Spatially Aggregated Gaussian Processes with Multivariate Areal Outputs
Authors: Yusuke Tanaka, Toshiyuki Tanaka, Tomoharu Iwata, Takeshi Kurashima, Maya Okawa, Yasunori Akagi, Hiroyuki Toda
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments on real-world data sets demonstrate that our model can 1) accurately refine coarse-grained areal data, and 2) offer performance improvements by using the areal data sets from multiple domains. |
| Researcher Affiliation | Collaboration | Yusuke Tanaka1,3, Toshiyuki Tanaka3, Tomoharu Iwata2, Takeshi Kurashima1, Maya Okawa1, Yasunori Akagi1, Hiroyuki Toda1 1NTT Service Evolution Labs., 2NTT Communication Science Labs., 3Kyoto University |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its own source code or a direct link to a repository containing the implementation of the described methodology. |
| Open Datasets | Yes | We evaluated SAGP using 10 and 3 real-world areal data sets from two cities, New York City and Chicago, respectively. They were obtained from NYC Open Data 2 and Chicago Data Portal 3. 2https://opendata.cityofnewyork.us 3https://data.citychicago.org/ |
| Dataset Splits | Yes | The number L of the latent GPs was chosen from {1, . . . , S} via leave-one-out cross-validation [1]; the validation error was obtained using each held-out coarse-grained data value. |
| Hardware Specification | Yes | The average computation times for inference were 1728.2 and 115.1 seconds for the data sets from New York City and Chicago, respectively; the experiments were conducted on a 3.1 GHz Intel Core i7. |
| Software Dependencies | No | The paper mentions the use of 'Sci Py' for the L-BFGS method but does not provide a specific version number for this or any other software dependency. |
| Experiment Setup | Yes | In our experiments, we used zero-mean Gaussian processes as the latent GPs {gl(x)}L l=1, i.e., l(x) = 0 for l = 1, . . . , L. We used the following squared-exponential kernel as the covariance function for the latent GPs, γl(x, x0) = 2 l exp( kx x0k2/2β2 l )... Here, we set 2 l = 1 because the variance can already be represented by scaling the columns of W. For simplicity, the covariance function for the Gaussian noise process n(x, x0) is set to (x, x0) = diag(λ2 1δ(x x0), . . . , λ2 Sδ(x x0)), where δ( ) is Dirac s delta function. The model parameters, W, {λs}S s=1, , {βl}L l=1, were learned by maximizing the logarithm of the marginal likelihood (8) or (12) using the L-BFGS method [15] implemented in Sci Py (https://www.scipy.org/). For approximating the integral over regions (see (18)), we divided a total region of each city into sufficiently fine-grained square grid cells, the size of which was 300 m 300 m for both cities... The number L of the latent GPs was chosen from {1, . . . , S} via leave-one-out cross-validation... |