Social recommendation utilizes social relationships to alleviate data sparsity and improve recommendation quality. Nevertheless, existing approaches still encounter several limitations. First, user preference structures in social networks are often hierarchical and non-Euclidean, whereas most existing models rely on Euclidean embeddings that are insufficient for capturing such complex geometries. Second, noisy social connections may propagate unreliable information and degrade representation quality. Third, commonly used multi-view fusion strategies based on multilayer perceptrons (MLPs) often lack the expressive capability required to effectively integrate heterogeneous user signals. To address these challenges, this paper proposes HGKAN (Hyperbolic Graph-Based Kolmogorov–Arnold Network), a unified framework that combines hyperbolic representation learning, graph neural networks, and KAN-based nonlinear fusion for social recommendation. Specifically, a Hyperbolic Intent Attention module is introduced to learn fine-grained and hierarchical user intent representations in hyperbolic space. In addition, a Hyperbolic Attention GCN is designed to denoise and aggregate social information while preserving complex relational structures. Furthermore, a KAN-based Multi-View Fusion Layer is developed to replace conventional MLP-based fusion, enabling more expressive and interpretable integration of preference, social, and intent representations. Extensive experiments conducted on three public benchmark datasets, including Yelp, Ciao, and Douban, demonstrate that HGKAN consistently outperforms several state-of-the-art recommendation methods. Experimental results show that the proposed framework achieves substantial improvements in both Recall@5 and NDCG@5, with all performance gains being statistically significant (p < 0.05). These findings verify the effectiveness of integrating hyperbolic geometric learning with KAN-based multi-view fusion for robust and accurate social recommendation.

In this study, we propose AlphaGrad, a novel adaptive loss blending strategy for optimizing multi-task learning (MTL) models in motor imagery (MI)-based electroencephalography (EEG) classification. AlphaGrad is the first method to automatically adjust multi-loss functions with differing metric scales, including mean square error, cross-entropy, and deep metric learning, within the context of MI-EEG. We evaluate AlphaGrad using two state-of-the-art MTL-based neural networks, MIN2Net and FBMSNet, across four benchmark datasets. Experimental results show that AlphaGrad consistently outperforms existing strategies such as AdaMT, GradApprox, and fixed-weight baselines in classification accuracy and training stability. Compared to baseline static weighting, AlphaGrad achieves over 10% accuracy improvement on subject-independent MI tasks when evaluated on the largest benchmark dataset. Furthermore, AlphaGrad demonstrates robust adaptability across various EEG paradigms—including steady-state visually evoked potential (SSVEP) and event-related potential (ERP), making it broadly applicable to brain-computer interface (BCI) systems. We also provide gradient trajectory visualizations highlighting AlphaGrad's ability to maintain training stability and avoid local minima. These findings underscore AlphaGrad's promise as a general-purpose solution for adaptive multi-loss optimization in biomedical time-series learning.