The present Deliverable 5.5 “Meta-learning Report” (henceforth referred to as D5.5) is related to Task 5.3 “Implement meta-learning processes” (henceforth referred to as T5.3). The objective of this task is to create a Machine Learning (ML) component where automatic learning algorithms are applied on meta-data related to machine learning experiments (as defined in the D1.3)1.
This module will be integrated with the rest of the components of the RED-Alert solution according to the below schema:
According to this schema, the goal of the Meta-leaning module is to:
Improve the Data Acquisition module, by suggesting new keywords to detect new content.
Update the word vectors of the NLP module, to improve the accuracy of the NLP Deep Learning threat score classifier (see Deliverable 2.2).
Improve the SNA module with new links, by updating the networks with new information (see Deliverable 3.1).
Update the CEP module with new patterns / rules to detect more radical content (see Deliverable 4.2).
The complexity of the component requires it to be developed from scratch in an iterative manner. During the first iteration, reported in this document, a plan for development has been defined and the part related to the Deep Learning (DL) model has been started to develop.
The next iterations will involve the following activities:
1) During the following months, a demo of this DL model will be created, in order to demonstrate the usefulness of the module.
2) Each of the leaders of the WPs that are updated by the Meta-learning processes will create the mechanism related to their tasks:
INT, the mechanism to improve the NLP Ontologies by the users.
ELTE, the improvement of the SNA module by the user’s feedback.
CITY, the mechanism to update the Data Anonymization Tool, including new data and data formats.
ICE, the method to update the CEP engine.
3) All these mechanisms will be integrated into the Meta-learning module by INSKT.
4) In parallel of these tasks, a UI specific for the meta-learning process will be created, in order to create a way to get the information from the users.
5) Finally, SIV will integrate this module into the RED-Alert solution.