Glossary

OF

AI & Machine
Learning
ESSENTIALS

Second
edition

Brought to you by:

SmartMonkey Team
 SMARTMONKEY EDITIONS

Glossary
OF

AI & Machine
Learning
ESSENTIALS

Second Edition

Brought to you by:

SmartMonkey Team
Copyright ©2018 SmartMonkey Scalable Computing

All rights reserved. No part of this publication may be reproduced,

distributed, or transmitted in any form or by any means, including
photocopying, recording, or other electronic or mechanical
methods, without the prior written permission of the publisher,
except in the case of brief quotations embodied in critical reviews
and certain other noncommercial uses permitted by copyright law.
For permission requests, write to the publisher, addressed
“Attention: Permissions Coordinator,” at the address below.

Book design by SmartMonkey

Second edition April 2019

First edition March 2019

SmartMonkey
Carrer Badajoz 32,
08005 Barcelona

smartMonkey.io

smartMonkey.io 2
Accuracy
The rate of correct or incorrect predictions made by a model over any
given set of data. This is generally determined by dividing the “total
number of correct predictions” by “the total number of examples”.

Activation Function
Activation function is any nonlinear function applied to the weighted sum
of the inputs of a neuron in a neural network. The presence of
activation functions make neural networks capable of approximating
virtually any function.

Active Learning
Active learning is a kind of machine learning in which the algorithm
decides on the data it learns from.

Algorithm
An algorithm is a sequence of explicit, step-by-step instructions that
enables a computer to solve a problem.

Ambient intelligence
Ambient intelligence refers to electronic environments that
are sensitive and responsive to the presence of people, e.g. smart devices
like Alexa, Google Home and Apple Home Pod.

Anomaly Detection
Identifying anomalous elements within a given stream of data.
These are located based on how they differ from the rest of the dataset in
relevant
criteria.
Artificial Intelligence (AI)
The field of computer science in which machines are
enabled to simulate human cognition and learning.

Artificial Narrow Intelligence (ANI)

Refers to AI that is limited to a specific set of topics and capabilities.

Artificial Neural Network (ANN)

A model used in AI; it is loosely based on the human brain. It consists of
neural layers that are used for machine learning.

Autonomic computing
A system's capacity for adaptive self-management of its own
resources for high-level computing functions without user input.
Backpropagation
Short for “backward propagation of errors,” backpropagation is a way of
training neural networks based on a known, desired output for specific
sample case.

Backward chaining
A method in which machines work backward from the desired goal, or
output, to determine if there is any data or evidence to support those goals
or outputs.

Baseline
A minimum or starting figure used to test the efficacy of your later models.
This is usually based on a reasonable hypothesis that considers relevant
factors, or on the results of simpler models previously deployed with
similar functions.

Behavioral informatics
The use of technology/devices to detect and measure human behavior to
gain insights. For example, searching Google for “Route optimization” tells
data-collection folks that you may be thinking of adopting a route
optimization application, which means there’s a good chance you’ll start
seeing online ads for route optimization software.

Binary Classification
When a Machine Learning model outputs one of two mutually exclusive
classifications. An example would be to classify a text in positive or
negative categories.
Big Data
This is the massive amount of information we now generate about
ourselves — our interests and habits — as we move through the digital
universe. Some say the term “big data” should be retired, because so much
data is collected these days that all data is now part of big data.

Bag of Words
Bag of words is a method of feature engineering for text documents.
In this model, a text (a sentence or a document) is represented as the bag
(multiset) of its words, disregarding grammar and even word order but
keeping multiplicity. The bag-of-words model has also been used for
computer vision.

Bagging
Bootstrap aggregating, also called bagging, is a machine learning ensemble
meta-algorithm designed to improve the stability and accuracy of machine
learning algorithms used in statistical classification and regression. It also
reduces variance and helps to avoid overfitting. Bagging consists of training
different base learners on
different subsets of the training set randomly, by drawing (with
replacement) random training sets from the given sample. To obtain the
final prediction, the predictions of the base learners are averaged.

Base Learner
Base learner is a learning algorithm used to build models that are then
combined by an ensemble learning algorithm. A popular example of a base
learner is decision tree which is used as a base learner in the random
forest and gradient boosting ensemble learning algorithms.

Batch
Batch, or minibatch, is the set of examples used in one iteration of model
training using gradient descent.

Batch Normalization
Batch normalization is a technique that normalizes layer inputs
per mini-batch. The technique consists in providing any layer in a neural
network with inputs that have a zero mean and a unit variance (also see
standardization). In practice, it results in speeding up training, allows for
the usage of higher learner rates, and often has a regularization effect.
Batch normalization has been found to be very effective for convolutional
and general feedforward neural networks.
Bias
The bias is an error from erroneous assumptions in the learning algorithm.
High bias can cause an algorithm to miss the relevant relations between
features and target outputs (underfitting).

Binary Variable
A binary variable is a variable (a feature or the target) that can take
values either "Yes" or "No" (True or False, one or zero, etc.)

Binary Classification
A binary classification problem is a classification problem where the label
is only one out of two classes. For example, spam detection or
determining if a patient has a certain disease or not are examples of a
binary classification problem.

Binning
Binning (also called bucketing) is the process of converting a continuous
feature into multiple binary features called bins or buckets, typically based
on value range. For example, instead of representing age as a single
integer-valued feature, the analyst could chop ranges of gates into discrete
bins: all ages between 0 and 5 years-old could be put into one bin, 6 to 10
years-old could be the second bin, 11 to
15 years-old could be the third bin, 16 to 25 would be the fourth been and
so on.

Boosting
Boosting is an ensemble learning technique that iteratively combines a set
of simple and not very accurate classifiers (referred to as weak classifiers)
into a classifier with high accuracy (called a strong classifier) by giving a
higher weight to the examples that the model is currently classifying
incorrectly. Multiple boosting algorithms exist. The most widely used ones
are AdaBoost and gradient boosting.

Bucketing
See binning.
Chatbot
Also known as a smartbots, talkbot, chatterbot, Bot, IM bot, interactive
agent, Conversational interface or Artificial Conversational Entity, is a
computer program or an artificial intelligence which conducts a
conversation via auditory or textual methods.

Churn Factor (Churn Prediction)

The process of a customer choosing to switch product or service provider
is classified as churn. Simply put, an organization's Churn Factor is the rate
at which this occurs, and Churn Prediction, is the preemptive process of
understanding when this is going to happen based on historic and current
data.

Cluster analysis
Or clustering is the task of grouping a set of objects in such a way that
objects in the same group (called a cluster) are more similar (in some
sense) to each other than to those in other groups (clusters). It is a main
task of exploratory data mining, and a common technique for statistical
data analysis, used in many fields, including machine learning, pattern
recognition, image analysis, information retrieval, bioinformatics, data
compression, and computer graphics.

Cognitive computing
In general, the term cognitive computing has been used to refer to new
hardware and/or software that mimics the functioning of the human brain
and helps to improve human decision-making. In this sense, CC is a new
type of computing with the goal of more accurate models of how the
human brain/mind senses, reasons, and responds to stimulus.
Commonsense knowledge
In artificial intelligence research, commonsense knowledge consists of
facts about the everyday world, such as "Lemons are sour", that all
humans are expected to know. The first AI program to address common
sense knowledge was Advice Taker in 1959 by John McCarthy.

Commonsense reasoning
Is one of the branches of artificial intelligence that is concerned with
simulating the human ability to make presumptions about the type and
essence of ordinary situations they encounter every day.

Computational intelligence
(CI), usually refers to the ability of a computer to learn a specific task from
data or experimental observation.

Computer Vision (CV)

The application of machine learning to tasks involving digital images or
video, such as identifying or tracking objects through a video sequence, or
segmenting images into distinct objects. Convolutional neural networks are
a powerful new tool widely used in computer vision.

Concept drift
In predictive analytics and machine learning, the concept drift means that
the statistical properties of the target variable, which the model is trying to
predict, change over time in unforeseen ways. This causes problems
because the predictions become less accurate as time passes.

Confirmation Bias
The act of intentionally seeking out results that consolidate pre-existing
beliefs or hypotheses. In Machine Learning this can extend to the process
itself, with developers preparing their data in a way that can skew any
eventual outcomes.

Connectionism
An artificial intelligence approach in the fields of cognitive science that
holds that human mental processes (such as learning) can be explained by
the operation of inherited or acquired bonds between stimulus and
response.
Convolutional Neural Network (CNN)
A special neural network architecture especially useful for processing
image and speech data. The difference between a normal feed-forward
network and a convolutional network is primarily in the mathematical
processing that takes place. Convolutional networks use an operation
known as convolution to help correlate features of their input across space
or time, making them good at picking out complex, extended features.
However, they still treat each input separately, without a memory.

Cross Validation
Cross-validation is a technique to evaluate predictive Machine Learning
models by splitting an original sample into one set to train the model, and
a test set to evaluate it against.
Data Augmentation
Supplementing datasets with more data, to ensure that there is an
abundance of information that a model can learn from. An example would
be training a model to distinguish between normal email and spam email,
but lacking enough to kick-start the learning process, so introducing
additional emails from elsewhere into the dataset to allow the model to
train effectively on.

Data Cleansing
The process of improving the data quality, usually by removing or
correcting data values that are incorrect. This is usually done before a
Machine Learning project, although throughout the knowledge discovery
process it may become apparent that further cleansing is important to
improve data quality.

Data Collection
The entire process of collecting relevant information in anticipation of a
Machine Learning project.

Data Fusion
Is the process of integrating multiple data sources to produce more
consistent, accurate, and useful information than that provided by any
individual data source.

Data Integration
Involves combining data residing in different sources and providing users
with a unified view of them.
Data Mining
Is the process of discovering patterns in large data sets involving methods
at the intersection of machine learning, statistics, and database systems.

Data Sourcing
Locating adequate avenues for the collection of data. Usually a
fundamental process in any Machine Learning project as good sourcing
methods can result in a higher quality of results later on.

Dark data
Is operational data that is not used. It is captured and stored, but not
analyzed, providing no value or insights.
According to Lucidworks, we create 7.5 septillion gigabytes of data
worldwide every day. Businesses typically analyze approximately 10% of the
data they collect. Gartner was the first to coin the term “dark data.”

Decision Tree
A decision support tool that uses a tree-like model of decisions to
illustrate every possible outcome of a decision.

Deep Blue
Was a chess-playing computer developed by IBM. It is known for being the
first computer chess-playing system to win both a chess game and a chess
match against a reigning world champion under regular time controls.

Deep learning
Also known as deep structured learning or hierarchical learning, is part of a
broader family of machine learning methods based on learning data
representations, as opposed to task-specific algorithms. Learning can be
supervised, semi-supervised or unsupervised.

Diagnosis
Is concerned with the development of algorithms and techniques that are
able to determine whether the behaviour of a system is correct. If the
system is not functioning correctly, the algorithm should be able to
determine, as accurately as possible, which part of the system is failing,
and which kind of fault it is facing. The computation is based on
observations, which provide information on the current behaviour.

Dialogue system – or conversational agent (CA)–

Is a computer system intended to converse with a human with a coherent
structure. Dialogue systems have employed text, speech, graphics, haptics,
gestures, and other modes for communication on both the input and
output channel.

Distributed artificial intelligence – (DAI)–

Also called Decentralized Artificial Intelligence, is a subfield of artificial
intelligence research dedicated to the development of distributed solutions
for problems. DAI is closely related to and a predecessor of the field of
Multi-Agent Systems.
Eager learning
Is a learning method in which the system tries to construct a general,
input-independent target function during training of the system, as
opposed to lazy learning, where generalization beyond the training data is
delayed until a query is made to the system.

Echo State Network

Echo state network deals with a random, large, fixed recurrent neural
network where each neuron gets a non-linear response signal, and the
connectivity and weights of neurons are fixed and assigned randomly. By
dealing with input weights this way, the echo state network achieves a sort
of flexible type of learning.

Embodied AI
Robots that are equipped with AI functionality.

Ensemble averaging
In machine learning, particularly in the creation of artificial neural
networks, ensemble averaging is the process of creating multiple models
and combining them to produce a desired output, as opposed to creating
just one model.

Error-driven learning
Is a sub-area of machine learning concerned with how an agent ought to
take actions in an environment so as to minimize some error feedback. It is
a type of reinforcement learning.
Ethical AI
Ethical AI uses artificial intelligence to enhance the human condition by
performing tasks that are menial or impractically slow to accomplish
manually. Key factors in maintaining ethical AI include:
● Explainability – the algorithms used can be understood by
humans, with calculations that can be explained in plain language.
This is essential to verifying that AI is serving its intended
purpose.
● Boundedness – Ethical AI is set to operate within predetermined
boundaries, and does not have the ability to create its own
pathways to exploring or learning unintended information.
● Purpose – Ethical AI is modeled to produce only specific answers
to well-defined problems.

Evolutionary computation
Is an artificial intelligence subfield and closely linked to computational
intelligence, involving lots of combinatorial optimization problems and
continuous optimization.

Explainable AI (XAI)
AI that reveals to human users how it arrived at its conclusions.

Extreme learning machine (ELM)

Is a particular kind of machine learning setup in which a single layer or
multiple layers apply. The ELM includes numbers of hidden neurons where
the input weights are assigned randomly. Extreme learning machines use
the concept of random projection and early perceptron models to do
specific kinds of problem-solving.
Forward Chaining
A method where AI looks back and analyzes the rule-based system to find
the “if” rules, and to determine which rules to use to find a solution.
Game AI
A form of AI specific to gaming that uses an algorithm to replace
randomness. It is a computational behavior used in non-player characters
to generate human-like intelligence and reaction-based actions taken by
the player.

Genetic algorithm
An evolutionary algorithm based on principles of genetics and natural
selection that is used to find optimal or near-optimal solutions to difficult
problems that would otherwise take decades to solve.

Generative Adversarial Networks (GAN)

A type of AI algorithm used in unsupervised machine learning where there
are two neural networks (generator and discriminator) trained on the same
data set. The generator produces output, and the discriminator compares
the output produced with the original data set in efforts to determine
which images are authentic. Based on those results, the generator adjusts
its parameters for creating new output. This process is iterated until the
discriminator is no longer able to distinguish the generator’s output with
the original data set. Used to create photorealistic images.
Heuristics
A technique designed for solving a problem more quickly when classic
methods are too slow, or for finding an approximate solution when classic
methods fail to find any exact solution.

Heuristic function
A function that ranks alternatives in search algorithms at each branching
step based on available information to decide which branch to follow.
Inductive reasoning
A logical process where multiple premises that are true or true most of the
time, are combined to form a conclusion. Often used in prediction and
forecasting.
Machine intelligence
An umbrella term that encompasses machine learning, deep learning, and
classical learning algorithms.

Machine Learning
A subset of AI Computers algorithms that learns from identified patterns in
data, and adjust their actions accordingly, without explicit programming.

Machine Learning Pipeline

Processes that allow for the automation of Machine Learning workflows.

Machine perception
The ability for a system to receive and interpret data from the outside
world similarly to how humans use our senses. This is typically done with
attached hardware, though software is also usable.

Multilayer Perceptron
An MLP can be viewed as a logistic regression classifier where the input is
first transformed using a learnt non-linear transformation. It consists of at
least three layers of nodes: an input layer, a hidden layer and an output
layer.
Natural Language Processing (NLP)
Applying computer algorithms to determine properties of natural human
language in efforts to enable machines to comprehend spoken or written
language.

Neural Network
This is essentially a two- (or more) heads-are-better-than-one approach to
problem solving. Neural networks — designed to be similar to the human
nervous system and brain — help AI solve complex problems by splitting
the work into levels of data. These networks can be used to recognize
handwriting or faces, for example.
Optical Character Recognition (OCR) :
Conversion of images of text (typed, handwritten, or printed) either
electronically or mechanically, into machine-encoded text
Perceptron
An early type of neural network developed in the 1950s. It received great
hype but was then shown to have limitations, suppressing interest in
neural nets for years.

Planning
A branch of AI dealing with planned sequences or strategies to be
performed by an AI-powered machine. Things such as actions to take,
variable to account for, and duration of performance are accounted for.

Pruning
The use of a search algorithm to cut off undesirable solutions to a problem
in an AI system. It reduces the number of decisions that can be made by
the AI system.
Reinforcement Learning
A type of machine learning method inspired by behavior psychology. The
reinforcement learning algorithm (agent) learns by interacting with its
environment and is either penalized or rewarded. The agent seeks to make
decisions to maximize reward over time.

Recurrent neural network (RNN)

A type of neural network that makes sense of sequential information and
recognizes patterns, and creates outputs based on those calculations.

Reactive machines
Can analyze, perceive, and make predictions about experiences, but do not
store data ; they react to situations and act based on the given moment
Strong AI
Is a form of machine intelligence that is equal to human intelligence. Key
characteristics of strong AI include the ability to reason, solve puzzles,
make judgments, plan, learn and communicate

Supervised learning
This is the very serious business of proving things. When you train an AI
model using a supervised learning method you provide the machine with
the correct answer ahead of time. Basically the AI knows the answer and it
knows the question. This is the most common method of training because
it yields the most data : it defines patterns between the question and
answer.
If you want to know why something happens, or how something happens,
an AI can look at the data and determine connections using the supervised
learning method.
Transfer learning
A system that uses previously-learned data and applies it to a new set of
tasks.

Turing Test
A test of a machine’s ability to behave in an intelligent manner that is
indistinguishable from human behavior. Developed by Alan Turing in 1950.
Unstructured data
This is what it sounds like. Disorganized chunks of data that appear
random and unconnected. Examples of unstructured data include email
messages, social media posts, photos, audio files, text messages, satellite
images and webpages
Variance
This refers to the consistency of predictions being produced by a given
model. Low variance indicates that there is not much deviation in
predictions after each iteration, whereas high variance indicates that the
models are varying a lot in terms of results
Weak AI
Also known as narrow AI, weak AI refers to a non-sentient computer
system that operates within a predetermined range of skills and usually
focuses on a singular task or small set of tasks. Most AI in use today is
weak AI.

Weights
A mathematical device used when performing a sum, integral, or average to
give some elements more "weight" or influence on the result than other
elements in the same set. It connects the strength between units, or
nodes, in a neural network. These weights can be adjusted in a process
called learning.