Meet Sri Lankan Researcher — Madhushi Bandara

What are you currently working on or worked on before?

I spent my Ph.D. candidature exploring how to engineer smart data analytics platforms, incorporating semantic web principles, knowledge-graphs, and linked open data. We identify several challenges faced by organizations today when implementing data analytics systems at the industrial scale.

Firstly, although there are numerous software tools and libraries to assist data analysts and software engineers in developing machine learning algorithms, organizations are struggling to deploy them as industrial-scale systems. Industries need systems that are long-lasting and match well with their organization’s operation domain, IT infrastructure, and analytics goals. Secondly, off-the-shelf data analytics products are challenging to customize and require high technical expertise to operate. In many cases, organizations outsource external teams or contractors to design and develop ML solutions for them. The resulting solutions are hard to maintain, especially when modifications are needed in the light of more domain knowledge or to stay up-to-date with the latest ML innovations.

Furthermore, the experience and knowledge accumulated by the external teams will not be available for future use by the organization. Thirdly, even a slight shift of the organizational analytics requirements, such as adding a new algorithm or integrating a new data source, requires full re-engineering of the existing ML systems and processes: “Change Anything — Changes Everything”.
We can understand this behavior through the lens of technical debt, the implied cost of additional rework caused by choosing an easy solution now, instead of using a better approach that would take longer.

Recent literature surveys that study existing analytics platforms as well as the experience of data scientists in large organizations such as Google and Microsoft repeatedly highlight the rising technical debt of ML systems and raise the need for investigating innovative approaches in designing and developing ML systems to reduce associated technical debt.

My current research goal is to reduce technical debt by incorporating provenance, traceability, and explainability into data analytics systems through knowledge-graphs and semantic web technologies. Semantic web technologies provide rich information models that can be used to create knowledge-graphs that represent and integrate knowledge related to different aspects for AI systems including domain knowledge, constraints, algorithm meta-data, performance attributes, and information about data sets. Such knowledge-graphs can capture how the AI system operated and evolves over time, enabling organizations to understand and maintain the ML system.
The findings of my research can be used in industry to develop new AI systems that are easy to maintain over a long period of time, saving money and resources.

Maintainability also means that the organizations can frequently innovate their ML systems, with new trends in ML or with changing business requirements, without stuck in legacy systems or being obsolete. By providing provenance and traceability within ML systems, it will be easy to identify errors and fix them quickly. Providing explainability increases the trust of the ML system, particularly in industries operating with critical systems, resulting in increased acceptance of ML systems and the insights they provide.

What encouraged you to pursue your research topic?

Data analytics is a profession of high demand today. There is going to be a shortage of experts who are capable of navigating the complexity of both software engineering and data science aspects when conducting data analytics. Coming from a software engineering background, I experienced how neglected data science is, as a branch of software engineering. So I want to bring best practices in software engineering, into data analytics system development, so that industry can develop sustainable and high-quality data analytics platforms.

I believe my research can contribute to change the way we think about data analytics systems: from a black-box produced behind closed doors to a more transparent white-box designed through multidisciplinary collaborations, where decision-makers can understand and justify the rationale behind the decisions made. We will pave the pathway to embed domain knowledge into analytics systems and make them understandable and usable for non-IT experts as well.

What is the name of your current institute?

University of New South Wales.

Where do you find your best inspiration for your work?

Seeing our research outcomes applied in industry and when other researchers use our work to create new research outputs, making a ripple effect.

Can you share with us some of your publications?

You can get my publications here.

What’s one of your biggest personal achievements so far?

Finishing my Ph.D. and building a fantastic network of research collaborators on the way.

What lessons would you share with a budding researcher?

Research needs patience, and especially in computer science, high-quality research is built with cross-discipline collaborations, working together as a team. Always believe in yourself and look at the big picture, how the smallest contribution can make the world a little better every day.

What motivated you to be a researcher?

In the long term, I see myself involved in knowledge engineering and data science research that add value to the organizational decision-making process and improve the usability of data engineering platforms while simplifying the platform development process. It will contribute to reducing the cognitive burden associated with developing and operating data analytics platforms in modern organizations.

If there is a chance, will you help build research in Sri Lanka?

Definitely.

According to your opinion, what are the changes that the Sri Lankan education system needs to do, in order to meet the requirement of the international industry and academia?

Sri Lanka education system is already capable of meeting international standards, with a lot of educators, mentors, and an industry that is willing to support students. We need to build on these strengths, expand the research capabilities, and look into new research funding opportunities.