Consistent Kernel Mean Estimation for Functions of Random Variables

Authors: Carl-Johann Simon-Gabriel, Adam Scibior, Ilya O. Tolstikhin, Bernhard Schölkopf

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We ran experiments on synthetic data to show how accurately ˆµ1, ˆµ2 and ˆµ3 approximate µf(X,Y ) with growing sample size N. We considered three basic arithmetic operations: multiplication X Y , division X/Y , and exponentiation XY , with X N(3; 0.5) and Y N(4; 0.5). As the true embedding µf(X,Y ) is unknown, we approximated it by a U-statistic estimator based on a large sample (125 points). For ˆµ3, we used the simplest possible reduced set method: we randomly sampled subsets of size n = 0.01 N of the xi, and optimized the weights wi and ui to best approximate ˆµX and ˆµY . The results are summarised in Figure 1 and corroborate our expectations: (i) all estimators converge, (ii) ˆµ2 converges fastest and has the lowest variance, and (iii) ˆµ3 is worse than ˆµ2, but much better than the diagonal estimator ˆµ1.
Researcher Affiliation Academia Department of Empirical Inference, Max Planck Institute for Intelligent Systems Spemanstraße 38, 72076 Tübingen, Germany joint first authors; also with: Engineering Department, Cambridge University
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not contain an explicit statement or link indicating the release of source code for the described methodology.
Open Datasets No We ran experiments on synthetic data to show how accurately ˆµ1, ˆµ2 and ˆµ3 approximate µf(X,Y ) with growing sample size N.
Dataset Splits No The paper mentions approximating the true embedding with a large sample, but does not specify train/validation/test splits for their own experiments. It states: 'As the true embedding µf(X,Y ) is unknown, we approximated it by a U-statistic estimator based on a large sample (125 points).'
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes For ˆµ3, we used the simplest possible reduced set method: we randomly sampled subsets of size n = 0.01 N of the xi, and optimized the weights wi and ui to best approximate ˆµX and ˆµY .