Does one need a Formal Degree in Machine Learning ?

This is a followup of my earlier post on whether it makes sense to do a part-time PhD: where I got asked if it makes sense to pursue a formal degree in Machine Learning at all... I see people with an academic background, solving complex problems using ML talk about how folks without formal training call themselves data scientists. At the same time, I see many good self made data scientists who do not really see any need for a formal degree in ML in this age and time when information is so freely available. Having been in both circles, here are some thoughts...

I would look at the evolution of Machine Learning as a field similar to the way computer science has evolved. In the beginning, there was no dedicated computer science discipline. Yet programming slowly became mainstream and people who picked up programming skills ruled. Slowly CS became a dedicated disciple in schools and folks started learning about the theory of computing, discrete mathematics and algorithms in great detail if they wanted to get into software development. However, after the advent of high-level programming languages, there has been a huge demand for developers. The field has been all encompassing and folks from all undergraduate disciplines still try getting a job as a developer even when they do not study the core CS topics, and they are largely successful.... 

Does this mean there is no place for dedicated formal education in CS ?

Of course there is. One might not need dedicated formal education to be able to write a basic web application. But could Dennis Ritchie have written such a beautiful and monumentally impactful programming language (The C language) if he did not have formal education and the mathematical exposure - probably not. 

Bottomline : There are some tasks for which you do not need rigorous formal training - Let’s be honest, these are most of the tasks ! Nevertheless, there are few where such rigorous training helps…

Coming back to Machine Learning, let us now answer :

Can I be a good data scientist through self-learning and online courses ? 

Absolutely. The key skills involve understanding the various techniques available and being able to formulate real world problems into those, so that statistical models can be applied as needed with the right features and the data for solving them

Machine Learning as a field has evolved to a stage where there are ready libraries and readymade tools leveraging a range of techniques - right from traditional decision tree and SVM based models to deep learning techniques (recently probabilistic programming is also picking up). These are sufficient to do most of the common tasks 

Let us recap what one needs to be a good data scientist.

• You need to have good coding skills in python or R. This should not be hard if you are already a developer. Else there are many online platforms to learn programming.
• You need to have good data intuition, a flair for data driven analysis of various hypotheses. This comes with practice, and doing a lot of projects, solving a lot of problems.. Very often, when you are relatively new, working with someone more experienced in the field helps develop these skills faster. In a lot of situations, coming up with ‘a’ solution is a good start. Often, the next step is to come up with more automated, data-driven scalable solutions.
• You need to have a basic understanding of some math concepts like linear algebra, calculus and probability and statistics to have a handle on various modeling techniques available in the ML toolkit and be able to apply them well.

Unlike 10 to 15 years back where self learning was much harder, there are tons of great resources onlilne today if you have been looking to pick up these skills. Absolutely possible to learn and practice on your own. You can have a successful career and get a good pay check if you are good at it. (Note, this is a good time to brush up your skills and learn new things if you are looking to transition - look at my previous article on Corona Virus and Data Science Jobs).

When should I go for a formal degree ? 

• When you want to dig deeper into the subject and learn more rigorously - most of the good grad schools facilitate this. Rigour and exposure to various techniques helps one to dive deep into the problem at hand and think of possibilities beyond the most obvious first cut solution.
• When you want to create new ML algorithms, or solve harder problems for which there are no ready-made solution templates. Trying new approaches is something you learn well in an academic research environment in most of the good grad schools.
• If you want to write efficient tools and libraries implementing ML algorithms and you need to have a deeper understanding than that required to apply them.
• If you are relatively new in your career and a college degree could add brand value or lead to campus placements
• When you think it is hard to pick up ML alongside your regular work, a full-time degree might be a clear path to career change

To summarize, both online learning/self-learning and a formal degree have their own merits. While a formal degree in ML or data science is clearly helpful in certain situations, there are numerous instances of self made data scientists who are great with ML modeling and problem solving and have built super cool stuff….

So, What do YOU think ?