The Brain–Computer Interface (BCI) has applications in smart homes and healthcare by converting EEG signals into control commands. However, traditional EEG signal decoding methods are affected by individual differences, and although deep learning techniques have made significant breakthroughs, challenges such as high energy consumption and the processing of raw EEG data remain. This paper introduces the Efficient Channel Attention Temporal Convolutional Network (ECA-ATCNet) to enhance feature learning by applying Efficient Channel Attention Convolution (ECA-conv) across spatial and spectral dimensions. The model outperforms state-of-the-art methods in both within-subject and between-subject classification tasks on MI-EEG datasets (BCI-2a and PhysioNet), achieving accuracies of 87.89% and 71.88%, respectively. Additionally, the proposed Spike Integrated Transformer Conversion (SIT-conversion) method, based on Spiking–Softmax, converts the Transformer’s self-attention mechanism into Spiking Neural Networks (SNNs) in just 12 time steps. The accuracy loss of the converted ECA-ATCNet model is only 0.6% to 0.73%, while its energy consumption is reduced by 52.84% to 53.52%. SIT-conversion enables ultra-low-latency, near-lossless ANN-to-SNN conversion, with SNNs achieving similar accuracy to their ANN counterparts on image datasets. Inference energy consumption is reduced by 18.18% to 45.13%. This method offers a novel approach for low-power, portable BCI applications and contributes to the advancement of energy-efficient SNN algorithms.

In recent years, joint entity–relation extraction (ERE) models have become a hot research topic in natural language processing (NLP). Several studies have proposed a span-based ERE framework, which utilizes simple span embeddings for entity and relation classification. This framework addresses the issues of overlap and error propagation that were present in previous entity–relation extraction models. However, span-based models overlook the influence of lexical information on the semantic representation of the span and fail to consider relations with a strong intrinsic connection between span pairs. To tackle these aforementioned issues, we present a new ERE model called ER-LAC (Span-based Joint Entity and Relation Extraction Model with Multi-level Lexical and Attention on Context Features). This model is designed with multi-granularity lexical features to enhance the semantic representation of spans, and a transformer classifier is employed to capture the internal connections between span pairs, thereby improving the performance of relational classification. To demonstrate the effectiveness of the proposed model, ablation experiments were conducted on the CoNLL04 dataset. The proposed model was also compared with other models on three datasets, showcasing its computational efficiency. The results indicate that the introduced lexical features and classifier enhance the F1 score for entity extraction by 0.84% to 2.04% and improve the F1 score for relationship classification by 0.96% to 2.26% when compared to the previous state-of-the-art (SOTA) model and the baseline SpERT model, respectively.

Model fusion is a technique of growing interest in the field of machine learning, which constructs a generalized model by merging the parameters of multiple independent models with different capabilities without the need to access the original training data or perform costly computations. However, during model fusion, when the number of parameters in a large language model is high, the dimension of the parameter space increases, which makes it more challenging to find the optimal combination of weights. Meanwhile, there is considerable potential for further development in sustainable optimization schemes for task-specific performance enhancement through model fusion in this area. In this paper, we propose a large-scale language model fusion approach based on task-enhanced reinforcement learning (RL–Fusion) to efficiently explore and optimize model fusion configurations. The key innovation of RL–Fusion lies in its use of reinforcement learning to guide parameter selection during model fusion, enabling a more intelligent and adaptive exploration of the parameter space. Additionally, RL–Fusion introduces a dynamic evaluation mechanism that adjusts the evaluation dataset in real-time based on feedback from SOTA models, ensuring continuous enhancement of domain-specific capabilities. RL–Fusion outperforms the baseline model by improving 1.75% in the MMLU benchmark test, 1.8% in the C-eval test, and 1.8% in the Chinese Named Entity Recognition (NER) test on the Yayi NER dataset by 16%. The results show that RL–Fusion is an effective and scalable model fusion solution that improves performance without increasing the computational cost of traditional optimization methods and has a wide range of applications in AI research and practice.

Studies on integrating Spiking Neural Networks (SNNs) with the Transformer architecture holds promise for enabling models to achieve ultra-low energy consumption while possessing the performance of the Transformer architecture. Currently, studies on ANN-to-SNN conversion of integrating Spiking Neural Networks (SNNs) with the Transformer architecture mainly focuses on simple activation functions in MLPs, and has not yet addressed the mismatch between the Softmax activation function in the self-attention mechanism and the computation rules of SNNs. Consequently, the ANN-to-SNN conversion efforts have consistently failed to make the Transformer architecture directly applicable to SNNs. To address this challenge, we propose the Spiking-Softmax method, which integrates Spiking Exponential Neuron (SI-exp) and Spiking Collaboration Normalized Neuron (SI-norm). The Spiking-Softmax method accurately simulates the Softmax activation function with only 12 time steps. Building upon this, we propose the Spike Integrated Transformer conversion (SIT-conversion) method, which enables the conversion of the Transformer architecture to SNNs. The SNNs generated by the SIT-conversion of Transformer models of various sizes achieve accuracy nearly identical to their ANNs counterparts, achieving nearly lossless and ultra-low-latency ANN-to-SNN conversion. This work represents the first implementation of simulating the Softmax activation function and fully converting the Transformer architecture into SNNs through spike firing.