Dynamic Manifold Learning for Land Deformation Forecasting
Authors: Fan Zhou, Rongfan Li, Qiang Gao, Goce Trajcevski, Kunpeng Zhang, Ting Zhong4725-4733
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We collected real-world In SAR point cloud data of slopes evolution and conducted extensive experiments to evaluate the effectiveness of our proposed model. The results show that Dy Land significantly outperforms other state-of-the-art models in forecasting land deformation, learning dynamic manifold and providing interpretable predictions. |
| Researcher Affiliation | Academia | 1 University of Electronic Science and Technology of China 2 Southwestern University of Finance and Economics 3 Iowa State University 4 University of Maryland, College park fan.zhou@uestc.edu.cn, rongfanli1998@gmail.com, qianggao@swufe.edu.cn, gocet25@iastate.edu, kpzhang@umd.edu, zhongting@uestc.edu.cn |
| Pseudocode | No | The paper describes its methods but does not include a dedicated pseudocode or algorithm block. |
| Open Source Code | No | The paper does not provide any links to open-source code or explicitly state that the code for their methodology is available. |
| Open Datasets | No | The paper uses |
| Dataset Splits | Yes | We split all datasets into three parts: 50% for training, 30% for validation, and the remaining 20% for testing. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions optimizing with Adam optimizer and using deep learning methods, but does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | All deep learning methods are optimized by Adam optimizer (Kingma and Ba 2017) with learning rate of 10 3 and weight decay of 10 5. Early stop is triggered when the loss has not declined for 100 consecutive epochs. ... In our Dy Land, the latent manifold U is 3D for evaluation and 2D for visualization. The dimension of spatio-temporal representation W is 5. The deformation for each location sτ i is scaled up by multiplying 10 and then passed to NF for conditioning, since the scale of sτ i is too small compared to V. The NF and neural ODEs have 3 layers and 64 dimensions, while both FC and GNN have 3 layers. |