When people think about artificial intelligence (AI) and deep learning, images of rogue computer programs, ready to take over humanity, might come to mind. However, the reality is less dramatic but more exciting. In recent years, deep learning has resulted in improvements in virtual assistants, self-driving cars and more. One area that is making advances is anomaly detection.

Our team, led by Charlie Abela from the Department of Artificial Intelligence within the Faculty of Information & Communication Technology (ICT) at the University of Malta, in collaboration with Parallels (Corel Malta Ltd.) and supported by the R&D (2014-2020) Malta Enterprise incentive, is working on a project to detect anomalous behaviours. Anomaly detection aims at finding outliers within data. Such outliers could lead and have been used to identify errors within a system, performance issues and unusual behaviours.

Machines maintain log files to keep track of important information and activities that are occurring within a system. Log files are automatically created and updated by the system, depending on the level of detail required by the engineers managing the system. Such files are important since any unauthorised, unusual or anomalous behaviour can be recorded and detected by engineers to avoid issues from occurring or debug problems if they occur. This, however, can be a laborious and time-consuming task and that usually only experienced analysts and engineers can perform.

The project will deepen research into AI and deep learning methods, while providing a practical setting that can showcase the importance of AI

In the project, we are addressing this task by leveraging several deep learning graph-based approaches, to detect anomalies within the Parallels Remote Application Server (RAS) user experience and provide IT teams with tools to help businesses deliver a better service in today’s always-connected world. In addition, the project will deepen research into AI and deep learning methods, while providing a practical setting that can showcase the importance of AI. New improvements will feature proactive problem detection and classification with advanced log analytics and monitoring of data generated by Parallels RAS and its infrastructure.

Graph data consists of nodes and edges which can be modelled to describe a variety of systems depending on the application. Graph Neural Network (GNN) use this graph data to learn patterns and behaviours. A server system and its log can be modelled as a graph, and this structure in turn can be used to learn patterns found in user and server interaction, and detect suspicious behaviour and anomalies within the system. One of the biggest advantages of using GNNs is that they have a better memory footprint than regular deep learning models since they only need to store information about connections between nodes instead of all nodes in the graph.

However, there are also other aspects of GNNs which require further research that the project will address. GNNs have a black box problem like deep learning models, meaning it is difficult to understand how a GNN comes to its conclusion because the complex algorithms’s internal processes are difficult to retrace from the outside. This project aims to broaden the understanding of graph-based methods over more conventional methods used. Further research in this area will be beneficial to the ICT sphere but also to other aspects of academia regarding predicting and detecting anomalies.

For further information on programmes at the Department of Artificial Intelligence, visit here or e-mail ai.ict@um.edu.mt for research collaborations.

Sound Bites

•        Two common viruses that lie dormant in neurons — herpes simplex virus (HSV), and varicella zoster virus (VZV) may be triggering the onset of Alzheimer’s disease. Lab models of the human brain show that activation or re-infection of VZV can trigger neuroinflammation and wake up HSV, leading to accumulation of Alzheimer’s-linked proteins and neural decline. Alzheimer’s disease can begin almost imperceptibly, often masquerading in the early months or years as forgetfulness that is common in older age. What causes the disease remains largely a mystery.

•        Researchers recently announced that they have figured out how to engineer a biofilm that harvests the energy in evaporation and converts it to electricity by using the moisture from your skin. This biofilm has the potential to revolutionise the world of wearable electronics, powering everything from personal medical sensors to personal electronics.

For more soundbites, listen to Radio Mocha www.fb.com/RadioMochaMalta/.

DID YOU KNOW?

•        Rainbows can exist at night when the light from the moon disperses through water droplets. Because moonlight is fainter, the bow itself may appear white, but long exposure cameras can capture the full array of colours.

•        Aristotle was one of the first thinkers to seriously consider how and why rainbows formed.

•        He devoted much of Meteorology, Book III to the topic. Although some of his theories were wrong (he posited that rainbows were made of only three colours), he rightly deduced that white light separated into the colours of the spectrum to form rainbows.

For more trivia, see: www.um.edu.mt/think.