基于 LSTM 的在线学习中解耦扩展卡尔曼滤波学习算法的稳定性
机器学习
2021-06-01 v4 信号处理
机器学习
摘要
我们研究了在基于长短期记忆网络 (LSTM) 的在线学习框架内，解耦扩展卡尔曼滤波学习算法 (DEKF) 的收敛性与稳定性性质。为此，我们将 DEKF 建模为受扰动的扩展卡尔曼滤波器，并推导出其在 LSTM 训练期间稳定的充分条件。我们表明，如果由于解耦引入的扰动保持有界，则 DEKF 以与全局扩展卡尔曼滤波学习算法相似的收敛和稳定性性质学习 LSTM 参数。我们通过若干数值仿真验证了我们的结果，并将 DEKF 与其他 LSTM 训练方法进行比较。在我们的仿真中，我们还观察到文献中用于 DEKF 的知名超参数选择方法满足我们的条件。
引用
@article{arxiv.1911.12258,
title = {Stability of the Decoupled Extended Kalman Filter Learning Algorithm in LSTM-Based Online Learning},
author = {Nuri Mert Vural and Fatih Ilhan and Suleyman S. Kozat},
journal= {arXiv preprint arXiv:1911.12258},
year = {2021}
}
备注
This paper was an early draft of the presented results. We have written and published another paper (arXiv:1911.12258) where we have improved on the material in this paper. The published paper covers most of the material presented in this paper as well. Therefore, we remove this paper from Arxiv and refer the interested readers to arXiv:1911.12258