Real-time Machine Learning is not available to all users yet. This feature is in alpha and still being tested. This document is subject to change.
Real-time Machine Learning can dramatically enhance the relevance of your digital experience content for your end-users. This is made possible by leveraging real-time inferencing and continuous learning on the Experience Edge.
A combination of seamless computation on both the Hub and the Edge dramatically reduces the latency that is traditionally involved in powering hyper-personalized experiences that are both relevant and responsive. Hence, Real-time Machine Learning provides inferences with incredibly low latency for synchronous decision-making. Examples include rendering personalized web page content or surfacing of an offer or discount to reduce churn and increase conversions on a web store.
The following diagrams provide a overview for the Real-time Machine Learning architecture. Currently, alpha has a more simplified version.
The following workflow outlines the typical steps and results involved in creating and utilizing a Real-time Machine Learning model.
Data is ingested and transformed with the Experience Data Model (XDM) on Adobe Experience Platform. This data is used for model training. To learn more about XDM, visit the XDM overview.
Create a Real-time Machine Learning model by authoring it from scratch or bringing it in as a pre-trained serialized ONNX model in Adobe Experience Platform Jupyter Notebooks.
Deploy your model to Experience Edge to create a Real-time Machine Learning service in the Service Gallery using the Prediction API endpoint.
Use the Prediction REST API endpoint to generate machine learning insights in real-time.
Marketers can then define segments and rules that map Real-time Machine Learning scores to experiences using Adobe Target. This allows for visitors of your brand’s website to be shown a same or next-page hyper-personalized experience in real time.
Real-time Machine Learning is currently in alpha. The functionality outlined below is subject to change as more features and nodes are made available.
df.valuesis called it returns an array that is acceptable by your DL model. This is because the ONNX model scoring node uses
df.valuesand sends the output to score against the model.
|Features||- Using the RTML notebook template, author, test, and deploy a custom machine learning model.
- Support for importing pre-trained machine learning models.
- Real-time Machine Learning SDK.
- Starter set of authoring nodes.
- Deployed to Adobe Experience Platform Hub.
|Authoring Nodes||- Pandas
|Scoring run times||ONNX|
You can begin by following the getting started guide. This guide walks you through setting up all the required prerequisites for creating a Real-time Machine Learning model.