ANN with Arduino: Quiz 4

In the IO control case, how are initial weights calculated before training?

• Set to zero

• Random values between -0.5 and 0.5 times INITIAL_WEIGHT_MAX

• Fixed at 1.0

• Copied from EEPROM

What determines the error in the robot navigation training?

• Difference between target and actual output

• Motor speed difference

• Compass heading shift

• Sonar distance reading

How is the weight update step size controlled in both case studies?

• Momentum alone

• Error threshold

• Learning rate and momentum

• Pattern count

In the IO control training, what’s squared to compute the error?

• Target minus output

• Input values

• Hidden weights

• Output delta

What multiplies the hidden delta during weight updates in the robot code?

• Learning rate

• • Output values

• Sonar distance

• Compass heading

In runtime for IO control, where are weights loaded from for calculation?

• Serial monitor

• EEPROM

• Input buttons

• Random seed

What’s used to calculate the error for each output in both case studies?

• Mean Squared Error (MSE)

• Sigmoid function

• Absolute difference

• Random offset

During training, how are weights updated after the backward pass in IO control?

• Adjusted by delta times learning rate plus momentum

• Reset to initial values

• Increased by a fixed step

• Copied from best weights

In the robot case, what triggers a weight update during training?

• Error calculation from forward pass

• Motor speed change

• Compass alignment

• Sonar threshold

What ensures weights improve in the robot navigation training loop?

• Random pattern shuffling

• Fixed cycle count

• Gradient descent with backpropagation

• EEPROM storage