You don't need the fancy.
With new ground-breaking research in the AI/ML domain, how relevant are the basic algorithms and methodologies?
The Modern Research
Even if you've never tried to learn machine learning, you must have encountered somewhere on the internet the cool new applications that are being built by machine learning.
A very famous example is GAN. The deepfakes are one of their applications and are quite famous over social & news media.
And not just GANs, if you go browse research sites on machine learning you'll find all these new more sophisticated methods people are coming up with.
Honestly, it's all thrilling and fun but for someone who is just starting out this could all be quite overwhelming; watching new stuff being developed to catch up on while you are growing from scratch and dealing with the basics.
You may or may not resonate with this feeling but I totally did and still do, although less so now.
Relevance of the Traditional
Now when I say it is overwhelming what are the impacts of it?
When I was still going through the Andrew Ng course, I was watching the domain grow at a tremendous rate, with more and more applications incorporating artificial intelligence. I was rushing through my learning process, not giving enough time and thought to the topics that require utmost attention because at the time it felt like I was lagging more than I was catching up each day and I wanted to do more but quickly.
But that feeling ended with college, ever since I have joined my organization, I came to understand that it is not an absolute necessity to always use state-of-the-art techniques, but it's important what you do you do it well, and have an in-depth understanding of.
The case is, most of the problems existing in real-world projects still match the criteria to apply the commonly known algorithms or processes.
I still see clients, at least in my country whose use cases are one of classification and regression problems, they don't require a tonne of knowledge gathered from studying research papers that came out last month.
All they require could just be a traditional algorithm or analysis but they require it to be done properly with no-nonsense.
If you are working for a customer-oriented organization the chances that you'll get to work on something cutting edge are relatively less.
You have learned GAN? Cool, wonderful. Now the client requires some regression analysis, can you give some insights?
I learned that even Linear Regression or subsequent algorithms with similar principles is one that very few have understood as well as it should be done to apply to a real-life use case.
When you apply regression to a problem, you don't just do some imports, scaling, fit a model and be on your way.
You check whether it has applicability, you check for assumptions, you thoroughly perform multiple types of analysis and hypothesis testing of your "insights" and then it becomes a complete solution.
The same goes for many other algorithms or techniques that people who are starting seem to care less about.
Unlike personal projects, professional projects require some time-spending, I am working on one regression problem currently, and it's been going on for the past 8 months, seems a lot right?
What have we been doing all this time in a regression problem?
Making sure that we leave no stone unturned, breaking down steps into smaller parts, understanding each and every aspect, looking for the caveats, validating our doubts, verifying our assumptions, and more.
This is the way!
Not a Discouragement
The point of this is not to discourage people from learning new stuff, not at all. It is one of the boons of machine learning that we are seeing every day the power of it unfolding before our eyes and have the chance to work on the latest applications. On occasion, I myself have been required to look for some latest techniques to solve some of the problems.
The argument I am making is to make sure that you don't overlook the non-fancy, do not skim through the basics, spend some time on the important stuff. You see them now because they are still very much relevant.
Thanks for your time and for bearing with me, I hope it was worth it!
Rate if you liked,
Subscribe if you haven’t,
Share if it seems helpful.
Until next time!