Constrained Local Latent Variable Discovery

Authors: Tian Gao, Qiang Ji

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the performance of our method on synthetic data to demonstrate its effectiveness in identifying the correct latent variables. We further apply our algorithm to feature discovery and selection problem, and show that the latent variables learned through the proposed method can improve the classification accuracy in benchmark feature selection and discovery datasets. ... We first demonstrate the performance of the proposed methods on synthetic datasets, and then apply our algorithm to benchmark feature selection and discovery datasets.
Researcher Affiliation Collaboration Tian Gao Rensselaer Polytechnic Institute, Troy NY and IBM Watson Research Center, Yorktown Heights NY tgao@us.ibm.com Qiang Ji IBM Watson Research Center, Yorktown Heights NY jiq@rpi.edu
Pseudocode Yes Algorithm 1 Latent MB Learning with Constrained Structure EM(LMB-CSEM) Algorithm... Algorithm 2 CSEM, the Constrained Structure EM(CSEM) Subroutine for Latent MB Learning
Open Source Code No The paper does not provide concrete access to source code, nor does it make any explicit statements about its release for the methodology described.
Open Datasets No The paper mentions using 'five feature selection datasets from the UCI machine learning repository and related works [Brown et al., 2012]' but does not provide specific links, DOIs, or formal citations for individual datasets to ensure public access.
Dataset Splits No The paper states 'We use half the data size for training and half for testing,' which indicates a train/test split but does not specify a validation set or explicit percentages for all three splits.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions using 'linear SVM classifiers' but does not specify any particular software, libraries, or programming language versions with explicit version numbers (e.g., Python 3.x, scikit-learn 0.xx, specific SVM implementation) that are necessary to replicate the experiment.
Experiment Setup Yes We fix the cardinality of latent variables to be 2 in all experiments. ... We run the proposed Algorithm 1 LMB-CSEM to find L = 20 different latent MB sets. ... We learn only one latent feature per latent MB set in LMB-CSEM. ... We also set the number of different initializations I in Algorithm 1 to be 80. ... we choose to use the Bi C score to learn different MB sets within each subspace, and the training errors as score(MB) to compare different MB sets across subspaces and from different initializations, which performs best compared to the mutual information and conditional likelihood of the target label.