Discriminative Features via Generalized Eigenvectors

Authors: Nikos Karampatziakis, Paul Mineiro

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate classifiers built from these features on three different tasks, obtaining state of the art results. and 4. Experiments
Researcher Affiliation Industry Nikos Karampatziakis NIKOSK@MICROSOFT.COM Paul Mineiro PMINEIRO@MICROSOFT.COM Microsoft CISL, 1 Microsoft Way, Redmond, WA 99 98052 USA
Pseudocode Yes Algorithm 1 Generalized Eigenvectors for Multiclass Require: S = {(xi, yi)}n i=1, θ 0 and γ 0 1: F 2: for (i, j = i) {1, . . . , k}2 do 3: Solve ˆCi V = ( ˆCj + γ d Trace( ˆCj)I)V Λ 4: F F {Vq|Λqq θ} 5: end for 6: ψv,α,δ(x) .= max(0, δv x)α/2 7: φ(x) .= [ψv,α,δ(x)|v, α, δ F {1, 2, 3} { 1, 1}] 8: w = Multi Logit({(φ(x), y)|(x, y) S})
Open Source Code No The paper does not contain any explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We begin with the MNIST database of handwritten digits (Le Cun et al., 1998), for which we can visualize the generalized eigenvectors, providing intuition regarding the discriminative nature of the computed directions. and Covertype is a multiclass data set whose task is to predict one of 7 forest cover types using 54 cartographic variables (Blackard & Dean, 1999). and TIMIT is a corpus of phonemically and lexically annotated speech of English speakers of multiple genders and dialects (Fisher et al., 1986).
Dataset Splits Yes To determine the hyperparameter settings γ and θ, we held out a fraction of the training set for validation. Once γ and θ were determined, we trained on the entire training set. and We use the standard training, development, and test sets of TIMIT.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions software like LAPACK and MATLAB, but it does not specify any version numbers for these or any other software dependencies needed to reproduce the experiments.
Experiment Setup No The paper mentions that hyperparameters γ and θ were determined using a validation set and that bandwidth and number of cosines were optimized for RFF. It also describes methods for selecting class pairs (e.g., random hypercube for TIMIT). However, it does not provide the specific numerical values for these hyperparameters in the text.