Understanding the data

FSM, Feature Space Mapping neurofuzzy network

Method based on FSM (Feature Space Mapping) neurofuzzy network.
Crisp rules: FSM + rectangular transfer functions.
Fuzzy rules: FSM + context-dependent fuzzy membership functions.

Transfer function

Examples of transfer function

Gauss function

Rectangular function

Bicentral functions - soft trapezoidal functions

New node conditions

Adaptation of parameters

Simple application of FSM network: logical rules for the Iris problem

FSM network with rectangular transfer function

R1:

C4

-4.89 Iris_setosa

+0.61 Iris_setosa

R2

C3 C4

0.66 0.65 Iris_versicolor

4.90 1.51 Iris_versicolor

5 incorrect classifications

Spiral data

FSM network with Gaussian functions, 53 neurons

FSM network with Gaussian functions, rotations enabled, 59 neurons

Other applications of FSM: as neural network, neurofuzzy system, prototype-based system or heuristics for search- based reasoning.

Example: any law of the form A=B*C or A=B+C, here Ohm's law V=I*R, has 13 true facts, 14 false facts.

Introduction