Within machine learning and all the fields related to it (data science, computer vision, NLP and etc) people usually come across 2 types of problems:
- Problems where ready-made algorithms work
- Problems where ready-made algorithms don't work
Most companies that work in the ML domain work on problems of the second type. If this were not true then companies providing ML based services would not exist in the first place because then any developer could simply run a couple of functions (pfft... how difficult is it to copy a couple of functions? Just use Stack Overflow!) and get the required work done. So if you wanted to work at such a company, you would be required to implement algorithms on your own. At the bare minimum that requires a knowledge of statistics, linear algebra and calculus. If you have no idea about statistics chances are you won't even understand what's happening (not to mention actually implementing the actual formula) and if you have no knowledge of linear algebra chances are you'll write some code that will take days to execute as opposed to hours (linear algebra libraries allow you to write extremely efficient code that has the capacity to maximise the utiliztion of your processor along with any GPUs)
"No worries! I'll just work on the first type of problems - where I can possibly use ready-made solutions!"
You've been presented with a problem that can be solved via ready-made algorithms. But hold on, should you use linear regression or neural nets? But there are so many type of neural nets, which one should you use? Technically speaking anomaly detection does the same thing as linear regression, so which one should you use? If you do choose anomaly detection with gaussian distribution then should you use the single-variate version or the multi-variate version? Is there any difference between the two? Oh, single variate assumes independence between variables. Hold on... what's independence? I also have all this data, how do I characterise it to fit the best possible machine learning model? You do that with distributions? What on earth is a distrubution?!
All these questions are answered by having mathematical intuition of the algorithm. Guess how you get to understand that mathematical intuition!?
"Okay, thanks for killing my optimism off. I'll just take a Coursera course I guess"
Stop! I guess you now understand that it's important for you to be able to apply what you learn or you're basically wasting your time! Most math courses on sites such as Coursera, Edx and etc. oversimplify the content and do not deliver the concepts required to apply your learning to real life situations. You're making the exact same mistake most Pakistani educational systems do by not learning to apply. Would you not want to know what the symbols mean in these mathematical courses? How to manipulate them correctly? Or what the intuition behind those symbols is? Hence we at Pakistan.ai started a guide to combat this issue by adding courses that are rigorous enough to teach you the concepts right. Find out more about these courses here.
"Okay, I understand what you mean. How about I drop everything and go do a PhD in math now -_-"
Machine learning does not require a PhD in mathematics, neither a masters and nor even a bachlors. But what it does require is that you understand the basic fundamentals of Probability and Statistics, Linear Algebra, and Calculus. Calculus is taught rather rigorously in Pakistan across all boards so this tends to be less of an issue. However Linear Algebra, and Probability and Statistics is not.
Furthermore, there's another plus point to doing rigorous mathematical courses. Data Science and Machine learning are rather new fields (even abroad) that most employers think twice before hiring someone (especially in Pakistan) - it's difficult to tell whether a new employee within this domain can actually deliver value. However, individuals who are sharp and bright have a greater chance of getting over this barrier. Reason: employers know that such candidates will be capable of picking up things quickly and they will be able to make the correct critical decisions as to what needs to be done. Question is how does one become like that?
One reaches that potential by challenging himself/herself again and again and again. You must go through the fire and then only you will come out as shimmering gold. Those difficult maths courses are the challenges you can take to show your client, or your employer or your investor that you are capable of taking high-tier courses, that you have the intellectual capability of overcoming intellectual challenges. Some might say that such gifts (picking things up quickly, having extraordinary critical thinking skills) are genetic. They are not. All modern psychological research points in this direction.
"I give up. You've convinced me."
I'm glad I have.