Search

Federated Learning (FL) is a newly emerged federated optimization technique for distributed data in a federated network. The participants in FL that train the model locally are classified into client nodes. The server node assumes the responsibility to aggregate local models from client nodes without data moving. In this regard, FL is an ideal solution to protect data privacy at each node of the network. However, the raw data generated on each node are unlabeled, making it impossible for FL to apply these data directly to train a model. The large volume of data annotating work prevents FL from being widely applied in the real world, especially for online scenarios, where the data are generated continuously. Meanwhile, the data generated on different nodes tend to be differently distributed. It has been proved theoretically and experimentally that non-independent and identically distributed (non-IID) data harm the performance of FL. In this article, we design a semi-federated active learning (semi-FAL) framework to tackle the annotation and non-IID problems jointly. More specifically, the server node can provide (i) a pre-trained model to help each client node annotate the local data uniformly and (ii) an estimation of the global gradient to help correct the local gradient. The evaluation results demonstrate our semi-FAL framework can efficiently handle unlabeled online network data and achieves high accuracy and fast convergence.

Machine learning (ML) has transformed neuroimaging research by enabling accurate predictions and feature extraction from large datasets. In this study, we investigate the application of six ML algorithms (Lasso, relevance vector regression, support vector regression, extreme gradient boosting, category boost, and multilayer perceptron) to predict brain age for middle-aged and older adults, which is a crucial area of research in neuroimaging. Despite the plethora of proposed ML models, there is no clear consensus on how to achieve better performance in brain age prediction for this population. Our study stands out by evaluating the impact of both ML algorithms and image modalities on brain age prediction performance using a large cohort of cognitively normal adults aged 44.6 to 82.3 years old (N = 27,842) with six image modalities. We found that the predictive performance of brain age is more reliant on the image modalities used than the ML algorithms employed. Specifically, our study highlights the superior performance of T1-weighted MRI and diffusion-weighted imaging and demonstrates that multi-modality-based brain age prediction significantly enhances performance compared to unimodality. Moreover, we identified Lasso as the most accurate ML algorithm for predicting brain age, achieving the lowest mean absolute error in both single-modality and multi-modality predictions. Additionally, Lasso also ranked highest in a comprehensive evaluation of the relationship between BrainAGE and the five frequently mentioned BrainAGE-related factors. Notably, our study also shows that ensemble learning outperforms Lasso when computational efficiency is not a concern. Overall, our study provides valuable insights into the development of accurate and reliable brain age prediction models for middle-aged and older adults, with significant implications for clinical practice and neuroimaging research. Our findings highlight the importance of image modality selection and emphasize Lasso as a promising ML algorithm for brain age prediction.

The neuroscience community has developed many convolutional neural networks (CNNs) for the early detection of Alzheimer’s disease (AD). Population graphs are thought of as non-linear structures that capture the relationships between individual subjects represented as nodes, which allows for the simultaneous integration of imaging and non-imaging information as well as individual subjects’ features. Graph convolutional networks (GCNs) generalize convolution operations to accommodate non-Euclidean data and aid in the mining of topological information from the population graph for a disease classification task. However, few studies have examined how GCNs’ input properties affect AD-staging performance. Therefore, we conducted three experiments in this work. Experiment 1 examined how the inclusion of demographic information in the edge-assigning function affects the classification of AD versus cognitive normal (CN). Experiment 2 was designed to examine the effects of adding various neuropsychological tests to the edge-assigning function on the mild cognitive impairment (MCI) classification. Experiment 3 studied the impact of the edge assignment function. The best result was obtained in Experiment 2 on multi-class classification (AD, MCI, and CN). We applied a novel framework for the diagnosis of AD that integrated CNNs and GCNs into a unified network, taking advantage of the excellent feature extraction capabilities of CNNs and population-graph processing capabilities of GCNs. To learn high-level anatomical features, DenseNet was used; a set of population graphs was represented with nodes defined by imaging features and edge weights determined by different combinations of imaging or/and non-imaging information, and the generated graphs were then fed to the GCNs for classification. Both binary classification and multi-class classification showed improved performance, with an accuracy of 91.6% for AD versus CN, 91.2% for AD versus MCI, 96.8% for MCI versus CN, and 89.4% for multi-class classification. The population graph’s imaging features and edge-assigning functions can both significantly affect classification accuracy.

Many previous works only focused on the cascading failure of global coupling of one-to-one structures in interdependent networks, but the local coupling of dual coupling structures has rarely been studied due to its complex structure. This will result in a serious consequence that many conclusions of the one-to-one structure may be incorrect in the dual coupling network and do not apply to the smart grid. Therefore, it is very necessary to subdivide the dual coupling link into a top-down coupling link and a bottom-up coupling link in order to study their influence on network robustness by combining with different coupling modes. Additionally, the power flow of the power grid can cause the load of a failed node to be allocated to its neighboring nodes and trigger a new round of load distribution when the load of these nodes exceeds their capacity. This means that the robustness of smart grids may be affected by four factors, i.e., load redistribution, local coupling, dual coupling link and coupling mode; however, the research on the influence of those factors on the network robustness is missing. In this paper, firstly, we construct the smart grid as a two-layer network with a dual coupling link and divide the power grid and communication network into many subnets based on the geographical location of their nodes. Secondly, we define node importance ( N I ) as an evaluation index to access the impact of nodes on the cyber or physical network and propose three types of coupling modes based on N I of nodes in the cyber and physical subnets, i.e., Assortative Coupling in Subnets (ACIS), Disassortative Coupling in Subnets (DCIS), and Random Coupling in Subnets (RCIS). Thirdly, a cascading failure model is proposed for studying the effect of local coupling of dual coupling link in combination with ACIS, DCIS, and RCIS on the robustness of the smart grid against a targeted attack, and the survival rate of functional nodes is used to assess the robustness of the smart grid. Finally, we use the IEEE 118-Bus System and the Italian High-Voltage Electrical Transmission Network to verify our model and obtain the same conclusions: (I) DCIS applied to the top-down coupling link is better able to enhance the robustness of the smart grid against a targeted attack than RCIS or ACIS, (II) ACIS applied to a bottom-up coupling link is better able to enhance the robustness of the smart grid against a targeted attack than RCIS or DCIS, and (III) the robustness of the smart grid can be improved by increasing the tolerance α . This paper provides some guidelines for slowing down the speed of the cascading failures in the design of architecture and optimization of interdependent networks, such as a top-down link with DCIS, a bottom-up link with ACIS, and an increased tolerance α .