H2O AI Cloud key terms

AI Unit: A measure of the consumption of H2O AI Cloud by end users. For example, if the platform software is running no apps, models, or AI Engines, then 0 AI Units are being used.
Managed Cloud: A managed offering featuring a complete set of H2O Hybrid Cloud capabilities on a fully managed cloud environment that handles infrastructure provisioning, scale, and software updates. H2O Managed Cloud is designed for high availability and provides tools to develop, deploy, manage and maintain AI-based applications.
Hybrid Cloud: A deployment option where the customer installs and manages the platform on their own cloud / on-prem infrastructure.

For more information, see H2O AI Cloud offerings.

Full access: Users who can do almost anything in the H2O AI Cloud including see all ALL_USERS apps, see all ALL_USERS instances of ALL_USERS apps, and upload their own apps to the App Store.
Visitor: Any user who does not have the FULL_ACCESS role.
Admin: Users who have access to the H2O admin commands of the CLI and who are able to see the admin pages on the top-right corner of the H2O AI Cloud UI. These users may be either FULL_ACCESS or VISITOR users. VISITOR admins will not be able to see or run apps not explicitly granted to them from the App Store. However, they will be able to view all apps using the admin CLI.

App: Runnable bundles of application code and metadata for how to run this code. Apps on H2O AI Cloud are built on the Wave software development kit.
App instances: A specific instance of an app in the H2O AI Cloud. Instances can be running or paused.
Managed instance: An instance of an app in the H2O AI Cloud that is created and managed by the app owner or administrators. An app with a managed instance lifecycle only allows the app owner or administrators to create a new instance of the app.
AI Engine: A software platform for building machine learning models. Usually, but not always, this is AutoML-related.
The H2O AI engines are H2O Driverless AI, H2O-3 Open Source, and H2O Hydrogen Torch (which is currently an app and users run app instances via the App Store). Users cannot create their own AI engines; these are provided by H2O.
AI Engine instances: A specific instance of an AI engine in the H2O AI Cloud. AI instances can be running or paused.

Machine Learning terms

Supervised Learning: A machine learning mechanism that uses 'labeled' training data to train the algorithm. This means that some of the input data is already tagged with the correct output data so that the algorithm can use these accurate examples as a means of predicting other data.
Unsupervised Learning: A machine learning mechanism that analyzes and identifies patterns using the whole dataset. The most common form of unsupervised learning is identifying and clustering the data into groups, however unsupervised learning can also involve detecting outliers or anomalies in the dataset, or finding relationships between observations.
Semi-supervised Learning: This is a mix of supervised and unsupervised learning which uses a small amount of labeled data and generally a larger amount of unlabeled data.
Deep Learning: A machine learning mechanism that uses neural networks of three or more layers that attempt to imitate the learning behaviour of the human brain and learn by example. It is usually used for large datasets where the inputs are less structured (i.e., large amounts of unstructured text or images).
AutoML: A mechanism that automates repetitive machine learning tasks involved in building an AI model.

AI model/ modeling: Modeling is the process of extracting data insights from your dataset, and a model is a computed representation of the relationships and trends between the features in your input dataset. The model can then be used on similar datasets to help make business decisions based on the outcomes of the model.
Feature: A measurable column within your input dataset that can be used for analysis.
Feature engineering: A machine learning technique to generate additional derived features in a dataset in order to improve a model's accuracy.
Feature transformation: Applying a mathematical formula to data within a measurable column in your dataset to transform the values and create a data column that can be used by machine learning algorithms for further analysis.
AI experiment: An AI experiment involves adjusting the parameters of your model to leverage the data insights you have in order to test and assess the performance and accuracy of your model.
Algorithm: Machine learning algorithms are mathematical model mapping methods used to learn or uncover underlying patterns embedded in the data.
Pipeline: An automated linear sequence workflow that specifies the data preparation options, modelling operations, and transformation operations.

Model deployment: Integrating a model into the existing production environment.
Drift: Situations where the input values for features during scoring differ from the input values for features during training. When a drift score increases, it means that the model is seeing data that it was not trained on, and so the performance and results of the model may not be accurate.

For details on these components, see What is H2O AI Cloud.

Feedback