The ubiquitous power of trivial knowledge

When I was in my second year of BE in Nitte University, I had an unusual opportunity to work with a research group consisting of my seniors working on a Machine Learning problem for a biomedical application. I was the only junior member in the team with zero experience in scientific computing. I did not have the functional knowledge of the theoretical underpinnings of machine learning or data science. Yet, for some reason, I was specifically asked for by the then head of the computer science department, Dr.Nileema Reddy.

My job was simple, I was to learn and teach the mathematics behind algebraic topology to the remaining members of the research group. Several presentations were done. I learnt algebraic topology well and delivered my understanding of the topic to the best of my ability to the team. As I recall now, they were some of the best months of my life in Nitte where I really learnt something.

Although, owing to the rest of the members in the team being in their final year (And only some of them interested to continue the research), I guess the project was left midway. I still do not know what happened with it. It was unfortunate for me specially since I found no gratification in terms of publishing out of whatever little efforts I had put in.

Today, I’m working on my master’s thesis at IIT-Delhi. My research is in Celiac Disease identification using Deep Learning. Recently, I was posed with a problem in my work to develop a differentiable loss function that would incorporate the clinically relevant information in the learning path of the DNN. I was stuck for months with this problem and just a couple of weeks ago, I revisited algebraic topology and its application in representation learning.

Couple of weeks later, I have succeeded in designing my experiments based on the principles of algebraic topology and I’m waiting for their results from my super-computing node.

Looking back, I don’t think even for a second that all the knowledge I gained on algebraic topology as a novice engineer has gone to waste. If anything, it is coming back to help me in ways I never imagined. That’s the end of this story.

Here is the barcode representation of the ground truth of a particular group of cells called the epithelial cells. My loss function incorporates some attributes of the statistic of persistent homology.

The Barcode representation of the ground truth image for the analysis of persistent homology.

Leave a Reply

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