Correction
Open
mLaSDI: Multi-stage latent space dynamics identification
ref [41] · 2506.09207 · notice #6600 · dispute
Raw extraction · bibliography line
J. Zabalza, J. Ren, J. Zheng, H. Zhao, C. Qing, Z. Yang, P. Du, and S. Marshall. Novel segmented stacked autoencoder for effective dimensionality reduction and feature extraction in hyperspectral imaging. Neurocomputing, 185:1–10, 2016. ISSN 0925-2312. doi: https: //doi.org/10.1016/j.neucom.2015.11.044. 12 Table 1: Hyperparameters and training time of autoencoders for multiscale oscillating example in Section 4.1, using Tanh activation function. Toy Problem Training Iterations Learning Rate β1 β2 Architecture Training Time GPLaSDI 50,000 10−3 10−1 10−3 600-100-5 169.11s mLaSDI (1st stage) 10,000 10−3 10−1 10−3 600-100-5 35.63s mLaSDI (2nd stage) 10,000 10−3 100 10−3 600-100-5 34.87s A Experiment Details In this section we provide full experiment details, including methods for data generation and training of our autoencoders. For all examples in the manuscript, our autoencoders were trained using the PyTorch implementation of the Adam optimizer [25] with Tanh activation function. We provide hyperparameters for each neural network in the sections below. When optimizing the GP parameters for our SINDy coefficients, we use the GaussianProcessRegressor from scikit-learn [30]. We use the