Module 5 - Lecture 11 - Neurofuzzy systems

Question 1

Neuro-fuzzy systems

In a neuro-fuzzy Mamdani-type model, what algorithm is used?

Answer 1

Backpropagation.

Question 2

Neuro-fuzzy systems

In a neuro-fuzzy Mamdani-type model, what is learned?

Answer 2

Weights and I/O MFs

Question 3

Neuro-fuzzy systems

Is a neuro-fuzzy Mamdani-type model or ANFIS more complicated?

Answer 3

Mamdani

Question 4

Neuro-fuzzy systems

What are some synonyms for fuzzy inference systems?

Answer 4

• Fuzzy model
• Fuzzy associate memory (FAM)
• Fuzzy controller

Question 5

Neuro-fuzzy systems

What parts does a fuzzy inference system consist of?

Answer 5

See image

Question 6

Neuro-fuzzy systems

What is ANFIS short for?

Answer 6

Adaptive network-based FIS

Question 7

Neuro-fuzzy systems

In the image, describe what the circles and squares are.

Answer 7

• Squares: nodes with parameters
• Circles: nodes without parameters

Question 8

Neuro-fuzzy systems

What is the goal of an adaptive network?

Answer 8

To learn an I/O mapping specified by training dataset

Question 9

Neuro-fuzzy systems

What’s the basic idea when creating a neuro-fuzzy FIS?

Answer 9

Given I/O data pairs (training dataset) (x1, …, xn; y), construct a FIS to fit them.

Question 10

Neuro-fuzzy systems

What are the two stages of Fuzzy Modeling?

Answer 10

1) Structure identification — input selection, # of MFs
2) Parameter estimation — optimal parameters

Question 11

Neuro-fuzzy systems

Describe what happens in the highlighted part. (See image)

Answer 11

See image. The green box multiplies the input by a weight.

Question 12

Neuro-fuzzy systems

Describe what happens in the highlighted part. (See image)

Answer 12

See image. Summation of weighted inputs.

Question 13

Neuro-fuzzy systems

Describe what happens in the highlighted part. (See image)

Answer 13

See image. Summation of unweighted inputs.

Question 14

Neuro-fuzzy systems

Describe what happens in the highlighted part. (See image)

Answer 14

Calculating sum of centroids?
(Or weight normalization?)

Question 15

Neuro-fuzzy systems

How does ANFIS learn - what error is used?

Answer 15

Least-Squares Estimator (LSE)

Question 16

Neuro-fuzzy systems

How does ANFIS learn - what learning algorithm is used?

Answer 16

Gradient descent

Question 17

Neuro-fuzzy systems

With what methods does FIS partition area? (GTS)

Answer 17

Grid/Tree/Scatter partition

Question 18

Neuro-fuzzy systems

What is grid partitioning?

Answer 18

Each region is included in a square area ⇒ hypercube. See image.

Question 19

Neuro-fuzzy systems

What is tree partitioning?

Answer 19

Each region can be uniquely specified along a corresponding
decision tree.

Question 20

Neuro-fuzzy systems

What is scatter partitioning?

Answer 20

Each region is determined by covering a subset of the input space that characterizes a region of the possible occurrence of the input vectors.

Question 21

Neuro-fuzzy systems

How does the number of rules grow for grid partitioning?

Answer 21

If we have k inputs & m MFs for each ⇒ m^k rules!!

Question 22

Neuro-fuzzy systems

How does the number of rules grow for tree partitioning?

Answer 22

No exponential increase in # of rules.

Question 23

Neuro-fuzzy systems

What is this an example of?

Answer 23

Grid partitioning.

Question 24

Neuro-fuzzy systems

What is this an example of?

Answer 24

Tree partitioning

Question 25

Neuro-fuzzy systems

What is this an example of?

Answer 25

Scatter partitioning