Assignment 7 - 5/05

bannerclubAI and Robotics

Oct 20, 2013 (4 years and 23 days ago)

52 views

Matthew Derenski

Assignment 7 Alt.

5/15
/08

Neural Network



A neural network “is a mathematical model or computational model based on biological
neural networks. It consists of an interconnected group of artificial neurons and processes
information using a

connectionist approach to computations. In most cases it is an adaptive
system that changes its structure based on external or internal information that flows through
the network during the learning phase.” There are six main types of neural networks and
three
learning paradigms. The main network types are: feedforward, radial basis function, kohonen
self
-
organizing, recurrent, stochastic, and modular. The three learning paradigms are:
Supervised, unsupervised, and reinforcement.


The feedwordard network i
s the “
simplest type of artificial neural
network devised. In this network, the information moves in only one
direction, forward, from the input nodes, through the hidden nodes (if
any) and to the output nodes. There are no cycles or loops in the
network.


The Radial Basis Function network uses functions that have


powerful techniques for interpolation in multidimensional space. A
RBF is a function which has built into a distance criterion with
respect to a centre.
” The kohonen self organizing network is a

self
-
organizing map (SOM) invented by Teuvo Kohonen
and
uses a form of
unsupervised learning. A set of artificial neurons learn to map points
in an input space to coordinates in an output space. The input space
can have different dimensions and topology f
rom the output space, and
the SOM will attempt to preserve these.

“ A recurrent networks are
“c
ontrary to feedforward networks,
they
are models with bi
-
directional
data flow. While a feedforward network propagates data linearly from
input to output, RNs al
so propagate data from later processing stages
to earlier stages.

A stochastic neural network differs from a typical
neural network because it introduces random variations into the
network. In a probabilistic view of neural networks, such random
variations

can be viewed as a form of statistical sampling, such as
Monte Carlo sampling.
” Finally, a modular network is where “
several
small networks cooper
ate or compete to solve problems
.



There are three main learning paradigms for a neural network the
first is

supervised

learning
. This method uses inference to determine
the solution for a given problem. Some common applications that use
supervised learning are pattern recognition and regression. Supervised
learning is well suited for work with sequential data p
rocessing. Next
is unsupervised learning
,
in this

learning
method “we are given some
data
, and the cost function to be minimized can be any function of the
da
ta and the network's output
.
” This method is well suited for
estimation problems and can be used f
or clustering, statistical
distributions, compression, and filtering. The final learning method
is “
reinforcement learning,
data

is usually not given, but generated
by an agent's interactions with the environment.
” The cost and
solution of this learning me
thod are based on the observations of the
data generated by the agents. This method of learning is well suited
to control problems, games, and sequential decision
-
making. Due to
that face that data is generated based on the agents interaction with
the envi
ronment it will dynamically adapt the solution to changes in
the environment.

(
http://en.wikipedia.org/wiki/Artificial_neural_network
)