Kernel Mean Estimation via Spectral Filtering

Authors: Krikamol Muandet, Bharath Sriperumbudur, Bernhard Schölkopf

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The paper includes a dedicated section '5 Empirical studies' which describes experiments on synthetic and real data, presenting results in figures and tables, and discussing performance metrics like risk and negative log-likelihood, thus indicating experimental research.
Researcher Affiliation Academia Krikamol Muandet MPI-IS, T ubingen krikamol@tue.mpg.de Bharath Sriperumbudur Dept. of Statistics, PSU bks18@psu.edu Bernhard Sch olkopf MPI-IS, T ubingen bs@tue.mpg.de
Pseudocode Yes Table 1: Update equations for β and corresponding filter functions.
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the described methodology.
Open Datasets Yes The datasets were taken from the UCI repository1 and pre-processed by standardizing each feature. 1http://archive.ics.uci.edu/ml/
Dataset Splits Yes For these two algorithms, we evaluate the leave-one-out score and select βt at the iteration t that minimizes this score (see, e.g., Figure 3(a)).
Hardware Specification No The paper does not provide specific hardware details (e.g., 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 with version numbers.
Experiment Setup Yes To allow for an analytic calculation of the loss L(β, X, P), we assume that the distribution P is a d-dimensional mixture of Gaussians [1, 8]. Specifically, the data are generated as follows: x P4 i=1 πi N(θi, Σi)+ε, θij U( 10, 10), Σi W(3 Id, 7), ε N(0, 0.2 Id) where U(a, b) and W(Σ0, df) are the uniform distribution and Wishart distribution, respectively. As in [1], we set π = [0.05, 0.3, 0.4, 0.25]. Throughout, we use the Gaussian RBF kernel k(x, x ) = exp( x x 2/2σ2) whose bandwidth parameter is calculated using the median heuristic, i.e., σ2 = median{ xi xj 2}. The parameters (πj, θj, σ2 j ) are initialized by the best one obtained from the K-means algorithm with 50 initializations. Throughout, we set r = 5 and use 25% of each dataset as a test set.