Competitive Learning

COMPETITIVE LEARNING
Learning
• The main property of a neural network is an
ability to learn from its environment, and to
improve its performance through learning.
• Learning may be viewed as the change in
behaviour acquired due to practice or experience
and it lasts for relatively long time.
• In case of artificial neural network, it is a process
of modifying neural network by updating its
weight, biases and other parameters if any.
Definition of learning:-
“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
manner in which the parameter changes take place.”
- By Mendel and McClaren(1970)
*stimulation – encouragement of something to make it
develop or become more active.
Types of Learning
learning
Supervised Unsupervised Reinforcement

Unsupervised Learning
• Learning without a teacher:-
only input is given , no output is told.
during the training session, the neural network
receives a number of different input patterns, discovers
significant features in these patterns and learns how to
classify input data into appropriate categories.
Unsupervised learning tend to follow the neuro-
biological organization of the brain.
It’s aim is to learn rapidly and can be used in real-time.
Competitive Learning
• In competitive learning, neurons compete among
themselves to be activated.
• Only a single output neuron is active at any time.
• The output neuron that wins the “competition” is called
the winner-takes-all neuron.
• The basic idea of competitive learning was introduced in
the early 1970s.
• In the late 1980s, Teuvo kohenen introduced a special
class of artificial neural networks called self-organizing
feature maps. These maps are based on competitive
learning..
#Basic Elements
• –A set of neurons that are all same except
synaptic weight distribution
• •respond differently to a given set of input
pattern
• •A mechanism to compete to respond to a
given input
• •The winner that wins the competition is
called“winner-takes-all”
Input layer Output layer (Single)
Competitive learning rule:-
the output signal yk of the winning neuron k is set equal to one .the output signal
of all the neurons that loose the competition are set equal to zero.
yk = 1 if vk > vj for all j , j≠k

0 otherwise
Here
vk represent all the combined action of all the forward and feedback input to
neuron k.
W(kj) synaptic weight connecting input node j to neuron k.
∑w(kj)=1 for all k
Competitive Learning Rule:-
∆w(kj) = ἠ(x(j)- w(kj)) if neuron k wins
0 if neuron k looses
Applicaton of competitive learning :-
• In action-selection for intelligent agents:-
Winner-takes-all is a concept that has been widely applied in behavior-
based robotics as a method of action selection for intelligent agents.
Winner take-all systems work by connecting modules(task designated areas)
In such a way that when one action is performed it stops all other actions
from being performed, so only one action is occurring at a time.
For example:- if agent’s goal were to reduce thirst, the behaviour “drink”
would require the pre-condition of having water available, so the module
would activate the module in charge of “find water”.
So the activations organize the behaviours into a sequence, even though only
one action is performed at a time.
Maxnet(fixed weight competitive net)
• MAXNET [Lippmann, 1987] is a specific example
of a neural net based on competition.
• It can be used as a subnet to pick the node
whose input is the largest.
• The m nodes in this subnet are completely
interconnected, with symmetric weights.
• There is no training algorithm for the MAXNET;
the weights are fixed.
• The architecture of MAXNET is as shown in the
figure
Maxnet Architecture
• The activation function for the MAXNET is
x if x ≥ 0
f(x) =
0 if x < 0
• The application procedure is as follows:

1)Initialize activations and weights ( 0 < ε < 1/m)
aj(0) input to node Aj
wij = 1 if i = j
- ε if i ≠ j
2- Update the activation of each node: j = 1, … , m
aj(new) = f [aj(old) – ε Σ ak(old)] k≠ j
3- Save activation for use in next iteration

aj(old) = aj(new), j = 1, … , m
4- If more than one node has a nonzero activation, then updating the
activation will continue, otherwise we should stop.
• Example -1: Consider the action of a MAXNET with four neurons
and inhibitory weights ε = 0.2. The neurons given initial activation
(input signal)
al(0) = 0.2, a2(0) = 0.4, a3(0) = 0.6, a4(0) = 0.8
Solution:-
aj(new) = f [aj(old) – ε Σ ak(old)]
a1(new) = f [a1(old) – 0.2(a2(old)+ a3(old)+ a4(old)]

• a1(1) = f [a1(0) – 0.2(a2(0)+ a3(0)+ a4(0)]
= f [0.2 - 0.2(0.4 + 0.6 + 0.8)] = f [- 0.16] = 0
• a2(1) = f [a2(0) – 0.2(a1(0)+ a3(0)+ a4(0)]
= f [0.4 - 0.2(0.2 + 0.6 + 0.8)] = f [0.08] =0.08
• a3(1) = f [a3(0) – 0.2(a1(0)+ a2(0)+ a4(0)]
= f [0.6 - 0.2(0.2 + 0.4 + 0.8)] = f [0.32] =0.32
• a4(1) = f [a4(0) – 0.2(a1(0)+ a2(0)+ a3(0)]
= f [0.8 - 0.2(0.2 + 0.4 + 0.6)] = f [0.56] =0.56
• The activations found as the net iterates are
• al(l) = 0.0, a2(l) = 0.08 ,a3(l) = 0.32, a4(l) = 0.56
• al(2) = 0.0 ,a2(2) = 0.0 ,a3(2) = 0.192 ,a4(2) = 0.48
• al(3) = 0.0, a2(3) = 0.0, a3(3) = 0.096 a4(3) =0.442
• al(4) = 0.0 a2(4) = 0.0 a3(4) = 0.008 a4(4) = 0.422
• a1(5) = 0.0 a2(5) = 0.0 a3(5) = 0.0 a4(5) = 0.421,

Competitive Learning

Uploaded by

Copyright:

Available Formats

Competitive Learning

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Competitive Learning

Uploaded by

Copyright:

Available Formats

What are the different types of learning in neural networks?

What is competitive learning?

COMPETITIVE LEARNING

Supervised Unsupervised Reinforcement

yk = 1 if vk > vj for all j , j≠k

• The application procedure is as follows:

3- Save activation for use in next iteration

a1(new) = f [a1(old) – 0.2(a2(old)+ a3(old)+ a4(old)]

You might also like