Personalization Enablers
Target audience: software developers
Personalization Enablers proposed in the XR2Learn ecosystem aim to personalize education scenarios in XR by enabling adaptive learning components that dynamically adjust to users based on their proficiency level, affective state (emotions), and challenge level of an educational scenario.
Personalization Enablers are organized into four main domains based on their functionalities. The figure below illustrates a high-level diagram of the proposed components. All the components within the different domains are cross-platform applications that can be hosted on a local or remote machine.
A modularized software engineering architecture approach was utilized to deploy enablers to foster scalability, flexibility, and dynamic network topology. Different components can be deployed in separate machines, allowing for a heterogeneous and dynamic deployment of the components. This is essential to provide the necessary computation resources for each component. For instance, deep learning training components need heavy computational resources. With the proposed enabler’s architecture, these heavy components can be deployed in a more robust computational machine. On the other hand, other components that do not require as many resources can be deployed in other machines and they can all communicate with each other.
In the proposed architecture, four domains have been implemented to cover the ER enablers' functionalities:
The Training domain covers all enablers associated with Deep Learning model training. Specifically, under the context of the Training domain, Enabler 2 was implemented to provide tools for pre-processing, handcrafted feature extraction, and learning emotion representations using self-supervised learning techniques. The emotion representations are lower-dimensional descriptive features extracted from data in an automated manner by optimized Neural Networks. Moreover, Enabler 3 is delivered as a set of tools to be used to extract features (pre-trained or handcrafted) from raw data. Finally, pre-trained models can later be used to build emotion classifiers given annotated data with emotions as required by Enabler 4.
The Personalization Tool exploits the outputs of Training and Inference tools to provide personalization of XR scenarios to the users interacting with them. Utilizing the user’s predicted emotions as the output of the Training and Inference domain, together with contextual information, e.g., a user and activity difficulty (challenge) levels, the personalization tool provides personalized suggestions on the recommended activity level for the user in educational XR applications.
The Personalization Tool exploits the Publisher/Subscriber messaging protocol implemented using Redis to provide asynchronous, real-time communication between the Personalization Tool, Inference domain and an XR educational software implemented using Unity.
Command-line interface (CLI) is an automated interface that facilitates access to the enablers’ functionalities. It allows users to quickly and easily access the enablers’ use cases. CLI includes simplified installation and commands, pre-configured scripts for common use cases, and benchmarks to evaluate the end-to-end workings of the whole pipeline, working as an integration test for the system.
|
Wiki - Immerse yourself in the world of XR2Learn |