But where is the science in that? Machine Learning is not science.
-By old-school profs
“If you are not smart enough to deeply understand a phenomenon, then yes you can use ML models. But where is the science in that? Machine Learning is not science.”
In 2018, one of the professors at MIT told this to his PhD student who was my good friend. My friend wanted to incorporate machine learning into his research, but the professor was very old-school, like many professors were at that time and before. (I am curious to know what these professors are thinking since the awarding of Nobel prize in Physics and Chemistry to ML).
During COVID however, I saw an interesting shift happen, not just at MIT but across many universities. More and more pure experimentalists from mechanical engineering, material science, physics, chemical engineering etc. were incorporating machine learning into their research.
There could have been a few reasons for this shift.
It was hard to do physical experiments during lockdown and maybe this forced researchers to look at alternative options and ML was there.
Scientific Machine Learning (SciML) was gaining popularity. Everyone started finding intersections between ML and their field. For example at MIT, one lab started studying the fluid mechanics with the help of ML. Another lab in mechanical engineering department developed better solar cells using ML and another lab built a better nano-sensors using ML.
FOMO (Fear Of Missing Out) because everyone was getting into ML. Yes professors can also get FOMO.
I don’t know the reason why I saw traditional experimental labs incorporating ML into their research. But whatever it was, it was a good thing. Because science or not machine learning is powerful. It can provide results. No-one can deny that.
At around the same time (end of 2019), I was a hard-core experimentalist with a strong computational background. My time at IIT Madras made me very good at modeling and simulations and my time at MIT made me a strong experimentalist.
And just like many professors and students, I was also fascinated by ML. Initially it was FOMO when my peers at IITM ventured into CNN and RL, while I was still working on tolerance of precision grinding machines. There was a time when I felt like I was living in 1950s whereas my peers are living the modern life.
But my FOMO disappeared at MIT. Because my lab had spun-off big companies that had nothing to do with ML when they started. I was being convinced that I can add value to the world without knowing ML.
But yet I was fascinated by ML and I did not want to miss out the opportunity to learn ML from one of the best places on earth - MIT.
So I did 4 things.
I decided to enroll directly into a graduate-level ML course at MIT. I was sure to get my ass kicked by smart undergrads.
I started to think myself as a serious ML engineer, although I was nowhere close. This was my trick to convince my mind to take-up hard ML problems without feeling intimidated.
I decided to stay away from toy projects (especially those on Kaggle) that were a waste of my time. I decided that I will collaborate with people on research-level problems in ML. I was not looking for resume points. I was looking for depth.
Since SciML was gaining a lot of word of mouth, I decided to dip my feet into it.
Venturing into ML has been the best decision I have ever taken in my life. 5 years later, now I running an AI-first company with my co-founders. If I look at the top 4 things I do on a daily basis, these are those.
Learn AI/ML
Work on novel research (and publish papers eventually)
Build AI products and features for Vizuara’s customers
Teach AI/ML
When I look back, one thing that helped me was my strong self belief. I strongly believed and still do (even more so now) that anyone, I mean anyone can transition to ML if they want to. It doesn’t matter your department, CGPA, age, job status etc. It does’t matter if you currently do not know linear algebra and calculus. You just need to commit time, ignore noise and follow your path.
If you are someone looking to transition to ML but do not know exactly where to start, you can join this free webinar happening today (Feb 12th, 7pm IST) where I will be sharing my view of what should and what shouldn’t someone be doing if they want to become a serious ML engineer/scientist. You can register for free here to get notified: https://topmate.io/sreedathpanat/1432422