Cross-City Transfer Learning for Deep Spatio-Temporal Prediction
Authors: Leye Wang, Xu Geng, Xiaojuan Ma, Feng Liu, Qiang Yang
IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Using crowd flow prediction as a demonstration experiment, we verify the effectiveness of Region Trans. Table 1 shows our results for D.C. Chicago and D.C. NYC. Region Trans can consistently outperform the best baseline, where the largest improvement is reducing RMSE by up to 10.7%. |
| Researcher Affiliation | Collaboration | Leye Wang1,2 , Xu Geng2 , Xiaojuan Ma2 , Feng Liu3 and Qiang Yang2,4 1Key Lab of High Confidence Software Technologies, Peking University, Ministry of Education, China 2Hong Kong University of Science and Technology, Hong Kong, China 3SAIC Motor, Shanghai, China 4We Bank, Shenzhen, China |
| Pseudocode | Yes | Algorithm 1 Region-based cross-city network parameter optimization |
| Open Source Code | No | The paper does not contain any explicit statement about making its source code publicly available, nor does it provide a link to a code repository. |
| Open Datasets | Yes | Three bike flow datasets collected from Washington D.C., Chicago, and New York City are used. Each dataset covers a two-year period (2015-2016). ... Following previous studies on crowd flow [Hoang et al., 2016; Zhang et al., 2017; Zhang et al., 2016], we use bike flow data for evaluation. |
| Dataset Splits | No | The paper mentions that "The last two-month data is chosen for testing" and that the target city has "limited crowd flow data (e.g., one day)" (implied for training). However, it does not specify a distinct validation set or explicit percentages for training/validation splits. |
| Hardware Specification | Yes | We use a server with Intel Xeon CPU E5-2650L, 128 GB RAM, and Nvidia Tesla M60 GPU. |
| Software Dependencies | No | We implement Region Trans with Tensor Flow (Cent OS). The paper mentions TensorFlow and CentOS but does not provide specific version numbers for these software dependencies, which are necessary for full reproducibility. |
| Experiment Setup | Yes | Our network structure implemented in the experiment has two layers of Conv LSTM with 5 5 filters and 32 hidden states, to generate Xrep t R20 20 32. With Xrep t as the input, there is one layer of Conv2D1 1 with 32 hidden states, followed by another layer of Conv2D1 1 linking to the output crowd flow prediction. For the external context factors, e.g., temperature, wind speed, weather, and day type, we use the same feature extrac-tion method as [Zhang et al., 2017] and obtain an external feature vector with a length of 28. We also need to set w in Eq. 14 to balance the optimization trade-off between representation difference and prediction error. We set w to 0.75 as the default value. ADAM is used as the optimization algorithm [Kingma and Ba, 2015]. |