Nonstationary Sparse Spectral Permanental Process

Authors: Zicheng Sun, Yixuan Zhang, Zenan Ling, Xuhui Fan, Feng Zhou

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on both synthetic and real-world datasets demonstrate the effectiveness of our approach, particularly in scenarios with pronounced data nonstationarity. Additionally, ablation studies are conducted to provide insights into the impact of various hyperparameters on model performance.
Researcher Affiliation Academia Zicheng Sun1 , Yixuan Zhang2 , Zenan Ling3, Xuhui Fan4 Feng Zhou1,5 1Center for Applied Statistics and School of Statistics, Renmin University of China 2School of Statistics and Data Science, Southeast University 3School of EIC, Huazhong University of Science and Technology 4School of Computing, Macquarie University 5Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing
Pseudocode Yes E Pseudocode The pseudocode of our inference algorithm is provided in Algorithm 1. Algorithm 1: Inference for (D)NSSPP
Open Source Code Yes Code is publicly available at https://github.com/SZC20/DNSSPP.
Open Datasets Yes Coal Mining Disaster [13]: This is a 1-dimensional dataset containing 191 incidents that occurred between March 15, 1875, and March 22, 1962. ... Redwoods [27]: This is a 2-dimensional dataset... Porto Taxi [21]: This is a large 2-dimensional dataset...
Dataset Splits Yes For each intensity, we simulate ten datasets and use each dataset alternately as the training set and the remaining ones as the test sets. For each dataset, we randomly partition the data into training set and test set of approximately equal size.
Hardware Specification Yes We perform all experiments using the server with two GPUs (NVIDIA TITAN V with 12GB memory), two CPUs (each with 8 cores, Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz), and 251GB memory.
Software Dependencies No The paper does not explicitly state specific version numbers for software dependencies such as programming languages or libraries.
Experiment Setup Yes For NSSPP, the number of frequencies (network width) is set to 50. For DNSSPP, we experimented with three different configurations: DNSSPP-[50,30], DNSSPP-[100,50], and DNSSPP-[30,50,30]. Each number represents the width of the corresponding network layer. We investigate DNSSPP s performance with varying network sizes by adjusting width and depth on nonstationary synthetic data. We conduct experiments on the nonstationary synthetic data to investigate the coordination of epoch and learning rate.