Gaussian process nonparametric tensor estimator and its minimax optimality
Authors: Heishiro Kanagawa, Taiji Suzuki, Hayato Kobayashi, Nobuyuki Shimizu, Yukihiro Tagami
ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We apply our proposed method to multi-task learning and show that our method significantly outperforms existing methods through numerical experiments on real-world data sets. |
| Researcher Affiliation | Collaboration | Heishiro Kanagawa KANAGAWAH.AB@M.TITECH.AC.JP Taiji Suzuki , SUZUKI.T.CT@M.TITECH.AC.JP Tokyo Institute of Technology, Tokyo 152-8552, JAPAN PRESTO, Japan Science and Technological Agency (JST), JAPAN Hayato Kobayashi HAKOBAYA@YAHOO-CORP.JP Nobuyuki Shimizu NOBUSHIM@YAHOO-CORP.JP Yukihiro Tagami YUTAGAMI@YAHOO-CORP.JP Yahoo Japan Corporation, Tokyo 107-6211, JAPAN |
| Pseudocode | No | The paper describes algorithmic steps but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | Restaurant data set (Blanca et al., 2011) and the School data set (Goldstein, 1991). |
| Dataset Splits | No | The paper mentions 'minimizing the validation error' but does not specify the exact percentages or counts for training, validation, and test splits needed for reproducibility. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment. |
| Experiment Setup | Yes | The tensor rank for ALS and GP was fixed d = 3 in both data sets. For our Gaussian process method, we employed the linear kernel (linear) and the GRBF kernel (GRBF) as the kernel function for the GP on fr. We also tested a mixture of them, i.e., some of three kernels for fr were the linear and the rest were the GRBF. We did this with the number of the linear kernels 1 and 2 (indicated by GRBF(2)+lin(1) and GRBF(1)+lin(2) respectively). The kernel width σ for the GRBF kernel was set at 100, and the delta kernel was chosen for the kernel functions for Task 1 (restaurant) and Task 2 (aspect). |