Applying Ai and machine learning techniques to analyse huge amounts of data, e.g. within medical diagnostics, is not new. However, in a 5-year research project funded by a DKK 10 million RECRUIT grant from the Novo Nordisk Foundation, Associate Professor at the Department of Computer Science at Aalborg University Arijit Khan and his team will apply machine learning to new types of data in order to generate better results when analysing the data.

- Our research is about knowledge management and how to get as much relevant and useful information out of the available data. It is foundational research, which may be applied to a number of different domains like finance, social network data, etc., but in this project, we will be looking particularly at applications within medical diagnostics and biology, where the potential benefits of applying machine learning technique are substantial, Arijit Khan explains and continues:

- There has been a lot of interest in and a lot of work done to develop what you may call medical assistant systems. These systems have typically relied on text data, e.g. from various types of medical documents, but we want to explore if applying machine learning techniques to so-called graph data can provide new insights and better results.

EXTRA INFORMATION IN GRAPH DATA

- Since graph data contains connections to other pieces of data for example through a link in a text, you can obtain extra information that will potentially provide more accurate results, e.g. within medical diagnostics. The technique that we will be applying is so-called graph embedding or graph representation learning, Arijit Khan explains.

A more dialogue-based approach will also be explored in the project.

- Sometimes, a medical assistant system may not be able to provide useful answers straight away, but through a number of iterations where information is exchanged between the user and the system over time, just like in a conversation between two people, a machine learning-based framework could be used to provide a better understanding of the context and more accurate information, says Arijit Khan.

EXPLAINABLE AI

In the project, Arijit Khan and his team will also be addressing the issue of explainable AI.

- Machine learning is often used as a black box, but since these tools are used in many critical applications, e.g. within healthcare, you have to be very accurate. So, it is not enough to just tell the doctor of the biologist that this is the result, but you should also be able to point to the reasons why. Without providing reasons for the results, it very difficult to create trust in these systems, says Arijit Khan.

PROMISING AND CHALLENGING

The perspectives of applying machine learning techniques to graph data are thus very promising. According to Arijit Khan, there are, however, also a number challenges that have to be met, before the full potential can be realised:

- Technically, graph data is more complex than text data. You have more information and more complexity, so you need more advanced algorithms that are both more efficient and scalable. And then there is always the challenge of adopting these kinds of techniques into a domain. It will be a new technology, so you need to convince domain scientists that it is both usable and useful.

 

FURTHER INFORMATION

• Meet Arijit Khan – Recipient of the Novo Nordisk Foundation Recruit Grant and New Associate Professor
• RECRUIT – Grants for international recruitment 2022
   

CONTACT

Arijit Khan
Associate Professor
Department of Computer Science
Aalborg University
Mail: arijitk@cs.aau.dk

 

Stig Andersen
Communications Officer
Department of Computer Science
Aalborg University
Mail: stan@cs.aau.dk
Phone: +45 4019 7682,