Introduction to Machine Learning and Artificial Intelligence 301

activation function

A logical rule that determines how nodes in a neural network calculate inputs and produce outputs. Activation functions determine a neural network's level of complexity.

agent

The entity in a reinforcement learning environment that performs actions. The agent in reinforcement machine learning is the artificial intelligence seeking to learn the correct actions to achieve a goal.

algorithms

A logical or mathematical process designed to systematically solve a problem. Algorithms are coded into software to form the rules by which artificial intelligence functions.

anomalies

An irregularity in a data set that does not fit with general trends or patterns in the data. Anomalies in a data set could indicate erroneous data or outliers that can be studied to learn new information.

artificial intelligence

AI. The ability of a machine or computer to imitate intelligent human behavior. Artificial intelligence allows machines to perform a process with autonomy.

artificial neural networks

A web of connected nodes designed to resemble the structure and function of neurons in the human brain. Artificial neural networks have been key to building advanced machine learning models for artificial intelligence.

artificial neuron

A conceptual model for artificial intelligence designed to resemble the basic structure and function of a neuron in the human brain. An artificial neuron stems from the idea that a computing device could process information similar to a human.

association

The process of analyzing a data set to discover unknown relationships between items in the data. Association algorithms are effective for analyzing unlabeled data in unsupervised machine learning.

autonomous

Self-governing. Autonomous systems can be programmed to make decisions without human interaction.

best fit line

A straight line on a regression chart that cuts through all plotted data points most accurately. The best fit line is drawn using input and output variables from the training set of data and is used to predict the output variable of the testing set of data.

binary classification

The process of analyzing and dividing inputs from a data set into two groups based on specified conditions that are either true or false. Binary classification is often the goal of supervised machine learning.

binary classifier

An algorithm that separates data into two categories using either a 1 or a 0 to define each group. Binary classifiers can be used to simplify more complex data into binary categories.

binary digit

The smallest unit of information on a computing device. Binary digits are either 1s or 0s.

causal relationship

A dynamic in which one variable is directly affected by another. Causal relationships between variables can be depicted on a regression chart.

classification

A set of steps used to analyze or predict whether or not a data output falls in a specified category. Classification algorithms can be used in supervised machine learning models.

clean

Refers to information that is free from errors, inconsistencies, missing values, and irrelevant features. Clean data ensures that machine learning models can learn patterns accurately.

clustering

The processes of analyzing a data set to discover similar patterns, then dividing the data into groups based on these patterns. Clustering algorithms for unsupervised machine learning allow an artificial intelligence to analyze large, unlabeled data sets.

clustering algorithm

A set of steps or calculations used to analyze an unlabeled data set in order to create groups within the data based on trends. Clustering algorithms can be used in unsupervised machine learning to discover unknown information about the data.

computer vision

A deep learning capability that enables an artificial intelligence to detect and identify objects by recognizing visual patterns through a camera or on a screen. Computer vision can be useful for supervised or unsupervised machine learning but requires very large amounts of visual data.

confusion matrix

A table used to evaluate the performance of a classification model by comparing the actual and predicted labels. Confusion matrices display the counts of true positives, true negatives, false positives, and false negatives, helping to assess metrics like accuracy, precision, recall, and F1 score.

continuous variables

Input or output data that can be any numerical value. Continuous variables are more specific than other types of variables.

data mining

The process of analyzing data with software in order to draw out unknown patterns in the data. Data mining is used to discover useful information in large sets of unlabeled, unorganized data.

data patterns

Similarities or trends discovered among items within a data set. Data patterns are key to enabling machine learning for artificial intelligence.

data scientist

An individual who uses computing systems to perform data analysis in order to interpret and present insights gained from data. Data scientists can help manufacturers operate more efficiently by adjusting processes based on data analysis.

decision trees

A binary classification method in which data is divided based on whether conditions for the data are either true or false. Decision trees can convert continuous variables into binary outputs.

deep learning

Deep learning: A subset of machine learning that uses multilayer neural networks with multiple hidden layers. Deep learning algorithms can enable machines to exhibit advanced, humanlike behaviors but require significant amounts of data.

dependent variable

Represents the output data that may be impacted by various input factors. The dependent variable changes in response to the independent variable.

digital

Consisting of information that is input or output electronically as a series of pulses or signals, often resulting in binary strings of 0s and 1s. Digital computing devices interpret various programming commands as binary digits.

F1 score

A metric used to assess the accuracy of a machine learning classification model. F1 scores make their evaluation by balancing precision and recall.

F1 score

The F1 score is a metric used to evaluate the performance of a classification model. F1 scores use precision and recall, providing a single score that accounts for both false positives and false negatives.

hidden layer

The middle row or rows of nodes in an artificial neural network that perform specified processes and calculations on the data stored in the input layer. Hidden layer nodes typically contain different weights and thresholds that affect each node’s calculated result of the inputs.

historical data

Previously recorded data for which all data outputs are known. Historical data can be used to train machine learning models to predict outputs in unknown data sets.

IF-THEN-ELSE command

A conditional command that executes when a particular portion of a program is met, and executes an alternate instruction if the same condition is not met. If-Then-Else commands are commonly used in computer programs.

independent variable

The input data in a regression algorithm that may influence the output, or end result. The independent variable causes the dependent variable to change.

input data

Information given to a computing device in order to be processed or analyzed. Input data is used by machine learning algorithms in order to produce an output.

input layer

The first row of nodes in an artificial neural network that store the collected data to be analyzed by the machine learning algorithm. The input layer can contain one or several input nodes.

iterative

Repeating actions and steps in processes in order to apply improvements or changes. Iterative design is an important aspect of smart optimization because it allows for constant and quick changes in the early adoption stages.

labeled data

A set of digital information that is defined prior to being analyzed by a computer. Labeled data is required in supervised machine learning algorithms.

linear regression

A statistical analysis process that models the relationship between one input or independent variable and one output or dependent variable to examine the correlation between them. Linear regression can be used in supervised machine learning to help an artificial intelligence predict outputs based on a single input.

linear regression chart

A chart or graph that depicts data points and a best fit line produced by a linear regression algorithm. Linear regression charts help data scientists visualize the causal relationship between a dependent variable and an independent variable.

machine learning

ML. The process that enables a digital system to analyze data in order to build predictive models and make decisions autonomously. Machine learning is a key benefit of smart manufacturing.

machine learning model

ML model. A computing function that uses one or more algorithms to perform specific calculations on data to produce an output. Machine learning models may use supervised, unsupervised, or reinforcement learning methods.

ML

Machine learning. The process that enables a digital system to analyze data in order to build predictive models and make decisions autonomously. ML is a key benefit of smart manufacturing.

ML model

Machine leaning model. A computing function that uses one or more algorithms to perform specific calculations on data to produce an output. ML models may use supervised, unsupervised, semi-supervised or reinforcement learning methods.

model drift

The degradation in a machine learning model’s performance over time due to changes in the underlying data distribution or relationships between features and target variables. Model drift can result from evolving real-world conditions, making previously learned patterns less relevant or inaccurate.

natural language processing

NLP. A deep learning capability that enables an artificial intelligence (AI) to detect and identify language patterns in audio speech or written text. Natural language processing requires very large amounts of data.

neurons

A conceptual model for artificial intelligence designed to resemble the basic structure and function of a neuron in the human brain. Neurons stemmed from the idea that a computing device could process information similar to a human.

nodes

A single unit in an artificial neural network that receives an input and produces an output. Nodes in a neural network can store information or perform a variety of actions.

noise

Random or extra variations in a dataset that distract from true data patterns. Noise in machine learning can degrade model accuracy by introducing misleading signals during training and evaluation.

normalize

The process of adjusting the scale of data input features so they have a consistent range or distribution. Normalizing techniques help prevent features with larger values from dominating the learning process.

output

A calculated result or action produced by a computing device after processing data inputs. Outputs produced by computing devices may be in the form of data or may include a response to data.

output data

The information that a computer system produces and sends to an output device after processing input data. Output data can take various forms such as text, images, audio, or signals.

output layer

The last row of nodes in an artificial neural network where the final processed result is produced. The output layer may contain one or more nodes depending on the algorithm's complexity.

outputs

A calculated result or action produced by a computing device after processing data inputs. Outputs produced by computing devices may be in the form of data or may include a response to data.

overfitting

An undesirable characteristic of a machine learning model in which the output is too specific and cannot be applied to other sets of data. Overfitting is corrected by adding regularization or reducing the complexity of the data sets used as inputs in the model.

positive reinforcement

Anything that is presented after the occurrence of a behavior to increase the likelihood of that behavior occurring again. Positive reinforcement is used in reinforcement learning to increase the chances of correct outputs.

principal component analysis

PCA. An analysis technique that helps simplify complex data and reduce noise. Principal component analysis improves processing efficiency while maintaining essential patterns for machine learning tasks.

raw data

Unstructured digital information produced directly from a data source that has not been analyzed. Raw data is produced in immense amounts from manufacturing processes and collected by sensors.

regression

A set of steps and calculations used to analyze or predict how inputs or independent variables impact outputs or dependent variables. Regression algorithms can be used in supervised machine learning models.

reinforcement machine learning

A process in which an artificial intelligence (AI) learns the best actions to achieve a goal through trial and error. Reinforcement machine learning enables an AI to learn details about inputs and produce correct outputs autonomously.

return on investment

ROI. A performance measure used to evaluate the efficiency or profitability of an investment. Return on Investment is the benefit of an investment divided by the cost of the investment.

reward

An instance of positive reinforcement given to an agent for carrying out a correct action in reinforcement learning. Rewards are used to train artificial intelligence to produce correct actions in a simulated environment.

scale

A uniform size decrease or increase that allows a large object to be accurately depicted in a smaller form or a small object to be accurately depicted in a larger form. Scaling an object up or down in size changes the size of the object but not the relationship of its dimensions to one another.

semi-supervised machine learning

A process where machines use both labeled data and unlabeled data to define desired outputs. Semi-supervised machine learning uses a combination of supervised and unsupervised machine learning.

sensors

A device that detects a change in a physical stimulus and turns it into a signal that can be measured or recorded. Sensors may be connected to a machine or system in order to collect operational data that is later analyzed.

simulated environment

A practice setting designed to mimic or resemble the environment in a real-life situation. Simulated environments can be physical or virtual environments.

single-layer perceptron

An algorithm first proposed by Frank Rosenblatt designed to resemble the function of a human neuron. A basic single-layer perceptron receives an input and produces a simple true or false output.

smart

Capable of sending and receiving data without human intervention. Smart technology generally requires internet connectivity to enable data processing.

speech recognition

The ability of an artificial intelligence (AI) to interpret audible speech by identifying speech patterns. Speech recognition technology requires complex machine learning algorithms.

state

The current status of a simulated learning environment that defines available actions. State changes in the environment can change what actions are available.

step function

An activation function that assigns threshold values to nodes in a neural network. A step function causes each node to produce a binary output.

sum

The total amount resulting from adding two or more numbers together. The sum of weighted values calculated by a node in a neural network must be above the node's assigned threshold value in order for the node to activate.

supervised machine learning

A process in which a human operator labels data inputs for a machine learning model and defines the desired outputs the model should produce. Supervised machine learning typically requires extensive human labor in order to prepare and label data sets.

target marketing

A marketing tactic that tailors different marketing messages to different customers based on purchasing habits or other information. Target marketing can be greatly enhanced using machine learning algorithms to analyze customer data.

target result

A known output value in the test data set that a machine learning model is designed to produce after calculating the test data inputs. Machine learning models that produce outputs close to the target result are considered effective models.

test data

A group of data for which the output variable is not known. A test data set is used to test the accuracy of a machine learning algorithm.

threshold value

A numerical limit, typically between zero and one, that is set for each node in the hidden layers of a neural network. The threshold value activates a node if that node calculates a weighted sum that is above the threshold.

training data

Previously recorded data used to train and improve machine learning algorithms. The training data for supervised learning includes historical data for both inputs and outputs.

Turing Test

A test designed by Alan Turing that could theoretically be used to rate the intelligence of a computer. The Turing Test essentially rated a computer's responses to text-based questions on whether the computer's response could be differentiated from a human's.

underfitting

An undesirable characteristic of a machine learning model in which the output is inaccurate and cannot be applied to any data sets. Underfitting is corrected through trial and error with additional data to refine the output so it draws closer to a target output.

unlabeled data

A set of digital information that is not defined prior to being analyzed by a computer. Unlabeled data can be analyzed using unsupervised machine learning algorithms.

unsupervised machine learning

A process in which a machine learning model processes and analyzes unlabeled data and produces outputs without human interaction. Unsupervised machine learning models help discover patterns in data that humans may not recognize.

weight

A specific numerical value assigned to a vector connecting the output of one node in a neural network to the input of another node on the next layer. Weights allow neural networks to assign different levels of importance to different inputs.

weighted values

A node's output value multiplied by its assigned weight. Weighted values are used as input values for connected nodes on the next layer.