Introduction to

Radial Basis Function Networks

Content
● Overview
● The Models of Function Approximator
● The Radial Basis Function Networks
● RBFN’s for Function Approximation
● Learning the Kernels
Overview
Cover’s Theorem (1965)

(Based on the separability of patterns)

A complex pattern classification problem that is

nonlinearly separable in a low dimensional space, is
more likely to be linearly separable in a high
dimensional space.
Example - XOR Problem

φ1

φ2
Typical Applications of NN

● Pattern Classification

● Function Approximation

● Time-Series Forecasting
Function Approximation
Unknown

f
Approximator

ˆf
Supervised Learning

Unknown
Function
+
+−
Neural
Network
Neural Networks as
Universal Approximators
● Feedforward neural networks with a single hidden layer of
sigmoidal units are capable of approximating uniformly any
continuous multivariate function, to any desired degree of
accuracy.
– Hornik, K., Stinchcombe, M., and White, H. (1989). "Multilayer
Feedforward Networks are Universal Approximators," Neural Networks,
2(5), 359-366.
● Like feedforward neural networks with a single hidden layer
of sigmoidal units, it can be shown that RBF networks are
universal approximators.
– Park, J. and Sandberg, I. W. (1991). "Universal Approximation Using
Radial-Basis-Function Networks," Neural Computation, 3(2), 246-257.
– Park, J. and Sandberg, I. W. (1993). "Approximation and
Radial-Basis-Function Networks," Neural Computation, 5(2), 305-316.
A feedforward neural network with one hidden layer using sigmoidal activation functions (like
the logistic sigmoid or tanh) can approximate any continuous function with arbitrary accuracy,
given enough neurons in the hidden layer.
This means that if we have a function () that maps inputs to outputs, we can construct a
neural network with a single hidden layer that closely approximates ()
The key insight is that neural networks can learn complex relationships even with a single
hidden layer, as long as they have enough neurons.
Why does this matter?
It proves that neural networks have great flexibility and can model almost any real-world
function.
This result laid the foundation for why deep learning is so powerful—because neural networks
can represent almost any function.

The Model of
Function Approximator
 Fixed
Basis We
Function igh
s ts
Linear Models
Linear Models
y
Linearly
Output
weighted
Units
output

w1 w2 wm
• Decomposition
Hidden φ φ φ • Feature Extraction
Units 1 2 m • Transformation

Inputs Feature Vectors

x = x1 x2 xn
Linear Models s e s?
e b a
y Can yo u say som
Linearly
Output
weighted
Units
output

w1 w2 wm
• Decomposition
Hidden φ φ φ • Feature Extraction
Units 1 2 m • Transformation

Inputs Feature Vectors

x = x1 x2 xn
Example Linear Models
o n a l b a s e s?
t h e y o r t h og
● Polynomial Are

● Fourier Series
Single-Layer Perceptrons as
Universal Aproximators
y

w1 w2 wm
With sufficient number of
Hidden φ φ φ sigmoidal units, it can be a
Units 1 2 m universal approximator.

x = x1 x2 xn
Radial Basis Function Networks as
Universal Aproximators
y

w1 w2 wm With sufficient number of

Hidden φ φ φ radial-basis-function units,
Units it can also be a universal
1 2 m
approximator.

x = x1 x2 xn
Adjusted
by the We
Learning igh
process ts
Non-Linear Models
The Radial Basis
Function Networks
Radial Basis Function Networks

Radial Basis Functions are feed-forward networks consisting of

❖A hidden layer of radial kernels and
❖An output layer of linear neurons
The two layers in an RBF carry entirely different roles [Haykin, 1999]
❖The hidden layer performs a non-linear transformation of input space
❖The resulting hidden space is typically of higher dimensionality than the
input space
❖The output layer performs linear regression to predict the desired targets
 As a function
approximator

The Topology of RBF

y1 ym

Output Interpolation
Units

Hidden Projection
Units

Inputs Feature Vectors

x1 x2 xn
 As a pattern classifier.

The Topology of RBF

y1 ym

Output
Units Classes

Hidden
Subclasses
Units

Inputs Feature Vectors

x1 x2 xn
Network Parameters

● : The radial basis function for the hidden layer.

This is a simple nonlinear mapping function (typically Gaussian) that
transforms the d- dimensional input patterns to a (typically higher)
H-dimensional space. The complex decision boundary will be
constructed from linear combinations (weighted sums) of these simple
building blocks.

● : The weights joining the first to hidden layer. These weights

constitute the center points of the radial basis functions. Also called
prototypes of data.

22
Network Parameters

● : The spread constant(s). These values determine the spread (extend)

of each radial basis function.

● : The weights joining hidden and output layers. These are the
weights which are used in obtaining the linear combination of the radial
basis functions. They determine the relative amplitudes of the RBFs when
they are combined to form the complex function.

● : the Euclidean distance between the input and the prototype

vector . Activation of the hidden unit is determined according to this
distance through .
Typical Radial Functions
● Gaussian

● Hardy Multiquadratic

● Inverse Multiquadratic
Gaussian Basis Function (σ=0.5,1.0,1.5)
Inverse Multiquadratic

c=5
c=4
c=3
c=2
c=1
RBFN’s for
Function Approximation
The idea
y Unknown Function
to Approximate

Training
Data

x
The idea
y Unknown Function
to Approximate

Training
Data

x
Basis Functions (Kernels)
The idea
Function
y Learned

x
Basis Functions (Kernels)
The idea
Nontraining
Sample Function
y Learned

x
Basis Functions (Kernels)
The idea
Nontraining
Sample Function
y Learned

x
Radial Basis Function Networks as
Universal Aproximators

Training set

Goal for all k

w1 w2 wm

x = x1 x2 xn
Learn the Optimal Weight Vector

Training set

Goal for all k

w1 w2 wm

x = x1 x2 xn
Learning the Kernels
 How to Train?
.
.
.
.
.

There are various approaches for training RBF networks.

Approach 1: Exact RBF – Guarantees correct classification of all
training data instances. Requires N hidden layer nodes, one for each
training instance. No iterative training is involved. RBF centers (u) are
fixed as training data points, spread as variance of the data, and w are
obtained by solving a set of linear equations
Approach 2: Fixed centers selected at random. Uses H<N hidden
layer nodes. No iterative training is involved. Spread is based on Euclidean
metrics, w are obtained by solving a set of linear equations.
Approach 3: Centers are obtained from unsupervised learning
(clustering). Spreads are obtained as variances of clusters, w are obtained
through LMS algorithm. Clustering (k-means) and LMS are iterative. This
is the most commonly used procedure. Typically provides good results.
Approach 4: All unknowns are obtained from supervised learning.
 Approach 1
.
.
.
.
.

Exact RBF
The first layer weights u are set to the training data; U=XT. That is the
gaussians are centered at the training data instances.
The spread is chosen as , where dmax is the maximum Euclidean
distance between any two centers, and N is the number of training data
points. Note that H=N, for this case.
The output of the kth RBF output neuron is then

Multiple Single output

outputs

During training, we want the outputs to be equal to our desired targets.

Without loss of any generality, assume that we are approximating a single
dimensional function, and let the unknown true function be f(x). The
desired output for each input is then di=f(xi), i=1, 2, …, N.
.
.
.
Approach 1
.
. (Cont.)

We then have a set of linear equations, which can be represented in

the matrix form:

Define:

Is this matrix always invertible?

.
.
.
Approach 1
.
. (Cont.)

Michelli’s Theorem (1986)

If {xi}iN=1 are a distinct set of points in the d-dimensional space, then the
N by N interpolation matrix Φ with elements obtained from radial basis
functions is nonsingular, and hence can be inverted!

Note that the theorem is valid regardless the value of N, the choice of the
RBF (as long as it is an RBF), or what the data points may be, as long as
they are distinct!
.
.
.
Approach1
.
. (Cont.)

The Gaussian is the most commonly used RBF (why…?).

Note that

Gaussian RBFs are localized functions ! unlike the sigmoids used by MLPs

Using Gaussian radial basis Using sigmoidal radial basis

functions functions
 Exact RBF Properties
.
.
.
.
.

Using localized functions typically makes RBF networks more suitable for
function approximation problems.
Since first layer weights are set to input patterns, second layer weights are
obtained from solving linear equations, and spread is computed from the
data, no iterative training is involved !!!
Guaranteed to correctly classify all training data points!
However, since we are using as many receptive fields as the number of
data, the solution is over determined, if the underlying physical process
does not have as many degrees of freedom 🡺 Overfitting!
The importance of σ: Too small will
also cause overfitting. Too large will
fail to characterize rapid changes in
the signal.
.
.
Too many
Receptive Fields?
.
.
.

In order to reduce the artificial complexity of the RBF, we need to

use fewer number of receptive fields.
How about using a subset of training data, say M < N of them.
These M data points will then constitute M receptive field centers.
How to choose these M points…?
At random 🡺 Approach 2.

Output layer weights are determined as they were in Approach 1, through

solving a set of M linear equations!
Unsupervised training: K-means 🡺 Approach 3
The centers are selected through self organization of clusters, where the
data is more densely populated. Determining M is usually heuristic.
.
. Approach 3
K-Means - Unsupervised
Clustering - Algorithm
.
.
.

Choose number of clusters, M

Initialize M cluster centers to the first M training data points: tk=xk, k=1,2,…,M.
Repeat
At iteration n, group all patterns to the cluster whose center is closest

tk(n): center of kth RBF at

nth iteration
Compute the centers of all clusters after the regrouping

New cluster center Instances that are grouped

for kth RBF. in the kth cluster
Number of instances
in the kth cluster
Until there is no change in cluster centers from one iteration to the next.
.
Determining the Output Weights:
.
.
.
Approach 3 LMS Algorithm
.

The LMS algorithm is used to minimize the cost function where

e(n) is the error at iteration n:

Using the steepest (gradient) descent method:

Instance based LMS algorithm pseudocode (for single output):

Initialize weights, wj to some small random value, j=1,2,…,M
Repeat
Choose next training pair (x, d);
Compute network output at iteration n:
Compute error:
Update weights:
Until weights converge to a steady set of values
.
. Approach 4:
Supervised
.
.
.
RBF Training
This is the most general form.
All parameters, receptive field centers (first layer weights), output layer weights and
spread constants, are learned through iterative supervised training using LMS /
gradient descent algorithm.
E

G’ represents the first derivative

of the function wrt its argument