AMA: About Me Anywho/Ask me Anything

Ask me Anything: A non-celebrity answers oft-asked questions .

While my previous articles and posts have focused on technical aspects of programming, machine learning/AI or research , this is a different beast. Confront this at your own peril! This self-inflicted AMA was influenced by questions candidates asked me after their interview, colleagues I worked with and also some college students. An indulgence if you may....

Today's Question

How did you transition from life sciences to a very different domain?

I have forever been interested in cross-disciplinary approaches that merge computation with analysis and a given domain. It so happened that in the 1990s, cross-disciplinary academic courses or applications of the same weren't prevalent in India. Heck, data science wasn't coined as term. Most practitioners were adherents of classical statistics and scientific computation, was done in physics departments using Fortran while mathematics/maple/matlab were used for Engg.

I had been teaching graduates basic programming in Qbasic and analysis using Excel (+add-ons)and a curve-fitting software called Curvefit . While not a great statistician, I also did some consulting stints to analyze agricultural data on the side. Ecology was big on using statistics and methods like PCA to analyze data and there were some fab books with diskettes and programs, to perform multivariate analysis for example. I remember writing a DOS based program for an Indian Pharma (Who shall not be named) in C and Qbasic for analyzing viral/bacterial assay data around 1996-97. All this gave me a foothold in LifeSciences. I still remember needing to calculate Eigenvalues and Eigenvectors to do a PCA without any libraries! Shudder, shudder!

This beachhead in Life sciences not only allowed me to leverage my expertise but also solved one vexed issue faced by most such efforts. There was enough open data, both structured and unstructured, to experiment with. This was particularly the case after completing my course in Bioinformatics , where we came to know about sequence and structure databases, text data (Pubmed) ,gene expression (Quantitative levels)and what not. This was much earlier than the kaggle era and not many domains could boast about such willingness to apply algorithms (BPNN's, Cart and SVM were in vogue)and also have relevant if non-standardized data.

It's my fundamental belief that unless a domain is highly esoteric and mathematical e.g. relativity and most areas of mathematics, key concepts can be absorbed relatively easily. It's important to understand that regardless of the domain, computational methods, algorithms , statistics , data crunching and advanced ML/AI and visualization are a universal tool-chain. Every domain has issues with data quality, ontology, profiling, business rules, feature engineering, classes/labels, quantities to be predicted, time as an important dimension etc. So if you understand how to use the right combination of these techniques and map it to particulars segments of the business problem to arrive at a final solution, half the battle is already won.

Most domains e.g. telecom, retail, banking aren't fundamentally hard. Its the terminology and the business rules that are initially hard to fathom and its important not to get intimidated. More importantly, even a domain expert, does not know everything & collaboration is key. Self learning the domain, general knowledge and asking questions that help problem solving using your set of computational techniques, makes a whole lot of difference. That's what I would heartily recommend anyone trying to take a similar leap. What one should refrain from, is boasting about success in analysis in one's domain but being utterly uninformed about the new domain one wishes to enter. Not implying that all this is trivial but it's certainly not as hard as sometimes made out to be.

This is not to say that only having the right attitude or self learning is the answer to successfully changing from one vertical to another. Some luck matters too! A big issue are 'keyword' matching mentality that filters out talent at an early stage and managers/companies who fixate on X yrs of experience- in applying Y technique to- Z domain, rather than intrinsic problem solving abilities..

Most often profiles like mine with deep experience in Ml AI and even coding used to get lost in the sea of keywords. I was rather fortunate that VODAFONE, GBIS VOIS leadership recognized the ability rather than the "Life Sciences" label that could have proved to be a millstone around my neck. So that ends a rather long winded answer to today's question!

Statutory Health Warning: I have a few more lined up to last till 31st of December 2021.