Hierarchical Deep Generative Models for Multi-Rate Multivariate Time Series

Authors: Zhengping Che, Sanjay Purushotham, Guangyu Li, Bo Jiang, Yan Liu

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on two real-world datasets demonstrate that our MR-HDMM model outperforms the existing state-of-the-art deep learning and state-space models on forecasting and interpolation tasks.
Researcher Affiliation Academia 1Department of Computer Science, University of Southern California, Los Angeles, California, United States.
Pseudocode Yes Algorithm 2 Learning MR-HDMM with stochastic backpropagation and SGD
Open Source Code No The paper does not provide any concrete access information (e.g., repository link, explicit statement of code release) for the implementation of the described methodology.
Open Datasets Yes MIMIC-III is a public de-identified dataset collected at Beth Israel Deaconess Medical Center from 2001 to 2012 (Johnson et al., 2016). The U.S. Historical Climatology Network Monthly (USHCN) dataset (Menne et al., 2010) is publicly available and consists of daily meteorological data of 54 stations in California spanning from 1887 to 2009.
Dataset Splits Yes For experiments on USHCN dataset, train/valid/test sets were split as 70/10/20. For experiments on MIMIC-III, we used 5-fold cross validation (train on 3 folds, validate on another fold and test on the remaining fold) and report the average Mean Squared Error (MSE) of 5 runs for both forecasting and interpolation tasks.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments, such as specific GPU or CPU models.
Software Dependencies No The paper mentions using "stats-toolbox (Seabold & Perktold, 2010) in python" and "pykalman (Duckworth, 2013)", but does not provide specific version numbers for these software components or any other key dependencies.
Experiment Setup Yes For generation model in MR-HDMM, we use multivariate Gaussian with diagonal covariance for both emission distribution and transition distribution. We parameterized the emission mapping gθx by a 3-layer MLP with Re LU activations, the transition mapping gθz by gated recurrent unit (GRU), and mapping gθs by a 3-layer MLP with Re LU activations on the hidden layers and linear activations on the output layer. For inference networks, we adopt filtering setting for forecasting and bidirection setting for interpolation from Table 1 with 3-layer GRUs. To update θs, we replace the sign function with a sharp sigmoid function during training, and use the indicator function during validation. We train all the deep learning models with the Adam optimization method (Kingma & Ba, 2014) and use validation set to find the best weights, and report the results on the held-out test set. All the input variables are normalized to be of 0 mean and 1 standard deviation.