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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
TrajMamba: An Efficient and Semantic-rich Vehicle Trajectory Pre-training Model
Authors: Yichen Liu, Yan Lin, Shengnan Guo, Zeyu Zhou, Youfang Lin, Huaiyu Wan
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on two real-world datasets and three downstream tasks show that Traj Mamba outperforms state-of-the-art baselines in both efficiency and accuracy. |
| Researcher Affiliation | Academia | 1School of Computer Science and Technology, Beijing Jiaotong University, China 2Department of Computer Science, Aalborg University, Denmark |
| Pseudocode | No | The paper describes its methodology in Section 4 and includes figures like Figure 2 and Figure 3, but it does not present any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code of Traj Mamba is available at https://github.com/yichenliuzong/Traj Mamba. |
| Open Datasets | Yes | In our experiments, we use two real-world datasets released by Didi3, called Chengdu and Xian. ...we fetch the information of POIs from the AMap API4 and road networks from Open Street Map5. ...3https://gaia.didichuxing.com/ 4https://lbs.amap.com/api/javascript-api-v2 5https://www.openstreetmap.org/ |
| Dataset Splits | Yes | We split the datasets into training, validation, and testing sets in an 8:1:1 ratio, with departure times in chronological order. |
| Hardware Specification | Yes | The experiments are conducted on servers equipped with Intel(R) Xeon(R) W-2155 CPUs and n Vidia(R) TITAN RTX GPUs. |
| Software Dependencies | No | The Traj Mamba model is implemented using Py Torch [31]. (PyTorch is mentioned, but no specific version number is provided in the text). |
| Experiment Setup | Yes | The four key hyper-parameters of Traj Mamba and their optimal values are L = 5, E = 256, N = 32, and H = 4. ... For model training, we use the Adam optimizer with a batch size of B = 128, and the initial learning rates of two pre-training procedures are 0.001 and 0.0001, respectively. We set the radius for neighboring POIs selection R = 300 meters. In Eq. 14, we set α = 1.0, β = 0.5 to balance each factor. |