My 4th Attempt at AI/ML

So this was my 4th attempt at it. I come from a development background, so it was more of a learning by doing/coding thing.

I was always good at maths (or so I think), and it made sense to me — the concepts, proofs, the language of maths itself. So, I decided to jump into what this field has to offer, as they say it had a bunch of maths. And yes, it was a bunch of maths and theory, which was very opposite of what I did as a developer. Paper-pen with concepts learning was the key, and intuition to what exactly and how exactly things (regression/classification, models, tensors, ANN, CNN, RNN) worked as they are.

So 1st attempt was in my 2nd year I guess. It was from Coursera’s Andrew NG’s course. TBH, I was just parrot-learning, making Notion notes, and because even if I understood the maths of schools and colleges, it was quite different when it was not on paper and done in head (as dev habit). I was quite overwhelmed, and I convinced myself that I knew what I was learning until I did not. Soon, of course, the spark was all gone. I, as almost many people do, dropped it.

Fast forward, I went on YouTube learning through tutors like Krish Naik, 100 days of ML, CampusX (he is the goat, btw) 100 days of DL. At this point, I think I made myself comfortable with all the jargons in this field, but still it wasn’t clicking as it should (resistance free).

Press enter or click to view image in full size

Then I had subjects like Numerical Methods in 4th sem I guess, which felt somewhat similar to what I’d studied in classical ML. Then came 5th sem, the foundational subject “Probability and Statistics.” I had learned fundamental concepts like probability distributions, random variables, central limit theorem, and it was quite fascinating.

Press enter or click to view image in full size

Finally, after my 5th sem, I am getting interested in Engineering as a whole. Subjects like COA, Data Communication were peak. I had a burning desire to learn anything on my way, and this time my rate of learning, ability to grasp concepts quickly overshoot — it felt good.

Now, the 4th attempt, which was mentioned at the starting context:

The day-to-day coding felt monotonous and not mentally stimulating enough — less of my interest, to put it correctly. Don’t get me wrong, there are fields like cloud, data, system design, LLD that I find fascinating. But the regular development work wasn’t scratching that itch.

I never understood this research thingy — getting papers published and going on to pursue masters in field of interest under TA/RA-ship. I wanted to read papers and get familiar with this world. PyTorch came into my mind, and it indeed was the go-to tool for modeling and referenced by many AI researchers for their research.

I found a playlist of CampusX on it: Practical Deep Learning with PyTorch. And trust me, it was gold. I finally feel much less resistance in learning and implementing stuff like ANN, CNN, RNN. It was creating a base for my deep learning understanding. I revisited concepts like activation functions, normal training pipeline (forward pass, loss calculation, back propagation, optimize). I understood the difference between SGD and batch-GD. The thing is, everything is clicking very nicely.

It isn’t that I’ve gained a lot of knowledge — some of my friends are into this field and are learning them from 1st principles, good for them. It’s that I’ve hopped onto too many fields: design, dev, editing and this. So, it has become quite clear to me as to what not to do and what to do, to some extent. And trust me, this clarity is very comforting. The math is mathing, my cognitive and critical thinking ability has improved quite a lot.

This was a very vague and not-so-coherent way of me dealing with this to document somewhere. So that was it, thank you :)