Federated multilabel learning enables the collaborative training of multilabel classification models while preserving client privacy. Existing federated multilabel learning methods either fail to effectively capture label correlations, which significantly affects model performance in scenarios where labels are interdependent, or increase the risk of client privacy leakage due to the transmission of unnecessary client data. To this end, we propose FML-DGCN, a federated multi-label learning approach based on dynamic graph convolutional networks. In the local training phase, a dynamic graph convolutional module is designed and employed to generate label representations specific to the input image, with parameters efficiently optimized by our designed correlation alignment loss. Within the module, we employ a fully connected layer-based network to merge static label embeddings and image features for computing dynamic label graphs, instead of leveraging complex attention-based networks, making it suitable for federated learning environments with limited computing resources. In the federated aggregation phase, clients transmit only model parameters, without sharing any additional information such as scene knowledge, thus lowering the risk of client privacy leakage. Experimental results on four typical multilabel image classification datasets demonstrate the superiority of our approach.

Significant progress has been made in the study of Alzheimer’s disease (AD) and Mild Cognitive Impairment(MCI) progression using multimodal approaches. Recent studies have identified peripheral blood gene expression data as valuable biomarkers for distinguishing AD and MCI progression subtypes. However, these studies either rely entirely on prior data for gene selection or are purely data-driven. These strategies are not conducive to discovering potentially important genes or may lead to results unrelated to brain neural function pathways. This study adopts a data-driven approach based on prior knowledge, using genes mapped onto morphologically different brain regions as features for selection. These features are then input into our Generative Adversarial Network framework to obtain attention masks for cortical morphological indicators, which are weighted and used in training the structural MRI feature extraction main network. Our method ensures that the selected genes are correlated with brain regions and group differences, and through post-hoc interpretability analysis, we identify potential biomarkers in both genes and brain regions across two modalities.

As Android apps become increasingly prevalent in daily life, a common issue in the development process is the configuration of UI colors, leading to color-related accessibility issues that make the text or non-text on the app’s UI difficult to see due to low color contrast. Such color-related accessibility issues are among the top issues in apps, having a negative impact on vision and user experience. However, state-of-the-art approaches are based on predefined rules and lack an understanding of strategies for alternative colors, therefore failing to generate patches acceptable to both app users and developers. To address this research gap, we first conducted an empirical study to explore common strategies used by app developers when fixing real-world color-related accessibility issues. Based on these findings, we proposed DroidPalette, an automated approach for repairing color-related accessibility issues in Android apps. DroidPalette encodes the common strategies used by app developers for selecting issue-fixing colors, as identified in our empirical study, and combines this with the candidate issue-fixing attributes identified from the Android framework and third-party libraries to generate patches. We evaluated DroidPalette on 497 color-related accessibility issues across 105 real-world Android apps, achieving a success rate of 66.60%. Encouragingly, out of 13 patches submitted to GitHub repositories, 8 have received positive feedback from app developers.

Federated multi-label learning can collaboratively train multi-label classification models without compromising user privacy. Compared to multi-class learning, one of the most critical issues of multi-label learning is how to capture the correlations between labels, which is often ignored by existing research on federated multi-label learning. In this paper, a scene-based federated multi-label learning framework is proposed, which effectively utilizes the dependencies among labels for model training on the client-side and aggregates diverse client information on the server-side. Specifically, in the local training phase, a scene recognition module is employed to detect the scene for each image and the corresponding label co-occurrence matrix is used to guide the propagation of image features on the label graph. In the aggregation phase, a scene-aware aggregation method is adopted to enrich the scene-label co-occurrence information of each client. Experiments on PASCAL VOC 2007 and MS-COCO show that our proposed method can significantly improve the accuracy of federated multi-label image classification.

Visual odometry aims to track the incremental motion of an object using the information captured by visual sensors. In this work, we study the point cloud odometry problem, where only the point cloud scans obtained by the LiDAR (Light Detection And Ranging) are used to estimate object's motion trajectory. A lightweight point cloud odometry solution is proposed and named the green point cloud odometry (GreenPCO) method. GreenPCO is an unsupervised learning method that predicts object motion by matching features of consecutive point cloud scans. It consists of three steps. First, a geometry-aware point sampling scheme is used to select discriminant points from the large point cloud. Second, the view is partitioned into four regions surrounding the object, and the PointHop++ method is used to extract point features. Third, point correspondences are established to estimate object motion between two consecutive scans. Experiments on the KITTI dataset are conducted to demonstrate the effectiveness of the GreenPCO method. It is observed that GreenPCO outperforms benchmarking deep learning methods in accuracy while it has a significantly smaller model size and less training time.