Sparse meta-Gaussian information bottleneck

Authors: Melani Rey, Volker Roth, Thomas Fuchs

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

Reproducibility Variable Result LLM Response
Research Type Experimental 4. Experiments Simulation: Comparison between different IB methods. We assess the efficiency of the different compression matrices A obtained by the above methods on a test set with 5000 observations. Each experiment is repeated to obtain the 50 curves for each method (shown in the top panel of Figure 2). 4.1. Real data
Researcher Affiliation Collaboration M elanie Rey MELANIE.REY@UNIBAS.CH University of Basel, Basel, Switzerland Thomas J. Fuchs FUCHS@CALTECH.EDU Jet Propulsion Laboratory, California Institute of Technology, Pasadena, USA Volker Roth VOLKER.ROTH@UNIBAS.CH University of Basel, Basel, Switzerland
Pseudocode Yes Algorithm 1 Optimisation of sparse MGIB
Open Source Code Yes A Matlab code for this method is available online1. Our algorithm (available in the supplementary material along with our R implementation)
Open Datasets Yes We generate training samples with n = 1000 observations (xi, yi), i = 1, . . . , n and dimensions fixed to p = 15, q = 15. Data was available in the form of immunohistochemical (IHC) expressions of 70 candidate biomarkers measured for 364 patients. A first promising approach to identify biomarkers important for survival prediction was reported in Meyer et al. (2012).
Dataset Splits No The paper mentions training and test sets but does not provide specific details on a validation split (percentages, counts, or explicit standard split references) for reproduction.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions 'Matlab code' and 'R implementation' but does not specify version numbers for these software environments or any libraries used within them.
Experiment Setup Yes We generate training samples with n = 1000 observations (xi, yi)... and dimensions fixed to p = 15, q = 15. By varying the parameter κ between 0.1 and 80 we can represent I(Y ; T) as a function of I(X; T) and obtain the information curves. Data was available in the form of immunohistochemical (IHC) expressions of 70 candidate biomarkers measured for 364 patients.