The distribution biases and scarcity of samples in multi-source data present significant challenges for few-shot learning (FSL) tasks based on brain-computer interface (BCI). Recent efforts have explored the application of diffusion mechanisms in FSL, typically utilizing labeled data to augment the support set. However, this approach has not effectively utilized unlabeled data nor addressed distribution biases. Inspired by the latest advancements in FSL, we propose the manhattan self-attention diffusion residual networks (MSADiff-Resnet) with dynamic bias rectification. This model explicitly adds the manhattan self-attention diffusion layer to resnet, using attention mechanisms and manhattan distance-based decay function to control local diffusion intensity, and adjusts the global diffusion strength through the parameter. This diffusion mechanism bridges labeled and unlabeled data, addressing the limitations associated with sample availability. Additionally, we effectively tackle the distribution biases of multi-source data through inter-class bias rectification and dynamic intra-class bias rectification. Moreover, this study presents for the first time a universal deep learning framework specifically designed for BCI-based FSL tasks. Extensive experiments on multi-source BCI task datasets have validated the effectiveness of proposed method.

Emotion recognition tasks based on physiological signals require the simultaneous capture of local features and global correlations of these signals. Although transformer-based models are widely used due to their superior ability to integrate information, their quadratic computational complexity limits their efficiency in processing large-scale or high-resolution data. Recently, state space models (SSM) with efficient hardware-aware designs have demonstrated significant potential in modeling long sequences. However, existing SSMs face limitations in processing global information due to window constraints. Therefore, this paper introduces a novel Global Context (GC) MambaVision model, which combines the linear time complexity advantage of SSMs with a new type of local-global attention mechanism. GC MambaVision maintains high computational efficiency in emotion recognition tasks while providing a more comprehensive understanding of the dynamic changes in local and global emotional states. Experimental results on the DEAP and SEED-V datasets show that GC MambaVision achieves superior performance compared to current state-of-the-art models, with accuracies reaching 98.62% and 85.88%, respectively.

Integrating mobility into WSNs can significantly reduce the energy consumption of sensor nodes. However, this may lead to unacceptable data collection latency at the same time. In our previous work, we alleviated the problem under the assumption of a mobile base station (BS). In this paper, we discuss how the problem can be solved when the BS itself is not capable of moving, but it can instead employ some mobile elements (MEs). The data collection latency is mainly determined by the longest tour of the MEs in this case. Each ME should be assigned a similar workload to reduce the latency. Furthermore, the total length of the tours should be minimized to decrease the working cost of MEs. We propose three methods to solve the problem with these two-fold objectives. In the first two methods, we cluster the network according to some criteria, and then construct the data collection tour for each ME. We apply a heuristic operator based on the genetic algorithm in the third method, whose fitness function is defined according to the two-fold objectives. These methods are evaluated by comprehensive experiments. The results show that the genetic method can provide us more steady solutions in term of data collection latency. We also compare the mobile BS model and the multiple MEs model, whose results show that the latter can get us better solutions when the number of MEs gets larger.