In medical image classification, data privacy constraints and the high cost of expert annotations pose significant challenges to building generalizable models. Federated semi-supervised learning (FSSL), which combines the privacy-preserving nature of federated learning with the label efficiency of semi-supervised learning, offers a promising direction. However, in real-world deployments, client data often exhibits highly non-independent and identically distributed (Non-IID) characteristics. This distributional heterogeneity undermines the reliability of pseudo-labels generated by global models, ultimately limiting model generalization. A key limitation of existing FSSL approaches lies in their reliance on a static labeled set fixed prior to training. Such strategies lack the ability to adaptively correct pseudo-label noise or address class imbalance throughout training, particularly under Non-IID settings. To address this, we propose FSSAL, a novel framework that introduces an active learning component into the FSSL pipeline. By continuously identifying informative and representative samples during training, our method adaptively refines the labeled set and enhances the model’s robustness to distribution shifts. FSSAL employs client-private models for pseudo-label generation to reduce global bias, applies a class-aware dynamic thresholding mechanism to ensure more reliable and balanced label selection, and incorporates a sample selection strategy guided by both feature diversity and model uncertainty. Extensive experiments on four public medical image classification datasets demonstrate that FSSAL consistently outperforms competitive FSSL methods in accuracy and F1-score, especially under highly Non-IID conditions, highlighting its robustness and practical potential.

Graph federated learning (GFL) is increasingly utilized in domains such as social network analysis and recommendation systems, where non-IID data exist extensively and necessitate a strong emphasis on personalized learning. However, existing methods focus only on the personality among different clients instead of the personality within a client which widely exists in the real social networks, where intra-client personality addresses the heterogeneity of known data, while inter-client personality always tackle client heterogeneity under privacy constraint. In this paper, we propose a novel automatic personalized graph federated learning (PGFL) scheme named FedCCH to capture both inter-client and intra-client heterogeneity. For intra-client heterogeneity, we innovatively propose the learnable Personalized Factor (PF) to automatically normalize each graph representation within clients by learnable parameters, which weakens the impact of non-IID data distribution. For inter-client heterogeneity, we propose a novel hash-based similarity clustering method to generate the hash signature for each client, and then group similar clients for joint training among different clients. Ultimately, we collaboratively train intra-client and inter-client modules to improve the effectiveness of capturing the heterogeneity of the graph data of clients. Experiment results demonstrate that FedCCH outperforms other state-of-the-art baseline methods.

Abstract Federated learning (FL) is a type of distributed machine learning that enables multiple participants to collaboratively build machine learning models without transferring data outside their local devices, thereby ensuring data privacy and security. However, free-riding (FR) attacks pose significant threats by sending false, erroneous, or malicious model updates to the central server, attempting to extract private information from other devices during the federated learning process. This results in privacy leakage and reduced model accuracy. Traditional defenses measures against FR attacks typically employ auditing methods to identify malicious clients, but these methods are ineffective when multiple FR clients collude to inflate each other’s scores mutually. This paper proposes a novel defense method against collusion-based FR attacks. We first design a grouping mechanism based on gradient norm to group clients and then update the groups using an inter-client audit system. Finally, the correlation analysis of all groups is carried out to eliminate the attack group to ensure the security of the training process. This method defends against standard FR attacks and effectively detects attackers in collusion scenarios. Experimental results demonstrate that our method significantly improves the detection of malicious clients and enhances model accuracy by 10–20% compared to existing methods. Moreover, the proposed defense mechanism maintains its efficacy even in large-scale client environments, where more than 50% of the clients may be compromised by attackers.

Federated Learning (FL) empowers multiple clients to collaboratively learn a global generalization model without the need to share their local data, thus reducing privacy risks and expanding the scope of AI applications. However, current works focus less on data in a highly nonidentically distributed manner such as graph data which are common in reality, and ignore the problem of model personalization between clients for graph data training in federated learning. In this paper, we propose a novel personality graph federated learning framework based on variational graph autoencoders that incorporates model contrastive learning and local fine-tuning to achieve personalized federated training on graph data for each client, which is called FedVGAE. Then we introduce an encoder-sharing strategy to the proposed framework that shares the parameters of the encoder layer to further improve personality performance. The node classification and link prediction experiments demonstrate that our method achieves better performance than other federated learning methods on most graph datasets in the non-iid setting. Finally, we conduct ablation experiments, the result demonstrates the effectiveness of our proposed method.

As the size of datasets and neural network models increases, automatic parallelization methods for models have become a research hotspot in recent years. The existing auto-parallel methods based on machine learning or graph algorithms still have issues with search efficiency and applicability. This paper proposes an automatic parallel method based on a dual-population genetic algorithm, TGA, which transforms model partitioning and placement into an integer linear programming problem and constructs a cost model to evaluate the solution. The solution space is built using the neural network’s dataflow graph and device cluster’s topology, and the dual-population genetic algorithm is used to search for the optimal model parallel strategy. Experiments with various models show that the proposed method can improve single-step execution time by up to 42% compared to the Baechi method and up to 37.7% compared to the Hierarchical method.

Extending the length of time series forecasting has a long-term impact on smart grid energy consumption planning, residential electricity monitoring, extreme weather warning, and other real applications. This article studies the reliable long-term load trend forecasting approaches in smart grid environment. In recent years, the improved neural networks models based on self-attention mechanism show good performances in many sequence tasks, but most studies focus on reducing the complexity of networks layer, and can not restrain the increase of calculation error in longer distance forecasting scenarios. Also, most models lack the ability to mine potential high-dimensional features of time series. Based on these problems, we design a reliable hierarchical self-attention model named as long-term stability network (LTSNet), which adopts a tree-shaped decomposition neural network architecture based on hierarchical residual self-attention blocks, to top-down incrementally mine high-dimensional features of temporal components. At the same time, the attention matrix is used for feature interaction at each layer, to reduce the distribution gap between time series fragments in different domains. Compared with existing study models, LTSNet maintains stable forecasting performance and speed in long-term forecasting services, achieved the most reliable multivariate and univariate forecasting results in multiple domain scenarios. Compared with the latest models, the forecasting accuracy of the proposed model is improved by 22.8% and 13.8%, respectively, covering three applications: energy consumption, residential electricity consumption, and weather forecasting. At the same time, our experiments verify that the hierarchical decomposition networks can be used as a backbone architecture to effectively extended to longer dimensional load trend forecasting scenarios.