The paper by Jaehyuk Yi and Jinkyoo Park introduces a hypergraph convolutional recurrent neural network (HGC-RNN), designed for structured time-series sensor network data prediction. This model is particularly adept at representing sensor networks in graph structures, capturing complex relationships among sensors that conventional graph structures may not adequately represent. The HGC-RNN performs hypergraph convolution on input data to extract hidden representations, considering the structural dependency of the data. It also employs a recurrent neural network structure to learn temporal dependencies from data sequences. The paper demonstrates the model's application in forecasting taxi demand in NYC, traffic flow in overhead hoist transfer systems, and gas pressure in gas regulators, showing its superiority over baseline models.
Key Points
HGC-RNN Model: A novel prediction model for structured time-series sensor network data.
Hypergraph Convolution: Utilizes hypergraph convolution for structural representation and feature extraction.
Temporal Dependency Learning: Employs RNN structure for learning from data sequences.
Applications: Demonstrated in forecasting taxi demand, traffic flow, and gas pressure in various systems.
Performance: Outperforms existing baseline models in the tested applications.
Citation
@inproceedings{Yi2020,
title={Hypergraph Convolutional Recurrent Neural Network},
author={Yi, Jaehyuk and Park, Jinkyoo},
booktitle={Proceedings of the 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’20)},
year={2020},
url={https://tmpfiles.org/dl/3316673/yirnn.pdf}
}
Title
Hypergraph Convolutional Recurrent Neural Network
URL
(https://tmpfiles.org/dl/3316673/yirnn.pdf)
Summary
The paper by Jaehyuk Yi and Jinkyoo Park introduces a hypergraph convolutional recurrent neural network (HGC-RNN), designed for structured time-series sensor network data prediction. This model is particularly adept at representing sensor networks in graph structures, capturing complex relationships among sensors that conventional graph structures may not adequately represent. The HGC-RNN performs hypergraph convolution on input data to extract hidden representations, considering the structural dependency of the data. It also employs a recurrent neural network structure to learn temporal dependencies from data sequences. The paper demonstrates the model's application in forecasting taxi demand in NYC, traffic flow in overhead hoist transfer systems, and gas pressure in gas regulators, showing its superiority over baseline models.
Key Points
Citation