neurons.

4. Write about the importance of the activation threshold (or bias).

5. Thinking about the features of artificial neural networks, explain what is

learning from experience and generalization capability.

6. Write about the main mathematical features which can be verified on both the

logistic and hyperbolic tangent activation functions.

7. Find the analytical expressions of the first order derivatives of the logistic and

hyperbolic tangent.

8. For a particular problem, it is possible to use the logistic or hyperbolic functions

as the activation function. Regarding hardware implementation, write about the

eventual features to be considered for selecting one of them.

9. Given that individual operation on artificial neurons are executed faster when

compared to biological neurons, explain why many tasks performed by the

human brain produce results faster than a microcomputer.

10. What are the main categories of problems which can be addressed by artificial

neural networks? 

learning.

2. Consider an application with four inputs and two outputs. The designers of this

application state that the feedforward network to be developed must present

exactly four neurons in the first hidden layer. Discuss about the pertinence of

this information.

3. Relating to the previous exercise, cite some factors that influence the determination

of the hidden layers number of a multiple layer feedforward network.

4. What are the eventual structural differences observed between recurrent networks

and feedforward networks.

5. In what application categories the employment of recurrent neural networks is

essential?

6. Draw a block diagram illustrating how the supervised training works.

7. Write about the concepts of training methods and learning algorithms, further

explaining the concept of training epoch.

8. What are the main differences between supervised and unsupervised training

methods?

9. What are the main differences between supervised and reinforcement learning

methods?

10. Considering a specific application, explain what performance criterion could be

used for adjusting the weights and thresholds of a network using reinforcement

learning method.