Variational Dependent Multi-output Gaussian Process Dynamical Systems
Authors: Jing Zhao, Shiliang Sun
JMLR 2016 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate the model on both synthetic and real-world data including motion capture data, traffic flow data and robot inverse dynamics data. Various evaluation methods are taken on the experiments to demonstrate the effectiveness of our model, and encouraging results are observed. |
| Researcher Affiliation | Academia | Jing Zhao EMAIL Shiliang Sun EMAIL Department of Computer Science and Technology East China Normal University 500 Dongchuan Road, Shanghai 200241, P. R. China |
| Pseudocode | No | The paper describes methods and derivations mathematically but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | Our code is implemented based on the framework of publicly available code for the VGPDS and CMOGP. |
| Open Datasets | Yes | In order to demonstrate the validity of the proposed model on real-world data, we employ ten sequences of runs/jogs from subject 35 (see Figure 4 for a skeleton) and two sequences of runs/jogs from subject 16 in the CMU motion capture database for the reconstruction task. |
| Dataset Splits | Yes | We randomly select 50 points from each output for training with the remaining 150 points for testing. This is repeated for ten times. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | Our code is implemented based on the framework of publicly available code for the VGPDS and CMOGP. However, specific version numbers for libraries or frameworks are not provided. |
| Experiment Setup | Yes | For approximation, 30 random inducing points are used. In addition, white Gaussian noise is added to each output. For the VDM-GPDS and VGPDS, the RBF kernel is adopted in this set of experiments to construct Kt,t which is a block-diagonal matrix because the sequences are independent. Moreover, the latent variables X in the VGPDS and VDM-GPDS with nine dimensions are initialized by using principal component analysis on the observations. For parameter optimization of the VDM-GPDS and VGPDS, the maximum numbers of iteration steps are set to be identical. We use 100, 300 and 500 points for training, respectively and the rest for testing. All the experiments are repeated for ten times. Particularly, 15, 20 and 30 inducing points are used for 100, 300 and 500 training points, respectively. For the VDM-GPDS and VGPDS, the dimensionality of the latent space is set to two. |