DeepUrbanMomentum: An Online Deep-Learning System for Short-Term Urban Mobility Prediction
Authors: Renhe Jiang, Xuan Song, Zipei Fan, Tianqi Xia, Quanjun Chen, Satoshi Miyazawa, Ryosuke Shibasaki
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the superior performance of our proposed model as compared to the existing approaches. |
| Researcher Affiliation | Academia | Center for Spatial Information Science, The University of Tokyo Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology {jiangrh, songxuan, xiatianqi, koitaroh, shiba}@csis.u-tokyo.ac.jp, {fanzipei, chen1990}@iis.u-tokyo.ac.jp |
| Pseudocode | No | The paper describes the model mathematically and textually, but it does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states that "Python and some Python libraries including Keras(Chollet 2015) and Tensor Flow(Abadi et al. 2015) are used to implement our system," but it does not provide a direct link to their own source code or an explicit statement about its public availability. |
| Open Datasets | No | A raw GPS log dataset was collected anonymously from approximately 1.6 million mobile phone users in Japan over a three-year period (August 1, 2010 to July 31, 2013)1. Footnote 1 clarifies: "Konzatsu-Tokei (R) from ZENRIN Data Com CO.LTD is used by us... through the docomo map navi service provided by NTT DOCOMO, INC." This indicates a proprietary dataset and no public access information is provided. |
| Dataset Splits | Yes | We randomly select 80% of the data for model training and use the remaining 20% for validation, which is used to early-stop our training algorithm if the validation error is converged. |
| Hardware Specification | No | The paper specifies the hardware for data storage ("this dataset is stored on a Hadoop cluster, containing 32 cores, 32 GB memory, and 16 TB storage, which can run 28 tasks simultaneously"), but it does not provide details on the hardware used for training the deep learning models (e.g., specific GPUs or CPUs). |
| Software Dependencies | No | The paper states that "Python and some Python libraries including Keras(Chollet 2015) and Tensor Flow(Abadi et al. 2015) are used to implement our system," but it does not provide specific version numbers for these software components. |
| Experiment Setup | Yes | For both Simple RNN and Deep RNN, a 64-dimension vector is used as the latent representation Zt of the entire Xt. The RMSprop algorithm is adopted in our system to govern the whole training process. Based on the empirical tuning result, we found the current urban mobility Xt.α = 3 and m = 3 in Deep RNN would be appropriate. |