TSK fuzzy system: pyTorch Version

This code is an implementation of the paper "Optimize TSK Fuzzy Systems for Classification Problems: Mini-Batch Gradient Descent with Uniform Regularization and Batch Normalization"

This paper proposes a mini- batch gradient descent (MBGD) based algorithm to efficiently and effectively train TSK fuzzy classifiers. It integrates two novel techniques: 1) uniform regularization (UR), which forces the rules to have similar average contributions to the output, and hence to increase the generalization performance of the TSK model; and, 2) batch normalization (BN), which extends BN from deep neural networks to TSK fuzzy classifiers to expedite the convergence and improve the generalization performance.

Run

./run.sh

File structure

• main_diff_data_split.py Sample code for running TSK-BN-UR
• lib.models.py Sample code for constructing TSK with BN and UR in pyTorch
• lib.tuning_training.py Sample code for training TSK fuzzy systems

Citing

This paper is under review, we will publish the bib text for citing as soon as it's accpeted.