Neural-Kernel Conditional Mean Embeddings

Authors: Eiki Shimizu, Kenji Fukumizu, Dino Sejdinovic

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To investigate the effectiveness of our approach, we conduct experiments on both toy and real-world datasets.
Researcher Affiliation Academia 1Department of Statistical Science, Graduate University of Advanced Studies (SOKENDAI), Tokyo, Japan 2The Institute of Statistical Mathematics, Tokyo, Japan 3School of Computer and Mathematical Sciences, The University of Adelaide, Australia.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks (clearly labeled algorithm sections or code-like formatted procedures).
Open Source Code Yes The implemented code can be found at https://github.com/tokorotenten/Neural-Kernel.
Open Datasets Yes To further investigate our approaches, we conduct experiments on 8 real-world regression benchmark datasets from the UCI repository. Details of the datasets are provided in Appendix F. ... Dua, D. and Graff, C. Uci machine learning repository, 2017. URL http://archive.ics.uci.edu/ml.
Dataset Splits Yes We follow experimental protocols of Han et al. (2022): (a) we employ the same train-test splits with a 90%/10% ratio, and use 20 folds for all datasets except Protein (5 folds) and Year (1 fold)
Hardware Specification Yes The experiment was conducted on Mac Book Pro with M2 system, using only a CPU.
Software Dependencies No The paper mentions software such as "Sci Py (Virtanen et al., 2020)" and "Gymnasium (Towers et al., 2023)" but does not consistently provide specific version numbers for multiple key software components or specialized packages used in the experiments.
Experiment Setup Yes We used NNs with two fully-connected hidden layers, each containing 50 Re LU activation units. For the optimizer, we used Adam W (Loshchilov & Hutter, 2019). Other architectural and hyperparameter choices for each model are provided below: Proposals: We set the number of location points M = 100, and ηa were chosen as uniformly spaced grid points within the closed interval bounded by the minimum and maximum values observed in the training data. The learning rate was set to 1e-4, the batch size was set to 50, and the number of training epochs was set to 1000, and σ was initialize to 1.0.