Question: What is the most interesting work-related thing you have learnt/done?

Sreejita Ghosh answered on 2 Jul 2020:

I will answer this in 4 parts because there are 4 main aspects of what I learned/done and found interesting: technical, application/ interpretation, communication, philosphical
Wrt technical learning, I learned to enjoy Maths now that I can see the applications. This in turn motivates me to learn new theories in Maths so that I can improve my machine learning (ML) models in terms of performance and interpretability. During school, to be honest, my motivation for Maths were to get higher grade in it, and so that I could secure a good rank for an engineering exam. During my research (which is highly dependent on calculus, linear algebra, probability and statistics, and statistical physics) I learned to really apply math to every problem and enjoy the thought process. It opened up a new way of looking at Maths.
Wrt the application of my ML model, well, it enabled knowledge extraction from medical datasets along with identifying the correct medical conditions it was trained on. In other words, the model enables doctors to learn things about the medical conditions which were not known before. This is possible because a ML algorithm can compute and find ‘hidden’ relations between 100s of features from 1000s of subjects, which is difficult for a human mind to ‘see’. Anyway, because of this, the explainable ML models my team and I developed found such never-before-noticed relations between hormones from adrenal glands (wrt the hormone disorder project), and between different clinical features related to different stages of asthma and recovery from it (wrt the asthma project).
Wrt communication. I learned over the course of research that to be a good data scientist it is equally important that there is clear communication between the dataset provider/ collaborators and the data scientist team. I learned how to talk to experts from different domains when asking them what they want, explaining to them what is technically feasible and what is not, and what our findings mean in their domain. For example, 1) the term optimisation is interpreted differently my computer scientists and mathematicians; 2) ‘Decision tree’ is a ML model used for classification and regression, but in the medical domain they often mean their diagnostic knowledge based logical questions when they use this term. So while talking to our collaborators from different field I have to use terms carefully so as not to confuse them or myself. It seems a small thing but it is interesting.
Wrt Philosophical aspects. When you bring in the humour filter to a difficult situation, whether in life, or explaining a complicated idea/ theory, they become easier to deal with/ understand.