Gradient Boosting Performs Gaussian Process Inference

Authors: Aleksei Ustimenko, Artem Beliakov, Liudmila Prokhorenkova

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments confirm that the proposed sampler from the Gaussian process posterior outperforms the previous approaches (Malinin et al., 2021) and gives better knowledge uncertainty estimates and improved out-of-domain detection.
Researcher Affiliation Collaboration Aleksei Ustimenko Share Chat research@aleksei.uk Artem Beliakov Yandex Research, HSE University belyakov.arteom2015@gmail.com Liudmila Prokhorenkova Yandex Research ostroumova-la@yandex.com
Pseudocode Yes Algorithm 1 Sample Tree(r; m, n, β), Algorithm 2 Train GBDT(z; ϵ, T, m, n, β, λ), Algorithm 3 Sample Prior(T, m, n), Algorithm 4 Sample Posterior(z; ϵ, T1, T0, m, n, β, σ, δ)
Open Source Code Yes The code of our experiments can be found on Git Hub.7 https://github.com/Take Over/Gradient-Boosting-Performs-Gaussian-Process-Inference
Open Datasets Yes For the experiments, we use several standard regression datasets (Gal & Ghahramani, 2016).
Dataset Splits Yes As a hyperparameter (that is estimated on the validation set), we consider β {10 2, 10 1, 1}., Here we perform cross-validation to estimate statistical significance with paired t-test and highlight the approaches that are insignificantly different from the best one (p-value > 0.05).
Hardware Specification No The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running the experiments. It only mentions 'Implementation details' in Appendix H, which does not include hardware specifications.
Software Dependencies No The paper mentions using 'the standard Cat Boost library' but does not specify a version number for it or any other software dependencies, which is required for a reproducible description.
Experiment Setup Yes For KGB, we fix ϵ = 0.3, T0 = 100, T1 = 900, σ = 10 2, δ = 10 4 β = 0.1, m = 4, n = 64, and sampled 100 KGB models., As a hyperparameter (that is estimated on the validation set), we consider β {10 2, 10 1, 1}., Finally, we set l2 leaf reg value to 0, as SGLB does.