Roadmap to Membership of RICS - Pathways and Routes
Learning Methods in a Neural Network
1. LEARNING METHODS
One of the most important properties of neural networks is to improve their
performances by taking into accountthe past experiences. This is achieved
through a process called learning. This improvement takes place over time in
accordancewith some prescribed measure. A neural network learns about its
environment through an interactiveprocess of adjustmentsapplied to its
synapticweightsand biaslevels. Ideally, the network becomes more
knowledgeableaboutitsenvironmentafter each iteration of the learning
process.
However, the process oflearning is a matter of viewpoint of an individual.
Hence, there is no universally agreeable definition of learning due to ambiguity
across all the available ones. Forexample, learning in the opinion of a
psychologist is quite different from learning in a classroom sense. Considering
the discussion to be bound within the domain of neural networks, the following
definition of learning adapted from Mendel and McClaren can be used:
“Learning is a process by which the free parameters of a neural network are
adapted through a process of stimulation by the environment in which the
network is embedded. The type of learning is determined by the parameter in
which the changes take place.”
This definition of learning process implies the following sequence of events:
1. The neural network is stimulated by an environment.
2. The neural network undergoes changes in its free parameters as a result of
this stimulation.
3. The neural network responds in a new way to the environment because of
the changes that have occurred in its internal structure.
Learning Paradigms:
There are three major learning paradigms: supervised learning, unsupervised
learning and reinforcement learning. Of these three, the first one can be viewed
as “learning with a teacher”, while the remaining two can be viewed as
“learning withouta teacher”. Usually they can be employed by any given type
of artificial neural network architecture. A prescribed set of well-defined rules
for the solution of a learning problem is called a learning algorithm. Each
learning paradigm has many learning algorithms.
2. SUPERVISED LEARNING:
Supervised learning, sometimes referred to as learning with a teacher is a
learning technique that sets parameters of an artificial neural network from
training data which serves as the “teacher” in this case. The task of the learning
artificial neural network is to set the value of its parameters for any valid input
value after having seen output value. The training data consistof labeled pairs
of input and desired output values that are traditionally represented in data
vectors, which may be conceptually thought of as an “environment” and
remains unknown to the neural network of interest. The neural network, after
learning from the training data or “teacher”, provides an output to a random
input which resembles the training data examples as close as possible. The
deviation of the actual output is called the error signal. Clearly, for good
performance, the value of this signal should be zero ideally and the least
possible in practical implementation cases.
Supervised learning can also be referred as classification, where we have a
wide rangeof classifiers, each with its strengths and weaknesses. Choosing a
suitable classifier (Multilayer perceptron, SupportVector Machines, k-nearest
neighbour algorithm, decision tree, radial basis function classifiers and others)
for a given problem is however still more an art than a science.
Fig. 1: Block diagram of Supervised Learning Model
3. In order to solve a given problem of supervised learning various steps has to be
considered:
1. Determine the type of training examples
2. Gather a training data set that satisfactory describe a given problem
3. Describe gathered training data set in form understandable to a chosen
artificial neural network
4. Do the learning and after the learning test the performance of learned
artificial neural network with the test (validation) data set. Test data set
consist of data that has not been introduced to artificial neural network
while learning
UNSUPERVISED LEARNING:
In unsupervised learning, there is no training set or “teacher” to monitor the
progress in the learning process ofthe neural network. In lieu of that, provision
is made for a task-independentmeasure of the quality of the representation that
the network is required to learn, and the free parameters of the network are
optimized with respect to that measure. The unsupervised-training model
consists of the environment, represented by a measurement vector. The
measurement vector is fed to the learning system and the system responseis
obtained. Based upon the system responseand the adaptation rule employed, the
weights of the learning system are adjusted to obtain the desired performance.
Note that unlike the supervised-training method, the unsupervised method does
not need a desired output for each input-featurevector. The adaptation rule in
the unsupervised training algorithm performs the error-signal generation role
the teacher performs in the supervised-learning system. Thus, the behaviour of
the unsupervised learning system depends in large measure on the
adaptation rule used to control how the weights are adjusted.
Fig. 2: Block Diagram of an Unsupervised Training Model
4. Unsupervised learning is mostly used for solution of estimation problems such
as statistical modelling, compression, filtering, clustering and others. In
unsupervised learning we seek to determine how the data is organized. It differs
from supervised learning and reinforcement learning in that the artificial neural
network is given only unlabeled examples.
REINFORCEMENT LEARNING:
Reinforcement learning is a learning technique that sets parameters of an
artificial neural network, where data is usually not given, but generated by
interactions with the environment. Reinforcement learning is concerned with
how an artificial neural network should take actions in an environment so as to
maximize some notion of long-term reward. The network is not told which
actions to take, but instead must discover which actions yield the most reward
by trying them. Thus reinforcement learning can be defined as:
“Reinforcement learning is a learning process in which the training set consists
of input patterns, after completion of a sequence a value is returned to the
network indicating whethertheresultwasrightor wrong and, possibly, how right
or wrong it was.”
One of the challenges that arise in reinforcement learning and not in other kinds
of learning is the trade-off between exploration and exploitation.To obtain a
lot of reward, a reinforcement learning aided network must prefer actions that it
has tried in the past and found to be effective in producing reward. But to
discover such actions, it has to try actions that it has not selected before. The
network has to exploit what it already knows in order to obtain reward, but it
also has to explore in order to make better action selections in the future. The
dilemma is that neither exploration nor exploitation can be pursued exclusively
without failing at the task. It must try a variety of actions and progressively
favour those that appear to be best. On a stochastic task, each action must be
tried many times to gain a reliable estimate its expected reward.
Reinforcement learning is particularly suited to problems which include a long-
term versus short-term reward trade-off. It has been applied successfully to
various problems, including robotcontrol, telecommunications, and games such
as chess and other sequential decision making tasks.