Neural Stochastic Differential Games for Time-series Analysis
Authors: Sungwoo Park, Byoungwoo Park, Moontae Lee, Changhee Lee
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Throughout the experiments on various datasets, we demonstrate the superiority of our framework over all the tested benchmarks in modeling time-series prediction by capitalizing on the advantages of applying cooperative games. |
| Researcher Affiliation | Collaboration | 1LG AI Research 2Artificial Intelligence Graduate School, Chung-Ang University 3Department of Information and Decision Sciences, University of Illinois Chicago. Correspondence to: Sungwoo Park <sungwoopark.lg@lgresearch.ai>. |
| Pseudocode | Yes | Algorithm 1 Deep Neural Fictitious Play |
| Open Source Code | Yes | Our code is available at https://github.com/LGAI-AML/Ma SDEs. |
| Open Datasets | Yes | We evaluated the time-series prediction performance of ours and the benchmarks on multiple real-world datasets: BAQD (Zhang et al., 2017), Speech (Warden, 2018), and Physionet (Silva et al., 2012). |
| Dataset Splits | No | We split each time-series in the interval [0, T] into two sub-intervals: the first 80% as the observation interval, i.e., O = [0, 0.8T], and the remaining 20% as the prediction interval, i.e., T = [0.8T, T]. We split time-series samples into two halves as training/evaluation sets in Physionet. For BAQD and Speech datasets, we divided time-series samples into 80/20 training/testing splits for training and evaluation, respectively. |
| Hardware Specification | No | No specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running experiments were provided. |
| Software Dependencies | No | No specific software dependencies with version numbers (e.g., library names or solver names with specific versions) were mentioned. |
| Experiment Setup | Yes | In all experiments with real-world datasets, we train each model for 500 epochs using the Adam optimizer with a learning rate of 10^-3 and batch size of 128. |