Glossary

This glossary page includes a list of definitions for terms that will be commonly referred to during the workshop. For more extensive and detailed definitions of terms relevant to Artificial Intelligence and Machine Learning, we recommend that you read the ‘Some Key Machine Learning Definitions’ post on Medium or check out the Machine Learning Glossary.

Artificial Intelligence

A field of study concerned with the simulation of aspects of human intelligence (for example, complex reasoning and adaptability) by computers.

Artificial Intelligence (AI) is a large field that encompasses a variety of topics, including: philosophy, ethics of AI and Machine Learning.

Attribute

A feature or characteristic of an observation (for example, a person in our data set could have the following attributes: weight, height and age).

The attributes that we include in the model usually have an impact on the outcome that we are trying to predict or classify. For example, if we were predicting how likely a student was to withdraw from university, one of the the attributes in the model could be the percentage of tutorials that they had attended.

Bucket

When using the Machine Learning for Kids tool we put the examples into different buckets to tell the computer how they should be classified.

For example, when training a model that recognises whether a photo contains a cat or a dog, we put the photos of cats in the ‘cat’ bucket and the photos of dogs in the ‘dog’ bucket. Buckets are sometimes referred to as labels or classes.

Classification

A type of Machine Learning algorithm that takes data and tries to recognise the class it belongs to (for example, whether an email is spam or is not spam).

When using Classification algorithms, the training data is split into different buckets and the model predicts which of the buckets new data belongs to.

Clustering

A type of Machine Learning algorithm that takes data and splits the data into groups (clusters) of data points that have similar attributes.

Clustering algorithms are commonly used when identifying customer segments, which are groups of customers that have similar characteristics. Segmenting customers can help marketers develop targeted advertising.

Computer Vision

A field of study that involves making computers ‘see’, which includes the emulation of humans’ visual abilities (such as the ability to recognise faces).

For more information about Computer Vision, see the Computer Science Field Guide’s Computer Vision Chapter.

General AI

A category of AI types that we only see in Science Fiction movies and that may never exist in the real world.

General AI includes two types of AI: Theory of mind and Self-awareness. Theory of mind AI is AI that has has the “understanding that people, creatures and objects in the world can have thoughts and emotions that affect their own behavior.” Self-awareness AI have this understanding as well but also “are aware of themselves, know about their internal states, and are able to predict feelings of others.” (like HAL 9000 in the film 2001: A Space Odyssey).

For more information about General AI and the four types of AI, see this Conversation article written by Arend Hintze.

Machine Learning

A large topic within AI that is concerned with ‘teaching’ computers to perform tasks, make predictions or classify data.

When ‘teaching’ a computer, we have to give it examples (data) and the computer will learn from these examples. There are different methods of Machine Learning (for example, reinforcement learning, supervised learning and unsupervised learning ) that are used for solving different problems.

Model

A simplified representation of a phenomenom that usually exists in the real world. Examples of models include statistical models, financial models and engineering models.

When we create Machine Learning solutions, we usually create a model that represents the problem we are trying to solve. Consequently, it is important that we model the problem appropriately by choosing an appropriate type of algorithm and identifying attributes that have an impact on the outcomes we are interested in.

Narrow AI

A category of AI types that can solve narrow and specific problems (such as playing chess), which are the most common types of AI that exist today.

General AI includes two types of AI: Reactive Machines and Limited Memory. Reactive Machines are the most common types of AI, which only have a representation a problem of one point in time. That type of AI does not have any memory but can use data to predict what will happen in the future (for example, the moves an opposing chess player may make). Limited Memory AI can record information and recall this but computers do not have the same ability to remember and learn from experiences that humans do.

For more information about Narrow AI and the four types of AI, see this Conversation article written by Arend Hintze.

Observation

An individual data point in a data set. For example, in data from a survey, each survey response would be an observation.

Another example of an observation would be an individual student in a data set that was used to predict students’ likelihood to drop out of university.

Regression

A method for finding the relationship between different variables (or attributes) in a model, which can give us a mathematical equation of the ‘line of best fit’.

There are a variety of different types of regression, including linear regression and logistic regression. Linear regression can be useful for answering questions that involve numbers. For example, we could predict album sales with a linear regression model that estimates how many sales we will get for a given amount of money spent on advertising.

Regression can be used as a supervised learning technique. Most statistics courses and resources will cover regression because it is a technique that is fundamental in ‘traditional’ statistics. For example, Khan Academy has a set of videos on regression.

Reinforcement Learning

A category of Machine Learning algorithms where computers learn by maximising the rewards they encounter in an environment (for example, scoring points in a videogame).

Algorithms in this category are used for a variety of purposes, including training AI for videogames. You can find out more about reinforcement learning in this Geeks for Geeks article.

Supervised Learning

A category of Machine Learning algorithms where the model is trained with examples that have been given labels or put into buckets.

The model is then trained to identify which of the buckets the input data belongs to. Examples of Supervised Learning Methods include classification and regression.

Unsupervised Learning

A category of Machine Learning algorithms where data is given to a model without labels and the model arranges the data into groups or identifies its own labels for the data.

A common type of Unsupervised Learning algorithms are clustering algorithms.