摘要
文章提出了融合稀疏注意力预处理、低秩近似处理和递归计算整合的优化方法,详细阐述了其设计思路与算法流程,并通过长文本分类和时间序列预测实验进行了验证。结果表明,该优化策略显著提升了模型的性能,为Transformer在长序列任务中的应用提供了有力的技术支持。
The paper proposes an optimized method that integrates sparse attention preprocessing,low-rank approximation processing,and recursive computation.It elaborates on the design concept and algorithm flow,and validates it through experiments in long text classification and time series prediction.The results show that this optimization strategy significantly enhances model performance,providing strong technical support for the application of Transformer in long sequence tasks.
作者
江雨欣
JIANG Yuxin(Institute of Artificial Intelligence,Hubei University,Wuhan Hubei 430062,China)
出处
《信息与电脑》
2025年第22期82-84,共3页
Information & Computer
