My project is about how we can explain the predictions made by machine learning algorithms.

For example, if we have an algorithm that tells us whether an image contains a polar bear, it’s important to know whether it detects the polar bear itself or just snow (which is likely to be included in most images of polar bears).

My research is centered around creating my own implementation of a technique that, given any machine learning algorithm, can provide an explanation about how the input would need to be changed to completely alter the prediction.

This has many uses, such as in a loan approval system, where someone who is declined a loan could request an explanation as to how they should change their application to get approved – it could suggest that they wait a few more years or that they build a credit history.

During my time as a JRA I worked around others from the Predictive Analytics Lab and the Text Analytics Group and have been able to attend their symposiums and reading groups. In fact on my final day I presented on the key paper from my research at the lab’s reading group!

These opportunities have given me openings to topics that I would’ve never known of from lectures alone.

I heard about the scheme from previous JRAs and was interested in doing one in machine learning. Part way through spring term I spoke to my lecturer, Dr. Novi Quadrianto, and asked about the possibility of doing a JRA with him as my supervisor.

After hearing about the different areas of his work, we decided on my project title and I drafted up a project proposal which was roughly 1,000 words. After that it was just the case of getting a reference from a previous lecturer and then I was able to submit my application.

I’ve found my experience as a JRA to be really enriching! I’ve had the chance to learn about interesting developments in machine learning.

Alongside this the events put on by the Undergraduate Research Office allowed me to meet JRAs from other subject areas. It was great to find out about what other projects people were working on and to make new friends.

Categories: News

Leave a Reply

Your email address will not be published. Required fields are marked *