DARBO: Automatic discovery of unconventional fuzzy rules: fuzzy temporal knowledge bases and variable structure TSK

This project seeks the implementation of evolutionary computation techniques for automatic learning of two categories of fuzzy rules that formalize knowledge through syntactic rather complex propositions (unconventional):

  1. Fuzzy temporal rules (FTRs), including spatial and temporal references between propositions that make the rules.
  2. Rules
  3. Takagi-Sugeno-Kang of variable structure (TSK-EV) that combine regression expressions with knowledge explicit.

Each rule has uses differentiated: RTRs are applied to the automatic design of behaviors for a mobile robot from the distance data measured directly by its laser sensors, and TSK-EV rules on the estimation of time in industrial production of the furniture sector and PET image preprocessing.

Objectives

  • Study of new models of operators to build complex rules that handle uncertainty, including fuzzy quantifiers. Applying them to build linguistic summaries of data sets.
  • Machine learning by evolutionary computation (genetic algorithms) of fuzzy temporal rules for classification problems on sensor data (dimensionality reduction and automatic clustering thereof).
  • Automatic construction of behaviors on mobile robot from simple data: "tracking" of people and contour following.
  • Machine learning by evolutionary computation (algorithms genetic) of TSK-EV rules for regression problems on approximate data and low quality (with uncertainty)