Session 7.1

Question 1

Deep Learning consists in…

Answer 1

training a neural network in which the inputs and outputs are the same

• Input and output are the same image

Question 2

Hidden layers have fewer perceptrons

Answer 2

• Each perceptron in a hidden layer has to represent a more elaborate
  concept
• Features are automatically created: no need to define them manually

Question 3

Deep learning uses relatively old methods:

Neural networks have been around for a long time (1950s).

Why now?

Answer 3

Big Data

• Large amounts of data are now available
• Possibility of using thousands of machines to solve a single problem

Question 4

Underfitting

Answer 4

A model that is too simple does not fit the data well (high bias)

e.g., fitting a quadractic function with a linear model

Question 5

Overfitting

Answer 5

A model that is too complex fits the data too well (high variance)

e.g., fitting a quadractic function with a 3rd degree function

Question 6

Bias

Answer 6

a model that underfits is wrong many times (high bias) but is not highly affected by slightly different training data

Question 7

Variance

Answer 7

a model that overfits is right on average, but is highly sensitive to specific training data

Question 8

We can reduce variance by

Answer 8

by putting many models together and aggregating their outcomes (without necessarily increasing bias)

-> this is the concept of ensemble methods

Question 9

Ensemble methods use

Answer 9

multiple algorithms to obtain better predictive performance than could be obtained from any of the algorithms by itself

Question 10

Using multiple algorithms usually increases model performance by:

Answer 10

reducing variance: models are less dependent on the specific training data

Question 11

Bagging (or bootstrap aggregation)

Answer 11

Creates multiple data sets from the original training data by bootstrapping – re-sample with repetition.

Runs several models and aggregates output with a voting system

Question 12

Other ensemble methods

Random Forest

Answer 12

combines bagging with random selection of features (or predictors)

Question 13

Other ensemble methods

Boosting

Answer 13

applies classifiers sequentially, assigning higher weights to observations that have been mis-classified by the previous methods

Question 14

We can use predictive models for predicting outcomes based on individual attributes

However, models based only on observational data…

Answer 14

do not inform how users would react to a specific intervention

Question 15

In most cases the best way to find out the effect of a specific intervention is to run an experiment:

Answer 15

1 Randomly assign customers to different treatment groups

2 Compare differences in behavior among treatment groups