Neural networks quiz — ScienceVerse

ScientistQuestion 1 of 3

What did Rumelhart, Hinton and Williams (1986) say back-propagation repeatedly adjusts?

The activation function, until it becomes linear

The weights of the connections, to reduce the difference between actual and desired outputs

The training data, until it matches the network

The number of layers, until the network is deep enough

AdvancedQuestion 2 of 3

What did the 1986 back-propagation paper highlight as the key advantage over the earlier perceptron-convergence procedure?

It works only with one layer

It needs no training data at all

Hidden units learn to represent important features of the task

It guarantees zero error on every task

AdvancedQuestion 3 of 3

The universal approximation theorem guarantees that gradient-based training will find weights that approximate any target function.

True

False

Neural networks quick check