ITS Artificial Intelligence Certification Key Terms'

A/B Test

A test in which different app versions are shown to different people and usage patterns and responses are compared to each other.

Accuracy

The percentage of correct predictions within a model.

Adversarial Testing

A test in which data is intentionally manipulated to evaluate an AI system's vulnerability to adversarial attacks or misleading inputs.

AI

A branch of computer science focused on creating systems capable of performing tasks that normally require human cognition, such as learning, reasoning, and problem-solving.

Algorithm

A set of rules used to generate calculations or perform problem-solving operations.

Anomaly

An unusual activity in a group of otherwise normal activities.

API

A set of rules that allows applications to communicate with each other.

Assumption

A belief or condition taken to be true for a system to work as intended.

Attack Surface

The parts of data or a system that are vulnerable to attack.

AUC

A metric that measures a model's probability of correctly ranking a positive instance higher than a negative instance, calculated from the region beneath the receiver operating characteristic plot.

Azure Machine Learning

A cloud-based tool used to build AI models.

Bias

An imbalance in data that can cause data to be skewed toward a demographic group, which can harm an AI machine learning model.

Binary

A form of data that stores data as ones and zeros.

Classification

A model type in which a result is categorized, usually with one of two values, such as a yes or no value.

Concept Drift

The change in the underlying concepts used to train AI models.

Confusion Matrix

A chart that defines true and false positives and true and false negatives for an AI model.

Constraint

A limitation or restriction on an AI system.

Correlation

A measure of connection between two points of data.

Data at Rest

Data stored on an on-premises or cloud storage drive.

Data Drift

The change in the statistical properties of the input data within an AI model.

Data in Transit

Data in the process of being copied or moved from one location to another.

Data Type

A characteristic of data, such as numeric, string, or date.

Decision Tree Algorithm

A machine learning method in which samples are split into two or more sets of data based on input variable differentiators.

Decommission

The act of no longer using an AI model in production due to its lack of effectiveness.

Deep Learning Algorithm

An algorithm used for neural networks, which work with AI models for human-based interaction.

Degraded Mode of Operation

A mode of operation in which an AI model is not performing to its maximum capabilities.

Derived Feature

A feature built using a formula, through splitting or combining data, or through assigning keywords to data based on conditions.

Drift Detector

A mechanism that identifies shifts in the underlying data used in an AI solution.

Edge Case Testing

A test in which an AI system's performance is pushed beyond its normal operating range.

Explainability Requirement

A clear explanation for a decision, prediction, or recommendation made from an AI model.

F1 Score

The harmonic mean of precision and recall.

False Negative

An incorrect negative result for an AI model.

False Positive

An incorrect positive result for an AI model.

Feature

An input to an AI model, used to generate output.

Feature Vector

An ordered list of numerical properties of observed phenomena.

Feedback Mechanism

A means in which feedback is collected for an AI system, whether that be through forms, a website, or directly within the system.

GDPR

The European Union (EU) regulatory framework governing how personal data must be collected, processed, and stored, applicable to all companies doing business in the EU.

GitHub

A repository used to store code for development projects.

Imbalanced Data

A dataset that has too much data from one or more groups of data.

Interpretability

The ability for one to understand the results of an AI model.

Iteration

A sequence of training an AI model, often needing to be repeated to fully train an AI model.

JSON

A lightweight, human-readable data interchange format that stores information in key-value pairs and ordered lists, commonly used for transmitting data between a server and a web application.

K-Means Clustering

A machine learning model that splits data into an unspecified number of groups.

KPI

A measurable metric used to set and evaluate performance standards, often visualized on dashboards and, in the context of machine learning, used to gauge model effectiveness.

Labeled Data

Data that is paired with output targets in an AI model.

Linear Regression

A form of regression in which an outcome is a continuous variable, like a number.

Log

A list of activities in an AI model and can be used to identify potential security problems with a model.

Logistic Regression

A form of regression in which an outcome uses binary dependent variables.

Model

An app used to generate predictions and outcomes based on the entering and training of data entered into the app.

Natural Language Processing

A form of AI that reads data from text and images and can include speech recognition and object detection.

Neural Network

A type of algorithm designed to mimic the human brain.

One-Hot Encoding

A form of encoding that assigns categorical variables to binary numbers.

OpenRefine

A tool used to cleanse data to ready it for an AI learning model.

Overfitting

A situation in which a model learns training data very well but performs poorly on new, unseen data.

Parameter

An input value used to help train an algorithm.

Performance Metrics

Data gathered on an AI model and used to ensure that an AI model is working properly.

Pipeline

A series of steps used to transform data used for an AI model into a final prediction or outcome.

Power Virtual Agents

A Microsoft tool used to build chatbots.

Precision

The percentage of true positive predictions among all positive predictions in an outcome.

Predicate

A logical statement or condition defining properties or relationships between different entities in an AI system.

R-Squared

A value that measures the proportion of variance of a dependent variable that is predictable from independent variables.

RBAC

A permission standard in which access privileges are assigned to defined roles, and those roles are then granted to individuals and groups rather than assigning privileges directly.

Recall

The measure of true positive predictions among all actual positives.

Regression

A model type that looks at the relationship between a dependent variable and one or more independent variables.

Reinforcement Learning

A type of machine learning model that involves an AI agent performing actions that maximize a certain reward.

Risk Register

A document that contains known risks for a project and the impact, probability, and mitigation strategy for those risks.

ROC

A diagnostic plot that graphs the True Positive Rate against the False Positive Rate at various classification thresholds, used to evaluate model performance.

Scalability

The ability to adjust machine resources up or down depending upon workload demand for one or more apps on that machine.

Self-Fulfilling Prophecy

A phenomenon to where biased data perpetuates a wanted or perceived bias within an AI model.

Sensitivity

The relative change of a model and its output.

Sentiment Analysis

A form of machine learning in which data is read and assigned a value for its positivity or negativity.

SME

A domain specialist who provides authoritative knowledge about the problem or problems being addressed through the development and use of a machine learning model.

Specificity

The process of evaluating a model's ability to accurately identify negative instances or nonrelevant outcomes.

Split Ratio

The ratio of data being used for training an AI model versus the data used for testing the model.

SQL Server

Microsoft's database server and often used to store data used to generate a pipeline for an AI model.

Stakeholder

An individual, group, or organization that may affect or be affected by a project.

Supervised Algorithm

An algorithm that uses labeled data, which means data where a target variable is known.

Tokenization

The process of converting words to numbers to be used for data in an AI model.

Transparency

In the context of AI, the act of being forthright about how data is collected and used for AI models.

Trend

A pattern of data within an AI model.

Trifacta Wrangler

A tool used to cleanse data to ready it for an AI learning model.

True Negative

A correct negative result for an AI model.

True Positive

A correct positive result for an AI model.

Underfitting

A situation in which a model does not perform well on training data or new, unseen data.

Unlabeled Data

Data that is not paired with output targets in an AI model.

Unsupervised

A type of machine learning model that looks for hidden patterns or structures in data.

Unsupervised Algorithm

An algorithm that does not use labeled data.

Usage Metrics

The extent to which people use an AI model to its fullest.

User Acceptance

The want of people to use an AI model and the outcomes it produces.

Visualization

A chart or report that explains the results of an AI model or the data used to build an AI model.

Based on the official exam objectives and the technical concepts often tested in the ITS AI Specialist certification, here are the critical missing terms formatted for direct import into your flashcard set.

Copy this block and add it to the bottom of your existing import list:

Backpropagation

The algorithm used in neural networks to calculate the gradient of the loss function, essential for updating model weights.

Epoch

One complete pass through the entire training dataset during the learning process.

Hyperparameter

Configuration settings used to tune the learning process (e.g., learning rate) that are set before training begins.

Generalization

The ability of a model to perform accurately on new, unseen data rather than just the training data.

Stochastic Gradient Descent

An optimization algorithm used to minimize errors by iteratively updating weights based on a subset of data.

Activation Function

A mathematical gate (e.g., ReLU, Sigmoid) that determines whether a neuron should be activated, allowing the model to learn non-linear patterns.