Sequential Kernelized Independence Testing
Authors: Aleksandr Podkopaev, Patrick Blöbaum, Shiva Kasiviswanathan, Aaditya Ramdas
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the power of our approaches on both simulated and real data. (From Abstract) and Synthetic Experiments. To compare SKITs based on HSIC, COCO, and KCC payoffs, we use RBF kernel with hyperparameters taken to be inversely proportional to the second moment of the underlying variables; we observed no substantial difference when such selection is data-driven (median heuristic). (From Section 3) |
| Researcher Affiliation | Collaboration | 1Statistics & Data Science and Machine Learning Departments, Carnegie Mellon University 2Amazon Research. |
| Pseudocode | Yes | Algorithm 1 Online Newton step (ONS) strategy for selecting betting fractions (and Algorithms 2, 3, 4). |
| Open Source Code | No | The paper does not provide any explicit statement about releasing source code or a link to a code repository for the described methodology. |
| Open Datasets | Yes | We analyze average daily temperatures in four European cities: London, Amsterdam, Zurich, and Nice, from January 1, 2017, to May 31, 2022. (From Section 4, footnote indicates data source: https://www.wunderground.com) and This experiment is based on MNIST dataset (Le Cun et al., 1998) where pairs of digits are observed at each step; under the null one sees digits (a, b) where a and b are uniformly randomly chosen, but under the alternative one sees (a, a ), i.e., two different images of the same digit. (From Appendix E.6) |
| Dataset Splits | No | The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) for training, validation, and testing. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper mentions software like PyTorch implicitly for kernel implementations, but it does not specify any software dependencies with version numbers. |
| Experiment Setup | Yes | We use RBF kernel with hyperparameters taken to be inversely proportional to the second moment of the underlying variables; we observed no substantial difference when such selection is data-driven (median heuristic). We consider settings where the complexity of a task is controlled through a single univariate parameter: (a) Gaussian model. For t 1, we consider Yt = Xtβ + εt, where Xt, εt N(0, 1). We have that β = 0 implies nonzero linear correlation (hence dependence). We consider 20 values for β, spaced uniformly in [0,0.3], and use λX = 1/4 and λY = 1/(4(1+β2)) as kernel hyperparameters. (From Section 3) and We select the kernel hyperparameters via the median heuristic using recordings for the first 20 days. (From Section 4) and Algorithm 5 a GRAPA strategy for selecting betting fractions Input: sequence of payoffs (ft(Z2t 1, Z2t))t 1, λa GRAPA 1 = 0, µ(1) 0 = 0, µ(2) 0 = 1, c = 0.9. (From Appendix C) |