CSCI 343 Quiz 3

Question 1

random forests is a(n) ? method

Answer 1

ensemble

Question 2

for each tree in random forests, you should

Answer 2

randomly choose a subset of features and build a decision tree using only those

Question 3

decision rules

Answer 3

if, then rules interpreted from a tree (ex: if income > 70, then risk)

Question 4

build trees with

Answer 4

training data

Question 5

test tree strength (quality) with

Answer 5

validation

Question 6

you can combine trees with high percentages to make

Answer 6

better trees

Question 7

NN stands for

Answer 7

neural network

Question 8

NN is named this way because it is

Answer 8

a model of a brain

Question 9

single neuron structure

Answer 9

inputs with weights -> neuron -> output

Question 10

the neuron contains some function

Answer 10

f(Σxiwi)

Question 11

weights are originally

Answer 11

randomly assigned

Question 12

deep neural nets have

Answer 12

a bunch of hidden layers

Question 13

recurrent neural nets have

Answer 13

cycles

Question 14

model of a neural net is the

Answer 14

structure of the net and learned weights

Question 15

activation function at node’s idea is to make the output

Answer 15

non-linear

Question 16

sigmoid function

Answer 16

f(y) = 1 / (1 + e^-y)

Question 17

which node activation function is most common

Answer 17

sigmoid

Question 18

ReLU stands for

Answer 18

rectified linear unit

Question 19

ReLU function

Answer 19

f(y) = max(0, y)

Question 20

in NN, the work is in (training/testing)

Answer 20

training

Question 21

structure of a neural net

Answer 21

input layer, 0 or more hidden layers, output layer

Question 22

NN are ? graphs

Answer 22

directed, acyclic

Question 23

typically the number of inputs in a neural net is equivalent to

Answer 23

the number of features

Question 24

it is typical for a neural net to be fully

Answer 24

connected

Question 25

typically the number of outputs in a neural net is equal to

Answer 25

the number of classes

Question 26

training process for NN

Answer 26

backpropagation (backprop) of errors

Question 27

Epochs

Answer 27

the amount of times you run through all training data

Question 28

you want the learning rate n to (increase/decrease) with time

Answer 28

decrease (so you only make minor tweaks as time goes on)

Question 29

ways to stop NN training

Answer 29

predetermine # epochs, look at error rate, use a validation set to determine if you can move on

Question 30

errors are backpropagated through; adjust ?

Answer 30

weights