Common AI Workloads

Question 1

What features and capabilities does Azure Machine Learning provide?

Answer 1

Automated Machine Learning

Azure Machine Learning designer

Data and compute management

Pipelines

Question 2

What is labelling?

Answer 2

The process of identifying raw data (images, text files, audio, etc.) and adding one or more meaningful and informative labels to provide context for machine learning.

Question 3

What is unsupervised learning?

Answer 3

Unsupervised learning is a subcategory of ML defined by its use of unlabelled datasets to train models that discover hidden patterns or data groupings without human intervention.

Question 4

What is supervised learning?

Answer 4

Supervised learning is a subcategory of ML defined by its use of labelled datasets to train models that classify data or predict outcomes precisely.

Question 5

What are two examples of supervised learning models?

Answer 5

Classification and Regression.

Question 6

What is an example of unsupervised learning models?

Answer 6

Clustering.

Question 7

What is a dataset?

Answer 7

A collection of data.

Question 8

What’s the difference between unsupervised ML labelling and supervised ML labelling?

Answer 8

With supervised ML, labelling is prerequisite to produce training data and each piece of data will generally be labelled by a human.

With unsupervised ML, labelling is produced by the computer and may not be human readable.

Question 9

What is regression?

Answer 9

A form of machine learning that is used to predict a numeric label based on an item’s features.

Question 10

What is time series forecasting?

Answer 10

Regression with a time-series element, that predicts numeric values at a future point in time.

Question 11

What is classification?

Answer 11

A form of machine learning that is used to predict which category, or class, an item belongs to.

Question 12

What is clustering?

Answer 12

A form of machine learning that is used to group similar items into clusters based on their features.

Question 13

What is a ground truth?

Answer 13

A properly labelled dataset used as the objective standard to train and assess a given model.

The accuracy of the trained model is dependant on the accuracy of the ground truth.

Question 14

What are the stages of the ML pipeline and what are they for?

Answer 14

Pre-processing - preparing data and feature engineering before passing the data to an ML model for training or inference.

Post processing - translating the output of a ML model back into a human readable format

Training - the process of training the model

Serving - the process of deploying the model to an endpoint to be used for inference

Inference - Invoking a ML model by sending a request and expecting back a prediction.

Question 15

What is data cleaning?

Answer 15

The process of correcting errors within a dataset.

Question 16

What is data reduction?

Answer 16

Reducing the volume of data, or applying dimensionality reductions to reduce the dimensions of inputted vectors

Question 17

What is feature engineering?

Answer 17

Transforming data into numerical values (vectors) to be ingested by a ML model.

Question 18

What is sampling?

Answer 18

Balancing a dataset to be uniform across labels by adding or removing records.

Question 19

How are features and labels used in ML?

Answer 19

ML uses features to predict labels.

Question 20

What is a training dataset?

Answer 20

A training dataset is used to train an ML model.

Question 21

What is a validation dataset?

Answer 21

A validation dataset is used to estimate the accuracy of an ML model.

Question 22

What is an algorithm in ML?

Answer 22

A procedure run on data to create an ML model.

Question 23

How do ML algorithms work?

Answer 23

By performing pattern recognition. They learn from data or are fit on a dataset.

Question 24

What are the 5 model evaluation metrics for classification?

Answer 24

Accuracy

Precision

Recall

F1 Score

AUC

Question 25

What does MAE measure?

Answer 25

The average difference between predicted values and true values.

The lower this value is, the better the model is predicting.

Question 26

What does RMSE measure?

Answer 26

The average square root of the mean squared difference between predicted values and true values.

When compared to the MAE, a larger difference indicates greater variance in the individual errors.

Question 27

What unit are MAE and RMSE based on?

Answer 27

The same unit as the label.

Question 28

What is RSE?

Answer 28

A relative metric based on the differences between predicted values and true values.

Question 29

What is RAE?

Answer 29

A relative metric based on the absolute differences between predicted values and true values.

Question 30

What is the range of RSE and RAE?

Answer 30

0 to 1.

The closer to 0 the metric is, the better the model is performing.

Question 31

What can you use RSE and RAE for and why?

Answer 31

As the metrics are relative, they can be used to compare models where the labels are in different units.

Question 32

What does accuracy measure?

Answer 32

The ratio of correct predictions (true positives + true negatives) to the total number of predictions.

Question 33

What does precision measure?

Answer 33

The fraction of positives correctly identified.

True positives / true positives + false positives

Question 34

What does recall measure?

Answer 34

The fraction of classified positives that were actually positives.

(true positives / true positive + false positives)

Question 35

What is F1 Score?

Answer 35

An overall metric that essentially combines precision and recall.

Question 36

What is AUC?

Answer 36

Area under curve. It is the metric that measures the area under the ROC curve.

It can be any value from 0 to 1.

The larger the AUC, the better the model is performing.

Question 37

What is knowledge mining?

Answer 37

A discipline in AI that uses a combination of intelligent services to quickly learn from vast amounts of information.

Question 38

What would you use knowledge mining for?

Answer 38

Content research

Auditing, risk, and compliance management

Business process management

Customer support and feedback analysis

Digital asset management

Contract management

Question 39

What is feature selection?

Answer 39

The process of deciding which relevant original features to include and which irrelevant features to exclude for predictive modelling.

Question 40

What is the difference between feature selection and dimensionality reduction?

Answer 40

In feature selection, the original features don’t change. In dimensionality reduction new features are created from original features.