Spotlight Story: Privacy-Preserving Machine LearningKiran Parmar, Wednesday 07 July 2021 | Annotate
As part of our summer series of interviews with KMi people, this week’s spotlight falls on Audrey Ekuban, who tells us about her PhD research on privacy in Educational Analytics.
What can you tell us about your PhD research?
I am exploring the solution space of Privacy-Preserving Machine Learning with the intention of creating a usable framework for Educational Analytics.
What is Privacy-Preserving Machine Learning?
Privacy-Preserving Machine Learning is an emerging suite of techniques that enables a machine learning model to be trained on raw data that it does not have direct access to. In reality, a combination of these techniques could be used, with each technique having some limitations. The trade-offs of each technique need to be understood so that the best fit for a specific project being undertaken can be communicated.
Why is this important in the field of Educational Analytics?
There is general agreement that Learning Analytics, for instance, if used correctly, can be beneficial to both students and academic institutions. However, as Learning Analytics leaves the controlled area of research and enters the mainstream and commercial space, various sociotechnical concerns have been raised. I believe that my work can alleviate some of those concerns.
What are the main concerns of Learning Analytics that you are hoping to address?
Learning Analytics requires personalised learning activity data. This presents an ethical dilemma, especially as the learner does not have the ability to switch the collection off. This means that a learner cannot opt into or opt out of the data collection. In addition, the learner does not have access to the data that is collected.
There are also concerns surrounding who has access to the data, who owns the data, where the data is stored and how long the data should be stored for.
What can you see this research area leading?
Currently, there is a large open-source community involved in both research and development of Privacy-Preserving Machine Learning solutions. In addition, anonymisation of data does not give the protection that it used to provide. These two factors lead me to believe that we are at the start of something that has a tremendous capacity for growth.
What is the end goal of your research?
The end goal of my research is to realise the value in learning activity data for the benefits of both educators and students in an ethical manner. With Privacy-Preserving Machine Learning, the benefits from Educational Analytics can be achieved without the concerns of data privacy.
Do you think that potential benefits to a student can outweigh the right of students to control their own data?
More and more organisations in more and more countries are entering the Learning Analytics space. There is the potential that students will be harmed by organisations using learning activity data for purposes other than Learning Analytics. Students having the right to share what they wish to share, whom they wish to share it with and when they wish to share it, I believe, will minimise that harm.
Would this control introduce algorithmic biases and impact the students who need it most?
Educational Analytics is also of benefit to the institution. In fairness to students, there should be incentives for them to take part in the training of machine learning models.