Rajaswa Patil is a research assistant at MIDAS-IIITD. He is a 4th year electronics student at BITS Goa. We recently talked to him to understand how he got into machine learning.
1. How did you decide that you want to start getting into the specific field you are in?
I started with ML due to motivation from my brother (who was working on self-driving cars back then). Like any other ML enthusiast, I started off with Computer Vision in ML as its the easiest to comprehend at a basic level and at the same time interesting enough. I caught a bit of interest in Natural Language Processing when my mother suffered through brain hemorrhage and was diagnosed with Aphasia soon after (a speech disorder). Further, I did a summer internship in NLP at IIT Bombay, which was an interesting experience for me.
This made me conclude that NLP is the thing I want to pursue.
2. How important is knowledge about specific subfields such as graphics or neurology when doing starting off in ML/AI. What are the prerequisites for starting ML?
The only pre-requisites to ML are high school level calculus, matrix calculations & linear algebra, and basics of probability & statistics. All of these are more or less covered completely by the end of your first year at BITS.
3. Is it any practical problem that motivated you to dive into this field, and if yes, what was it, and were you able to solve it?
During my time at IIT Bombay, I was supposed to build a data-domain and language-independent opinion-mining pipeline for online texts. It is a very challenging problem and almost demands human-level multi-lingual language comprehension skills. I wasn’t able to achieve near perfection in the task (nobody has), but I did manage to find a simple linguistics based workaround for it. We could say the problem was partially solved. This also made me notice that ML is not the answer to everything, some answers in AI will be answered by core disciplines like linguistics for NLP and Image Processing for Computer Vision. This gave me a new curious perspective on AI research and motivated me deeply towards the same.
4. What do you recommend to the people starting to get I to research after some reading? Where do they begin after they have some base knowledge to think about it?
The decision of jumping into ML research should be very very calculated. The field of ML research has a very high pace and is very competitive. You need to be exceptionally devoted to it in order to survive. Further, any research career will be more longer (2yrs MS + 5yr Ph.D.) than any other job. Keeping all this in mind, I would suggest that you work under a faculty or senior on some problem statement which is not too ambitious. This first experience should not be used to target a very good achievement, rather it should be used to measure how much interest are you building up working in a research environment.
5. Apart from computer science students can other branch peeps take ML/DL as their career? What is the scope for them?
Yes, you definitely can. ML has become quite interdisciplinary. Branches like EEE, Maths, Biology, and Physics have become an integral part of ML now. Apart from that even the core humanities branches like Linguistics, Psychology, and Cognitive Sciences cannot be ignored at any point in time in the future, lots of scope with these branches as well.
6. Other than online classes, what are some other areas that one should focus on if he wishes to choose a career in ML
Programming practice. This is where most of the ML enthusiasts fail.
Open-source experience. You should know how to re-implement existing work available online.
7. What are some things that differentiate a AI researcher and ML ?
I am not sure if I understand this question completely. I can tell you the differences between an AI researcher and ML Engineer though. An AI researcher will usually focus on improving the current state-of-the-art ML algorithms, whereas an ML Engineer will focus on deploying the existing algorithms for real-world application in a cost and compute effective way.
8. Can you explain how the application is as important as theory and in many cases more important? How does one start the application of the theory?
We know that necessity is the mother of invention. Without any incentive of real-world application of theory, there’s no driving force for the research to progress.
For ML, this relationship is quite visible in plain sight. Most of the research fundings and even research contributions in ML come from Industry (take some corporate giants, for example, Google AI, Microsoft Research etc.) Further, most of the ML conferences and journals are funded by Industry, which expects a certain return in terms of value addition to their products. Usually, any conference/journal will have a list of research topics of interest. Usually, more than half of these topics would be in the interest of the industry sponsors’ products and domain.
9. How did BITS Goa aid your interest towards machine learning?
The coursework in the first year: Prob-Stats, Calculus, CP helped a lot to cover/revise the pre-requisites without any extra efforts. Further, the on-campus seniors and clubs helped a lot too. I took my first comprehensive course in DL with QSTP (Quark and Saidl). After that, the on-campus organizations (Saidl, CTE, Quark) helped a lot in terms of community support, a platform for sharing & discussing my interests, and most importantly connecting with opportunities and ideas that I was not aware of.