Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Cross-City Latent Space Alignment for Consistency Region Embedding
Authors: Meng Chen, Hongwei Jia, Zechen Li, Wenzhen Jia, Kai Zhao, Hongjun Dai, Weiming Huang
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show Co RE outperforms competitive baselines, confirming its effectiveness for crosscity knowledge transfer via aligned latent spaces. ... 4. Experiments In this section, we evaluate the proposed Co RE on three cross-city urban prediction tasks with real-world datasets and compare its performance against various baselines. ... Table 1 presents the cross-city socioeconomic indicator prediction performance of various methods on the XA and CD datasets, where X denotes the source city and Y denotes the target city. ... 4.4. Ablation Study and Parameter Analysis |
| Researcher Affiliation | Collaboration | 1School of Software, Shandong University, Jinan, China 2Business School, Shandong Normal University, Jinan, China 3Walmart AI Lab, California, USA 4Department of Physical Geography and Ecosystem Science, Lund University, Sweden. |
| Pseudocode | No | The paper describes methods using mathematical equations and textual descriptions of steps, but it does not contain any explicitly labeled "Pseudocode" or "Algorithm" blocks. |
| Open Source Code | Yes | Data and source codes are available at: https://github.com/AIMUrban/Co RE. |
| Open Datasets | Yes | Region data is obtained from Beijing City Lab1. ... Human mobility data includes one-month taxi trips from three cities, with each trip providing the latitude and longitude of both the origin and destination (Jiang et al., 2023). ... Socioeconomic indicator data includes GDP data (Zhao et al., 2017), population data (Bondarenko et al., 2020), and carbon emission data (Oda etal., 2018) of the three cities, which are used in downstream tasks. 1https://www.beijingcitylab.com/data-released-1/ |
| Dataset Splits | Yes | The training data consists of region representations (ZX) and socioeconomic indicator labels ({y X i }NX i=1), including GDP, population, and carbon emissions, from the source city (X). The test data comprises region representations (ZY ) and corresponding labels ({y Y i }NY i=1) from the target city (Y ). |
| Hardware Specification | No | The paper does not explicitly mention any specific hardware used for running the experiments (e.g., GPU models, CPU models, or cloud computing instance types). |
| Software Dependencies | No | The paper mentions using an "Adam optimizer" and a "two-layer GAT as the encoder", but it does not provide specific version numbers for any software libraries, frameworks, or programming languages used in the implementation. |
| Experiment Setup | Yes | In the base region embedding model, we utilize a two-layer GAT as the encoder with an embedding dimension d set to 128 and the attention heads H set to 8. For cross-city latent manifold alignment, we initially generate 1000 anchors... The optimization of the model is conducted using the Adam optimizer, with a learning rate configured at 1 10 3. ... The terms α and β are hyperparameters that adjust the weights of the LMA loss and the LIA loss, respectively. ... Figure 7 illustrates the MAE results for the XA/CD pair, demonstrating that Co RE achieves optimal performance when both α and β are set to 1. ... Ridge regression is employed as the task predictor. |