Dr Marwa Mahmoud, Department of Computer Science and Technology
Machine Learning Models
Machine learning models are very important tools for predicting and managing the COVID-19 pandemic. With all the global efforts to make the data available, many challenges arise around the consistency and reliability of the available data. Models are usually as good as the data. When we are dealing with data collected on a global scale and in a time of global distress, we need to be careful with how the data was recorded, how our ground truth was collected, how to incorporate uncertainty and any confounding factors in the models we de-velop, and how the terms and criteria are defined across different countries. It is also important to think about how these models are disseminated to the laymen to avoid any un-intentional misinformation while bearing in mind the societal effects of these interpretations.
I believe that there will be lots of opportunities for accelerating technology development to cope with the demand and the trust that people put on technology to run their daily lives. This is anything from video calls with your boss, catching up virtually with friends and family, to even having a virtual counselling session. As a researcher in affective computing and social signal processing, I believe there is great potential and need for these technologies that make our interactions with the machines more "human". For instance, our video conference systems can detect and interpret our non-verbal cues and convey them along with our video and audio to foster the virtual interaction especially if the resolution, or camera angle is not ideal. However, more now than ever, technology developers will need to be transparent about their data, models and output in order to build trust in these technologies amongst users.
