten text recognition. However, most of the previous deep models need large scale training data and a long training time to obtain good results. In this paper, we propose a novel deep learning method based on “stretching” the projection matrices of stacked feature learning models. We call the proposed method “stretching deep architectures” (or SDA). In the implementation of SDA, stacked feature learning models are first learned layer by layer, and then the stretching technique is applied on the weight matrices between successive layers. As the feature learning models can be efficiently optimized and the stretching results can be easily computed, the training of SDA is very fast and no back propagation is needed. We have tested SDA on handwritten digits recognition, Arabic subword recognition and English letter recognition tasks. Extensive experiments demonstrate that SDA performs not only better than shallow feature learning models, but also state-of-the-art deep learning models.