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 Shift Toward Open and Reproducible AI Research," Under Review, 2026. doi: 10.48550/arXiv.2606.16974..
Kernel Interpolation for Scalable Structured Gaussian Processes (KISS-GP)
Authors: Andrew Wilson, Hannes Nickisch
ICML 2015 | Venue PDF | LLM Run Details | Input Tokens: 17,613 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 4,226 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate SKI for kernel matrix approximation (section 4.1), kernel learning (section 4.2), and natural sound modelling (section 4.3). |
| Researcher Affiliation | Collaboration | Andrew Gordon Wilson EMAIL Carnegie Mellon University Hannes Nickisch EMAIL Philips Research Hamburg |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | We have implemented code as an extension to the GPML toolbox (Rasmussen & Nickisch, 2010). For updates and demos, see http://www.cs.cmu.edu/ andrewgw/pattern |
| Open Datasets | Yes | We use SKI to model the natural sound time series in Fig 3(a), considered in a different context by Turner (2010). |
| Dataset Splits | No | The paper mentions |
| Hardware Specification | Yes | All experiments were performed on a 2011 Mac Book Pro, with an Intel i5 2.3 GHz processor and 4 GB of RAM. |
| Software Dependencies | No | The paper mentions the |
| Experiment Setup | Yes | For SKI, we use cubic interpolation and a 100 100 inducing point grid, equispaced in each input dimension. That is, we have as many inducing points m = 10, 000 as we have training datapoints. We use the same θ initialisation for each approach. |