RSS feed LinkedIn Watch us on Youtube

Activity title

Unsupervised Machine Learning in the Military Domain

Activity Reference

IST-ET-113 (AI2S)

Panel

Information Systems Technology

Security Classification

NATO UNCLASSIFIED

Status

Planning

Activity type

ET

Start date

2020

End date

2020

Keywords

Anomalous Behavior Detection, Artificial Intelligence, Big Data, Deep Machine Learning, Unsupervised Deep Learning

Background

Mimicking the human brain to achieve human-level cognition performance has been a core challenge in artificial intelligence research for decades. Humans are very efficient in capturing the most important information while being exposed to a plethora of different stimuli, a capability that is used to represent and understand their surroundings in a concise fashion. Machine learning research has made considerable progress towards cloning such human capability with innovative techniques like deep and feature learning, incremental learning, etc. The recent NATO IST-160 Specialists' Meeting on AI and Big Data for Military Decision Making in Bordeaux and the follow-on IST-173 Specialist Team are a testimony to the enormous interest we, researchers and innovators, currently have to make Artificial Intelligence work for the military. However, training of machine learning systems requires large amounts of annotated and labeled data, and of computing resources.

Objectives

In light of the above considerations, we aim to extend the supervised existing solutions in such a way that they can work in an unsupervised setting too. To do this, we can start from existing neural models (e.g., neural trees, CNN, etc.) and try to adapt them to work in an unsupervised way. For example, we can hypothesize to replace the internal nodes of a neural tree structure with ELM and to consider a different strategy to route any pattern that reaches a specific node to its children. we would like to establish innovative solutions to: • Develop new deep learning architectures able to work in an unsupervised way. • Develop new deep learning algorithms able to work with limited amount of annotated data. • Develop intelligent transfer learning methods for using available (also civil) data sets for military applications. • Reduce the costs for the provision of assets (e.g. data set annotation, QS, …). • Acquire new techniques for incremental learning in military applications.

Topics

Unsupervised Deep Learning, Intelligent Transfer Learning, Incremental Learning in military applications, Data security, Predictive analysis, Detection of threats in systems and networks, Prediction of enemy‘s behavior.

Contact Panel Office