Spherical Inducing Features for Orthogonally-Decoupled Gaussian Processes
Authors: Louis C. Tiao, Vincent Dutordoir, Victor Picheny
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on multiple benchmark datasets demonstrate the effectiveness of our approach. |
| Researcher Affiliation | Collaboration | Louis C. Tiao 1 * Vincent Dutordoir 2 3 Victor Picheny 3 1University of Sydney, Sydney, Australia 2University of Cambridge, Cambridge, UK 3Secondmind, Cambridge, UK. |
| Pseudocode | No | No pseudocode or algorithm blocks are present in the paper. The paper contains mathematical equations and descriptions but no formatted algorithm. |
| Open Source Code | Yes | The open-source implementation of our method can be found at: ltiao/spherical-orthogonal-gaussian-processes. |
| Open Datasets | Yes | We evaluate our method on a number of well-studied regression problems from the UCI repository of datasets (Dua & Graff, 2017). Finally, we consider a large-scale regression dataset concerning U.S. commercial airline delays in 2008. |
| Dataset Splits | Yes | Unless otherwise stated, for each method and problem, we perform random sub-sampling validation by aggregating results from 5 repetitions across 10% held-out test sets. ... To quantitatively assess performance, we report the test RMSE and NLPD evaluated on a 1/3 held-out test set. |
| Hardware Specification | Yes | All experiments were carried out on a consumer-grade laptop computer with an Intel Core i7-11800H (8 Cores) @ 4.6GHz Processor, 16GB Memory, and a NVIDIA Ge Force RTX 3070 Laptop (Mobile/Max-Q) Graphics Card. |
| Software Dependencies | No | Table 2 lists key software dependencies by name and GitHub URL but does not provide specific version numbers for these software components, which is necessary for reproducibility. While SciPy 1.0 is mentioned in a citation, it is not explicitly stated as a versioned dependency for the implementation. |
| Experiment Setup | Yes | The choices of the hyperparameters and other relevant dependencies are summarized as follows: Optimization. We use the L-BFGS optimizer ... Likelihood. The Gaussian likelihood variance is initialized to 1.0 across all experiments. Kernel parameter initialization. All stationary kernels are initialized with unit lengthscale and amplitude. Variational parameter initialization. The variational distributions q(u), q(v ) are initialized with zero mean and identity covariance m = 0, C = I. |