AFS: An Attention-Based Mechanism for Supervised Feature Selection

Authors: Ning Gui, Danni Ge, Ziyin Hu3705-3713

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that AFS achieves the best accuracy and stability in comparison to several state-of-art feature selection algorithms upon both MNIST, noisy MNIST and several datasets with small samples. A set of experiments are designed on both Large-dimensionality Small-instance dataset (denoted as L/S dataset) and Medium/large-dimensionality Large-instance dataset (short for M/L dataset).
Researcher Affiliation Academia 1School of Software, Central South University, China, ninggui@gmail.com 2 School of Informatics, Zhejiang Sci-Tech University, China, dongfangdecheng@gmail.com, fayssica@gmail.com
Pseudocode No The paper describes algorithms and includes mathematical equations but does not present a structured pseudocode block or algorithm labeled explicitly as such.
Open Source Code Yes Source code at https://github.com/upup123/AAAI-2019-AFS.
Open Datasets Yes The MNIST dataset2 consists of greyscale thumbnails, 28 x 28 pixels, of handwritten digits 0 to 9. The n-MNIST dataset3 consists of three MNIST variants... The Lung_discrete and Isolet4 are L/S datasets. 2 http://yann.lecun.com/exdb/mnist/ 3 http://csc.lsu.edu/~saikat/n-mnist/ 4 http://featureselection.asu.edu
Dataset Splits Yes For L/S datasets, 3 times 3 fold cross-validation is adopted to provide a fair comparison.
Hardware Specification No The paper does not specify the hardware used for experiments (e.g., specific CPU/GPU models, memory, or cluster configurations).
Software Dependencies No The paper mentions "scikit-feature selection repository" but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes Model parameters are initialized with truncated normal distribution with a mean of 0 and standard deviation of 0.1. The model is optimized by Adam, with batch size 100. The weight of regularizer is 0.0001. For the solution of AFS and Roy, the training step is both set to 3000 as they both begin to converge.