ARTIFICIAL NEURAL NETWORKS by nikeborome

VIEWS: 159 PAGES: 32

									ARTIFICIAL NEURAL
   NETWORKS
Introduction to Neural
      Networks
          Neural network
• It is an information processing
  paradigm
• It is based on the way in which
  biological nervous system works.
• It helps in processing information.
• e.g. ANN
      Use of Neural Networks
•    Remarkable ability to derive meaning
     from complicated data.
•    Used to extract patterns and detect
     complex trends.
•    It can be compared to an expert.
•    Advantages
1.   Adaptive learning
2.   Self organisation
3.   Real time operations
4.   Fault tolerance via redundant
     information coding.
       Neural network versus
      Conventional Computers
• Conventional computers use algorithmic
  approach i.e. computer follows a set of
  instructions in order to solve a problem
  which is in a way limit to solving capability.
• Neural networks process information like our
  brain does.
• Neural networks and conventional
  computers are not in competition but
  complement to each other.
Similarities between human
  and Artificial Neurons
  Learning of a Human Brain
• The structure of a Human Neuron is shown
  below
When a neuron receives excitatory inputs
larger than inhibitory input it sends an
electrical activity down its axon to the
synapses and thus the communication
between various neurons exists.
From human neurons to artificial
          neurons
• First we deduce essential features of
  neurons and their interconnection.
• Secondly, we program a computer to
  stimulate these features .
• Finally model achieved is a gross
  idealisation of real networks of
  neurons.
An Engineering Approach
             Artificial Neuron
•    It is a device with many inputs and one output.
•    Two modes of operation
1.   Training mode
2.   Using mode
                Firing Rule
• Important concept accounting for high
  flexibility in neural network.
• Firing rule can be implemented using
  hamming distance technique.
• Firing rule applied to a 3 - input neuron.
X1:       0    0    0     0     1     1   1     1
X2:       0    0    1     1     0     0   1     1
X3:       0    1    0     1     0     1   0     1


OUT:      0    0    0/1   0/1   0/1   1   0/1   1
• The truth table after generalisation :
X1:       0    0    0    0     1     1     1   1
X2:       0    0    1    1     0     0     1   1
X3:       0    1    0    1     0     1     0   1


OUT:      0    0    0    0/1   0/1   1     1   1
      Pattern Recognition
• An important application of neural
  networks
• can be implemented using a feed
  forward neural network that has been
  trained accordingly.
• Example: The figure is trained to recognize
  the following patterns:
The truth table for 3-neurons after generalisatio
  X11:     0   0     0      0        1     1   1     1
  X12:     0   0     1      1        0     0   1     1
  X13:     0   1     0      1        0     1   0     1


  OUT:     0   0     1      1        0     0   1     1
                     Top neuron


  X21:     0   0     0      0        1     1   1     1
  X22:     0   0     1      1        0     0   1     1
  X23:     0   1     0      1        0     1   0     1


  OUT:     1   0/1   1      0/1      0/1   0   0/1   0
                     Middle neuron
X21:   0   0    0     0        1   1   1   1
X22:   0   0    1     1        0   0   1   1
X23:   0   1    0     1        0   1   0   1


OUT:   1   0    1     1        0   0   1   0

               Bottom neuron
     From the tables following
   associations can be extracted




• Conclusion-The output is black and the total
  output of the network is still in favour of the
  “T” shape.
Architecture of Neural
      Networks
     Feed-forward Networks
• Allow the signal to travel in one direction.
• Are straight forward networks that associate
  inputs with outputs.
• Extensively used in pattern recognition.
       Feedback Networks
• Signal travel in both directions.
• Are dynamic in nature.
• Used to denote feedback connections in
  single layer organisations.
         Network Layers
• Three units-input, hidden, output.
• Activities of these units.
• Simple network is interesting because
  of hidden layers.
• Single and multi-layer architectures.
Applications of Neural
      Networks
    Neural Networks in Practice
•   They are best suited for prediction or
    forecasting including: industrial
    process control, data validation, risk
    management, etc.
•   Also used in specific paradigms:
    interpretation of multi meaning, texture
    analysis, facial recognition,
    recognition of speakers in
    communications ,etc.
 Neural Networks in medicine
• The research on modeling parts of the
  human body and recognizing diseases
  from various scans.
• Used effectively in recognizing
  diseases as no details are needed to
  how to recognize the and no specific
  algorithm need to be provided.
 Modelling and diagonising the
    cardiovascular system
• Potential harmful medical conditions
  can be detected at early stage using
  artificial cardiovascular system models.
• Ann technology is used as it provides
  sensor fusion which is combining of
  several values from different sensors
        Electronic Noses
• Neural networks have made possible to
  transmit various odours over long
  distances via communication links.
• This has help in enhancing
  telemedicine and telepresent surgery.
         Instant Physician
• An associative neural network to store
  a large number medical records
  including symptoms,diagnosis,and
  treatment of specific case.
• After training, the net can be presented
  with input consisting of a set of
  symptoms; it will then find the full
  stored pattern that represents the
  "best" diagnosis and treatment.
Neural Networks in Business
• Any neural network application would
  fit into one business area or financial
  analysis.
• Neural networks is used for dataminig
  purposes, for various business
  purposes including resource allocation
  and scheduling.
        Enhancing Trading
• The identification of specific patterns in
  stock price derived from technical
  stock analysis heuristics, which after
  occurring results in a predefined price
  movement.
• Neural networks are trained in the
  experiments to classify whether the
  outcome of an occurred pattern will
  result in a predefined price movement.
      ANNs in Water Supply
         Engineering
• Whenever this technology is applied for
  water supply engg. problems have reported
  findings that were beyond the capability of
  traditional statistical / mathematical
  modeling tools.
• Some of the applications performed
  includes: Forecasting salinity levels in River
  Murray, South Australia; Predicting
  gastroenteritis rates and waterborne
  outbreaks; Modeling pH levels in a eutrophic
  Middle Loire River, France;
 Understanding Brain Activity
• It provides a powerful new approach for
  neuroscience to study and manipulate signal
  propagation in neuronal networks

• It represents a new, powerful, and flexible
  approach for real-time cellular assays useful
  for drug discovery and other applications;
  and it opens the possibility for hybrid
  circuits that couple the strengths of digital
  nanoelectronic and biological computing
  components.

								
To top