Skip to content

Differentiable neural architecture search method for tabular data

License

Notifications You must be signed in to change notification settings

EricssonResearch/tabgns

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

TabGNS: Gated Neuron Selection for tabular data

TabGNS is a neural arhcitecture search method for multi layer perceptrons. It uses gradient descent to decide on a models architecture at level of neurons. The paper describing the method will be published at NOMS 2026.

This repository contains an implementation of TabGNS. The method can be run on simple task using eval.py. For comparison, the script trains two separete models; TabGNS and the largest MLP within the search. To run TabGNS using hyperparameters used in the paper, use configurations in conf/paper.yaml instead. This can be done by running python eval.py --config-name=paper.

An example use of TabGNS is provided in nb.ipynb.

About

Differentiable neural architecture search method for tabular data

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published