Artificial Neural Networks

1
Artificial Neural Networks

• Dr. Bimal Kumar

2
Artificial Neural Network (ANNs)

• What is Neural Network?
• Models that mimic the functioning of human brain
  to a very small scale
• Structure of Human Brain
• Billions of Neurons interconnected in a very
  complex manner

3
Artificial Neural Network (ANNs)

• A Biological Neuron
• Powerful Computational Units of the Brain
• Dendrites brings in Signal to the cell body
• Axons carry signals to the adjacent neuron

4
Artificial Neural Network (ANNs)

• Neurons The basic computational elements of ANN

5
Artificial Neural Network (ANNs)

• The node receives inputs (x1,x2......,xn etc.)
• The net input to a node is a weighted sum of
  these values as where wij is the
  weight of interconnection from ith node to the
  jth node.
• The neuron is fired if the net input exceeds a
  threshold.
• The output of a neuron is decided by a function
  (Activation function)

6
Artificial Neural Network (ANNs)

• Activation Function
• A neuron produced different types of outputs
  depending on the type of activation function
• The various types include
• A Simple Linear Function
• Linear Threshold Function
• Hard Limiter
• Non-Linear Logistic (Sigmoid Function)

7
Artificial Neural Network (ANNs)

• Simple Linear Function Output of the net is
  same as the input
• Hard Limited Takes either -1 or 1 as the
  output
• Linear Threshold A hybrid of Simple linear and
  hard limiter
• Logistic Function Passes the value through a
  non-linear function

8
Artificial Neural Network (ANNs)

• ANNs
• A collection Artificial Neurons
• Computationally powerful
• Maps the unknown non-linear relationships between
  the given inputs and outputs
• Successfully applied in many civil and structural
  engineering problems

9
Artificial Neural Network (ANNs)

• Multilayered Feedforward Network
• Most widely used network
• Back Propagation Algorithm is used
• Consists of different layers of nodes
• An input layer of nodes receives inputs from the
  external world
• The output of the net are produced at an output
  layer of nodes
• One or More hidden layers that transmit the data
  from the input nodes to the output nodes

10
Artificial Neural Network (ANNs)

• Multilayered Net with two Hidden Layers

11
Artificial Neural Network (ANNs)

• The net is trained to learn the relationship by
  presenting the training set
• Two types of learning
• Supervised learning The training set include
  the inputs and the outputs between which a
  relationship has to be formed
• Unsupervised learning The target outputs are
  not specified in the training set. The net learns
  to form classifications from the training set

12
Artificial Neural Network (ANNs)

• Some Civil Structural Engineering Applications
• Material Modeling
• Highway Maintenance
• Damage Assessment
• Seismic Liquefaction Assessment
• Determination of Pile Capacity
• Structural Analysis and Design
• Optimization of Structural Design
• Design of Masonry Walls etc.