Careers
I have interviewed many data scientists who are at the various stages of their careers and one question I ask all of them is how do they keep themselves abreast with the fast-paced changing world of AI and deep learning.
This question serves two purposes:
- It tells me how aware they are of the world of analytics and provides nodal points for further conversation.
- It is a window to their world through which I can get a glimpse of their problem-solving skills.
We are living in a world of fancy titles which range from data scientist to data evangelist and from data engineer to data ecologist(I hadn't heard of this one till a few days ago). Thanks to Harvard's article and readily available powerful computing machines, the field has seen a monotonic upsurge in anything and everything that is related to the domain. The swell has upskilled many people and has created many data analytics-related helpful microcosmic communities. With precipitous amplifications come to the problems and we do have a major one here.
The field is relatively new so a lot of folks create their own definition of the roles and responsibilities of a DS(Data Scientist) and treat them like a gospel; the definitions can range from a full stack developer to some data ninja sitting in a basement hacking the night away. Whatever definition you have concocted, stick with it because one time or the other your definition and you will be tested and the litmus test will let you know your results then and there itself.
The D-day generally poses you with a problem that you have never seen before and Stackoverflow, GitHub, or other websites won't come to your rescue. The only thing that will rescue you are your intrinsic skills that include the ability to breathe normally, programming, and creative problem-solving. There is one more in the list which isn't talked about much and aligns coherently with the term scientist in DS.
The skill is multifaceted and is the ability to:
1. Parse through esoteric and terse scientific literature quickly.
2. Understand and draft an algorithm on the central idea.
3. Program it on a test data of your choice and check for consistency.
All those mentors, blogs, video channels help you with the basic concepts that can get you started but scarcely anyone tells you how to build upon your knowledge and what to do when you are posed with an unseen challenge.
A month ago or so while working on audio data from oil rig sites, we were heading nowhere near the solution and that's when we decided to recalibrate the initial strategy. We read a paper on Transfer learning for classification, I used my Spotify's playlist data as seed, and we wrote quick and dirty code to reinforce our hypothesis that rather than working on raw audio, waveforms, or only fourier transformed data, we should focus on spectrograms.
I wouldn't spread a false narrative that the world was rosier and full of unicorns after we found the approach. We kept bumping into one issue or the other but it was all in the right direction. The decision to step back and changing the strategy was pivotal point for a robust model was a product of the decision to read and implement the paper in first place.
The difference between a good and average data scientist doesn't lie in the googling skills but the sheer ability to find the relevant research paper or literature and see whether it is the solution to your problem. It equips you with more analytical horsepower, hashes out the strategies that can work, weeds out the unwanted directions where you or your team can spill the efforts in.
So, adorn yourself with whatever titles, embellish your CV with keywords such as Machine Personality Developer, Data Ninja, or squirrel(an inside joke with a few of my friends :)) but fear of maintaining the status quo or past achievements shouldn't hold you back from the truth and from experimenting, because that's what data science is about.
You can mull over it and hold philosophical discussions on whether this paper fits the bill but you will find the answer only when:
- You will get your cup of coffee,
- Plug your earphones in (headphones if that's your preference),
- Straighten your back,
- And start translating that paper into a code.