Compression Method for Deep Diagonal State Space Model Based on $H^2$ Optimal Reduction

--Deep learning models incorporating linear SSMs have gained attention for capturing long-range dependencies in sequential data. However, their large parameter sizes pose challenges for deployment on resource-constrained devices. In experiments, the LRA benchmark results show that the model compression based on our proposed method outperforms an existing method using the Balanced Truncation, while successfully reducing the number of parameters in the SSMs to 1/ 32 without sacrificing the performance of the original models. Deep learning models that incorporate linear State Space Models (SSMs) [1] have achieved remarkable success across various fields, including text [2], audio [3], images [4], and videos [5]. Since the introduction of this class of models in works such as [6]-[8], these architectures have attracted increasing attention, with several studies demonstrating their potential to model long-range dependencies in sequential data [8]-[12].