2. Why Neural Networks?
State-of-the-art
solutions for the
diversity of applications
3. What do Neural
Networks do?
• Clustering, classification, categorization
• Function approximation
• Prediction
• Optimization
• Associative memory
• Control
4. What Applications?
Manufacturing process control
Credit application Manipulator controllers
evaluation
Image/data compression
Autopilot Breast cancer cell analysis
Vehicle scheduling Special Effects
5. What Industries?
Manufacturing
Robotics
Banking
Telecommunications
Aerospace Medical
Transportation Entertainment
10. Neural networks
Static Dynamic
The output is calculated The output depends also on
directly form the input the previous inputs, outputs,
through feedforward or states of the network
connections
11. Neural networks
Static Dynamic
The output is calculated The output depends also on
directly form the input the previous inputs, outputs,
through feedforward or states of the network
connections
13. The work flow for the NN design process
1. Collect Data
2. Create the network
3. Configure the network
4. Initialize the weights and biases
5. Train the network
6. Validate the network
7. Use the network
14. Train the Network
Tuning weights and biases of the NN to
optimize NN performance function,e.g.
Mean Square Error
N N
1 2 1 2
F = (ei ) = (ti − ai )
N i=1
N i=1
15. Optimization methods
• Use GRADIENT of the network
performance w.r.t. the network
weights
• Use JACOBIAN of the network
errors w.r.t. the network weights