Meta-Learning Reliable Priors in the Function Space

Authors: Jonas Rothfuss, Dominique Heyn, jinfan Chen, Andreas Krause

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In our benchmark study on meta-learning for Bayesian Optimization (BO), F-PACOH significantly outperforms all other meta-learners and standard baselines.
Researcher Affiliation Academia Jonas Rothfuss ETH Zurich jonas.rothfuss@inf.ethz.ch Dominique Heyn ETH Zurich heynd@student.ethz.ch Jinfan Chen ETH Zurich georgcjf@gmail.com Andreas Krause ETH Zurich krausea@ethz.ch
Pseudocode Yes Algorithm 1 F-PACOH-MAP: Meta-Learning Reliable Priors
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] See Appendix B.3.
Open Datasets Yes As practical application of meta-learning for BO, we consider hyper-parameter tuning of machine learning algorithms on different datasets... [67]. In these hyper-parameter experiments, we use the area under the ROC curve (AUROC) as test metric to optimize.
Dataset Splits Yes With this meta-training data D1,Ti, ..., D1,Tn, we meta-train F-PACOH and the considered baselines. To obtain statistically robust results, we perform independent BO runs on 10 unseen target functions / tasks and we repeat the whole meta-training & -testing process for 25 random seeds to initialize the meta-learner.
Hardware Specification Yes All experiments were performed on a cluster of Intel (R) Xeon (R) CPU E5-2698 v4 @ 2.20GHz machines with 40 CPU cores and 192 GB RAM.
Software Dependencies No The paper mentions software like 'GPyTorch [70]' and 'Adam optimizer [71]' but does not provide specific version numbers for these or other key software components, such as PyTorch itself, which would be needed for reproducible results.
Experiment Setup Yes Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] See Appendix B