Glossary
Below is an alphabetized list of technical terms and features used in Reality AI Tools, with detailed definitions to help you understand their usage across the platform.
A
AI Explore
Reality AI Tools function that analyzes data sample lists and automatically compares feature spaces and decision structures to suggest the best model architectures. This is the typical starting point for most users.
Anomaly Detection
A semi-supervised learning method to identify data patterns that deviate from normal behavior. Only normal samples are required for training; however, results can be automatically optimized when counter examples are also available.
B
Base Tool
Defines the foundation of an AI model, including feature space, decision structure, and inference target. A Base Tool is trained on specific data to create a Trained Tool. An initial Base Tool and Trained Tool are usually created from an AI Explore run, providing the foundation for further training and refinement.
C
Cortex Microcontroller Software Interface Standard – Neural Networks (CMSIS-NN)
A collection of optimized neural network kernels, used to accelerate embedded model inference. These may be leveraged when deploying a model to an ARM core device.
Confusion Matrix
A visual and tabular tool used to evaluate classification performance by comparing predicted labels to actual labels. See how to read a confusion matrix for more information.
Convolutional Neural Network (CNN)
A Neural Network that includes convolutional layers. Such networks can learn feature structures within the data that are not directly extracted by the feature space itself. CNNs have increasingly larger footprints and slower inference times compared to other model types.
Curate
The process of organizing your data files for use in training and testing AI models. This includes removing bad data, grouping and labeling data, and preparing raw sensor data by segmenting it into samples. See Data Readiness and Curate sections.
D
Data Sample List
A set of smaller data frames derived from raw files through segmentation, used in model training. Each frame comprises one test sample as seen by the inference model, and results are tracked and tabulated accordingly. Also called a “Segmented List.”
Decision Structure
The inference architecture of an AI classifier model, such as Neural Network, Convolution Neural Network, Support Vector Machine, or K-means, together with its internal structure, including network layers in the case of NN or ensemble strategy in the case of SVM.
Deployment
The process of packaging and exporting trained models to the cloud or embedded systems for real-world use. See Embedded section.
E
e² studio
An Integrated Development Environment (IDE) provided by Renesas Electronics. It is specifically designed for developing embedded applications on Renesas microcontrollers and microprocessors.
Energy Triggered
A segmentation method in Reality AI Tools that divides data into samples based on energy patterns or thresholds in the input signal rather than using a fixed sliding window approach. Click here to learn how to create a Data Sample List using energy triggering.
Explorer Tier
A free trial account that allows users to test Reality AI Tools using a limited set of features and data.
F
Feature Space
A set of derived attributes used to describe input data during model training, such as frequency components or statistical attributes. See Feature Spaces page for more information about those used by Reality AI Tools.
File Level Metadata
Descriptive labels assigned to individual files within your dataset. See Metadata.
Full-Access Account
A paid Reality AI Tools subscription offering unlimited access to all features including uploads, downloads, and model deployment.
K
K-Fold Validation
A cross-validation technique where the dataset is divided into K parts to evaluate model accuracy over multiple training/testing splits. Reality AI Tools applies 10-fold testing during AI Explore and enables users to set up their own k-fold tests in the Validation page.
K-Means
A clustering-based model decision structure that partitions data into K distinct groups based on similarity, typically using distance metrics. It is an unsupervised method that can be useful for anomaly detection and pattern recognition, especially when class labels are not available. K-means has a compact footprint and efficient inference, making it suitable for resource-constrained embedded deployments.
M
Metadata
Metadata refers to descriptive labels assigned to subsets of your data—such as group numbers, dates, machine sources, or colors—intended for reference and contextual understanding. Metadata can be defined at the file level (File-Level Metadata) or at the sample level, within blocks of data in a file (Sample-Level Metadata). A Target Class or Target Value is considered a special type of metadata.
N
Neural Network (NN)
A model decision structure based on connecting layers of neurons with learned weights that may have better performance for use cases with more complex class structures, at the cost of a larger deployed footprint.
O
One-vs-all
An SVM ensemble structure in which only one learner is created for each individual class versus all the remaining classes.
One-vs-one
An SVM ensemble structure in which one learner is created for each possible combinatorial pair of classes.
Optimize BOM
A set of Reality AI Tools functions that help the user optimize the hardware and sensor costs by finding critical limits of AI models. This includes Sensor Channel Selection to reduce sensor counts, as well as Sensitivity and Tolerance tests to determine minimum necessary sample rates, bit depths, and noise floors for reliable operation.
P
Project
A container within Reality AI Tools that holds datasets, models, configurations, and results for a specific use case. We recommend users organize their data and models by project.
R
Reality AI Tools
A cloud-based platform by Renesas for developing, training, validating, and deploying AI models tailored for embedded systems.
Remaining Useful Life (RUL)
A regression model output that predicts how long a component will operate before failure.
S
Sample Rate
The frequency, typically in Hertz (Hz), at which data points are collected from a sensor.
Sample Level Metadata
Descriptive labels assigned to blocks of data within your files. See Metadata.
Secure File Transfer Protocol (SFTP)
Secure File Transfer Protocol used to upload large datasets to Reality AI Tools through encrypted connections. Click here to learn how to use this feature.
Segmented List
A set of smaller data frames derived from raw files through segmentation, used in model training. Each frame comprises one test sample as seen by the inference model, and results are tracked and tabulated accordingly. Also called a “Data Sample List.”
Sliding CSV Window
A segmentation strategy that creates a data samples list by using a moving window across time-series data at a constant spacing or stride. These windows may be overlapping or non-overlapping. Click here to learn how to create such a list.
Support Vector Machine (SVM)
A model decision structure based on partitioning data with hyperplanes that, for many use cases, results in an extremely compact embedded footprint while still retaining good generalization properties.
T
Target
Each AI inference model has a target. Depending on the type of model, the target may be a Target Class or a Target Value. Data Sample Lists are labeled with a target that is compatible with the AI model type they will be used with.
Target Class
A categorical label used in classification tasks to denote the state or condition of the sample.
Target Value
A numerical value used in regression tasks representing quantities like temperature or pressure.
Trained Tool
An AI model derived from a base tool after training with user-supplied data. A Trained Tool includes the base tool architecture plus a training checkpoint. It may also include additional post-processing aspects such as Error Balance adjustment and Smoothing. Trained Tools fully define an AI inference model and are the basic unit within Reality AI Tools for testing, optimization, and deployment.