Upload
aeditya-yadav
View
165
Download
0
Tags:
Embed Size (px)
Citation preview
ContentIntroductionHistoryCapabilitiesApplicationsReal world implementationIssuesAdvantage/DisadvantageConclusionsReferences
IntroductionBrain : A highly complex, non-linear &
parallel computing
Structural constituent : ‘Neurons’
A computer program designed to model the human brain.
Conti…Our brains are made up of about 10
billion tiny units called neurons .
Tree like Nerve fibres are called dendrites .
Signals coming into the neuron are received via junctions called synapses .
Artificial Neuron•
Dendrites
Soma (cell body)
Axon
SynapsesThe information transmission happens at the synapses.
Conti…A network of interconnected functional
elements.
A NN is trained to recognize and generalize the relationship between a set of inputs and outputs.
HistoryInspiration for development came from
attempts to model the human central nervous system.
McCulloch-Pitts 1943 Introduced a simple NN model using electrical circuits.
Conti…Hebb 1949 Wrote that neural pathways
are strengthened every time they are used.
Minsky 1954: Learning Machine.
Rosenblatt’s Perceptron 1957.
Conti…
Widrow 1960 : Adaline model.
Recent work includes Boltzmann machines, competitive learning models, multilayer networks, and adaptive resonance theory models.
Artificial neuronsNeurons work by processing
information.
The McCulloch-Pitts model
Outputw2
w1
w3
wn
wn-1
. . .
x1
x2
x3
…
xn-1
xn
y)(;
1
zHyxwzn
iii == ∑
=
CapabilitiesNon-linearityInput/Output mappingAdaptivityEvidential responseFault tolerenceVLSI implementability
ApplicationsRoboticsImage processingSpeech/Pattern recogntionGamingTarget RecognitionMedical Diagnosis Voice and touch interface with
computers and other devices.
Real World ImplementationLogical reasoningPattern recognitionPlanningGenetic programmingCommon sense knowledgeRepresentationControl system
IssuesComplex programsDifficult to implementMachine prediction may not be accurate Human beings may lose their importance
Advantage
Pattern recognitionDoes not need to be reprogrammedImplemented in any application Adaptive learningSelf-OrganisationReal Time OperationFault Tolerance
ConclusionsOnce NNs are trained they can be
reapplied over and over again.
Can be linked with other models to solve complex problems.
Conti…Neural Networks LEARN. They are not
programmed.
Can be applied to areas where humans are often wrong too.
Referenceswww.wikipedia.org
www.ai-junkie.com
“Artificial Neural Network” by B.Yegnanarayana