Variational Elliptical Processes
Authors: Maria Margareta Bånkestad, Jens Sjölund, Jalil Taghia, Thomas B. Schön
TMLR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We highlight advantages compared to Gaussian processes through regression and classification experiments. We examined the variational elliptical processes using four different experiments. |
| Researcher Affiliation | Collaboration | Maria B ankestad EMAIL RISE Research Institutes of Sweden and Uppsala University, Sweden Jens Sjölund EMAIL Department of Information Technology, Uppsala University, Sweden Jalil Taghia EMAIL Ericsson Research, Sweden Thomas B. Schön EMAIL Department of Information Technology, Uppsala University, Sweden |
| Pseudocode | Yes | Algorithm 1 Py Torch implementation of the variational sparse elliptical process (VI-EP-EP). |
| Open Source Code | Yes | Accompanying code is found at . |
| Open Datasets | Yes | We conducted experiments on seven datasets from the UCI repository (Dua & Graff, 2017) |
| Dataset Splits | Yes | For all experiments, we created ten random train/val/test splits with the proportions 0.6/0.2/0.2, except for the two smallest datasets (mpg and concrete), where we neglected the validation dataset and used a train/test proportions of 0.7/0.3. |
| Hardware Specification | No | The paper mentions 'running it on a GPU' but does not specify any particular model or detailed specifications for the hardware used in the experiments. |
| Software Dependencies | No | The paper mentions 'Pyro library' and 'Py Torch' but does not specify their version numbers. |
| Experiment Setup | Yes | In the regression experiments in Section 4.3, we ran all experiments using the Adam optimizer (Kingma & Ba, 2015) with a learning rate of 0.01. For the full GP, we used the L-BFGS optimizer to train the hyperparameters. We run the optimizer for the large dataset for 500 epochs and the small dataset for 2000 epochs. The likelihood mixing distribution uses a spline flow with nine bins and Softplus as its output transformation. The elliptic posterior mixing distribution uses a spline flow with five bins and a Sigmoid output transformation. |