Instructions: Each question may have any number of correct choices. Clearly mark (tick)
all choices you believe to be correct (1 mark for each correct choice, −0.5 for each incorrect
choice).
Consider the following possible choices of error function in training a neural network for classification:
cross-entropy error (I), classification error (II), and sum-of-squares error (III). Which of the following
are true?

Question

Instructions: Each question may have any number of correct choices. Clearly mark (tick)
all choices you believe to be correct (1 mark for each correct choice, −0.5 for each incorrect
choice). 
 Consider the following possible choices of error function in training a neural network for classification:
cross-entropy error (I), classification error (II), and sum-of-squares error (III). Which of the following
are true?

Answer

(II) is problematic because it’s non-differentiable, but either of (I) or (III) should give the same
result.

Answer

(II) is problematic because it’s non-differentiable; (III) is preferred to (I) because the former
corresponds to maximising the likelihood of the data.

Answer

Any of the three could be easily used for backpropagation, but (I) is preferred because it
corresponds to maximising the likelihood of the data.

Answer

Any of the three could be easily used for backpropagation, but (III) is preferred because it
corresponds to maximising the likelihood of the data.

Answer

(II) is problematic because it’s non-differentiable; (I) is preferred to (III) because the latter
corresponds to an inappropriate noise model.

Crowdly

Instructions: Each question may have any number of correct choices. Clearly mark...

Want instant access to all verified answers on moodle.elct.lnu.edu.ua?