121st  Top numerical analysis topics 
A radial basis function network is an artificial neural network that uses radial basis functions as activation functions. It is a linear combination of radial basis functions. They are used in function approximation, time series prediction, and control.
Contents 
Radial basis function (RBF) networks typically have three layers: an input layer, a hidden layer with a nonlinear RBF activation function and a linear output layer. The output, , of the network is thus
where N is the number of neurons in the hidden layer, is the center vector for neuron i, and a_{i} are the weights of the linear output neuron. In the basic form all inputs are connected to each hidden neuron. The norm is typically taken to be the Euclidean distance and the basis function is taken to be Gaussian
The Gaussian basis functions are local in the sense that
i.e. changing parameters of one neuron has only a small effect for input values that are far away from the center of that neuron.
RBF networks are universal approximators on a compact subset of . This means that a RBF network with enough hidden neurons can approximate any continuous function with arbitrary precision.
The weights a_{i}, , and β are determined in a manner that optimizes the fit between and the data.
In addition to the above unnormalized architecture, RBF networks can be normalized. In this case the mapping is
where
is known as a "normalized radial basis function".
There is theoretical justification for this architecture in the case of stochastic data flow. Assume a stochastic kernel approximation for the joint probability density
where the weights and e_{i} are exemplars from the data and we require the kernels to be normalized
and
The probability densities in the input and output spaces are
and
The expectation of y given an input is
where
is the conditional probability of y given . The conditional probability is related to the joint probability through Bayes theorem
which yields
This becomes
when the integrations are performed.
It is sometimes convenient to expand the architecture to include local linear models. In that case the architectures become, to first order,
and
in the unnormalized and normalized cases, respectively. Here are weights to be determined. Higher order linear terms are also possible.
This result can be written
where
and
in the unnormalized case and
in the normalized case.
Here δ_{ij} is a Kronecker delta function defined as
In a RBF network there are three types of parameters that need to be chosen to adapt the network for a particular task: the center vectors , the output weights w_{i}, and the RBF width parameters β_{i}. In the sequential training of the weights are updated at each time step as data streams in.
For some tasks it makes sense to define an objective function and select the parameter values that minimize its value. The most common objective function is the least squares function
where
We have explicitly included the dependence on the weights. Minimization of the least squares objective function by optimal choice of weights optimizes accuracy of fit.
There are occasions in which multiple objectives, such as smoothness as well as accuracy, must be optimized. In that case it is useful to optimize a regularized objective function such as
where
and
where optimization of S maximizes smoothness and λ is known as a regularization parameter.
RBF networks can be used to interpolate a function when the values of that function are known on finite number of points: . Taking the known points to be the centers of the radial basis functions and evaluating the values of the basis functions at the same points the weights can be solved from the equation
It can be shown that the interpolation matrix in the above equation is nonsingular, if the points are distinct, and thus the weights w can be solved by simple linear algebra:
If the purpose is not to perform strict interpolation but instead more general function approximation or classification the optimization is somewhat more complex because there is no obvious choice for the centers. The training is typically done in two phases first fixing the width and centers and then the weights. This can be justified by considering the different nature of the nonlinear hidden neurons versus the linear output neuron.
Basis function centers can be randomly sampled among the input instances or obtained by Orthogonal Least Square Learning Algorithm or found by clustering the samples and choosing the cluster means as the centers.
The RBF widths are usually all fixed to same value which is proportional to the maximum distance between the chosen centers.
After the centers c_{i} have been fixed, the weights that minimize the error at the output are computed with a linear pseudoinverse solution:
where the entries of G are the values of the radial basis functions evaluated at the points x_{i}: g_{ji} = ρ(   x_{j} − c_{i}   ).
The existence of this linear solution means that unlike MultiLayer Perceptron (MLP) networks the RBF networks have a unique local minimum (when the centers are fixed).
Another possible training algorithm is gradient descent. In gradient descent training, the weights are adjusted at each time step by moving them in a direction opposite from the gradient of the objective function (thus allowing the minimum of the objective function to be found),
where ν is a "learning parameter."
For the case of training the linear weights, a_{i}, the algorithm becomes
in the unnormalized case and
in the normalized case.
For locallineararchitectures gradientdescent training is
For the case of training the linear weights, a_{i} and e_{ij}, the algorithm becomes
in the unnormalized case and
in the normalized case and
in the locallinear case.
For one basis function, projection operator training reduces to Newton's method.
The basic properties of radial basis functions can be illustrated with a simple mathematical map, the logistic map, which maps the unit interval onto itself. It can be used to generate a convenient prototype data stream. The logistic map can be used to explore function approximation, time series prediction, and control theory. The map originated from the field of population dynamics and became the prototype chaotic time series. The map, in the fully chaotic regime, is given by
where t is a time index. The value of x at time t+1 is a parabolic function of x at time t. This equation represents the underlying geometry of the chaotic time series generated by the logistic map.
Generation of the time series from this equation is the forward problem. The examples here illustrate the inverse problem; identification of the underlying dynamics, or fundamental equation, of the logistic map from exemplars of the time series. The goal is to find an estimate
for f.
The architecture is
where
Since the input is a scalar rather than a vector, the input dimension is one. We choose the number of basis functions as N=5 and the size of the training set to be 100 exemplars generated by the chaotic time series. The weight β is taken to be a constant equal to 5. The weights c_{i} are five exemplars from the time series. The weights a_{i} are trained with projection operator training:
where the learning rate ν is taken to be 0.3. The training is performed with one pass through the 100 training points. The rms error is 0.15.
The normalized RBF architecture is
where
Again:
Again, we choose the number of basis functions as five and the size of the training set to be 100 exemplars generated by the chaotic time series. The weight β is taken to be a constant equal to 6. The weights c_{i} are five exemplars from the time series. The weights a_{i} are trained with projection operator training:
where the learning rate ν is again taken to be 0.3. The training is performed with one pass through the 100 training points. The rms error on a test set of 100 exemplars is 0.084, smaller than the unnormalized error. Normalization yields accuracy improvement. Typically accuracy with normalized basis functions increases even more over unnormalized functions as input dimensionality increases.
Once the underlying geometry of the time series is estimated as in the previous examples, a prediction for the time series can be made by iteration:
A comparison of the actual and estimated time series is displayed in the figure. The estimated times series starts out at time zero with an exact knowledge of x(0). It then uses the estimate of the dynamics to update the time series estimate for several time steps.
Note that the estimate is accurate for only a few time steps. This is a general characteristic of chaotic time series. This is a property of the sensitive dependence on initial conditions common to chaotic time series. A small initial error is amplified with time. A measure of the divergence of time series with nearly identical initial conditions is known as the Lyapunov exponent.
We assume the output of the logistic map can be manipulated through a control parameter c[x(t),t] such that
The goal is to choose the control parameter in such a way as to drive the time series to a desired output d(t). This can be done if we choose the control paramer to be
where
is an approximation to the underlying natural dynamics of the system.
The learning algorithm is given by
where
