Today we’re here with Aditya…..
You might know Aditya from his numerous mentions of workshops on Python, Machine Learning and Research, on LinkedIn. Given his 7k+ followers, it is highly likely you already follow him on Linkedin. He is a Google AI ExploreML facilitator, founder and Team Leader of Cognitive Applications Research Lab, Technical Head and Research Mentor for GirlScript Foundation Chennai amongst other roles. You can also check out Aditya's personal website phreakyphoenix.tech.He also worked at SRM Team Robocon and has been a mentor for Google Code-in 2019 with TensorFlow. He loves teaching and communities and has volunteered with a number of organizations, GDGs and TFUGs. I’ve known him as a witty guy who is always up to hacks and absurd technology applications . He also posts about free software and cool tools on LinkedIn. On a (slightly more) serious note, we share our passion for research and ML, the two key tenets of this interview.
I also encourage you to check out my conversation with him about the Explore ML program. He had so much more to talk about research and his experiences, and that's what brings us here.
Q) What inspired you to get involved with research? What fields are you most passionate about?
Not money :P I always wanted to do astonishing things, which people wouldn’t usually do, I once just left home and went backpacking through Darjeeling and Bhutan, with no ticket and plan whatsoever, like people go to buy eggs. So yeah, I have a little adventurous streak and a fire to discover new things, it could have percolated into my career decisions as well and research gave me an outlet to pursue that. It guaranteed that I’d never be bored.
Yeah, this might not be as inspirational as you were expecting, but thrilling nonetheless, and here’s a fun fact, I was equally good with Bio and CS, the reason I took Computer Science was because they had air-conditioning :P Our biggest decisions sometimes have the weirdest reasons.
I’m broadly interested in all forms of intelligence, of both machines and humans, I’ve explored Expert systems and ML till now. I’m most passionate about research with images, be it recognition, generation, segmentation etc. and have also worked with encryption.
Q) I'd love to know what got you interested into research. Could you also highlight your research experiences and challenges you might have faced along the way?
I never meant to get into research per se, it was more of the consequences of how things turned out. During class 12, we used to take computer tuitions and there was a girl Shritama who had designed a TicTac Toe game in Java. In this game, you have to place 3 X’s or O’s in a row column or diagonal to win, which we’ll call a triad. Her program could complete the player’s triad to win as well as block the opponent’s triad to prevent it from winning. This inspired me to try to make this on my own, which I called ‘TicTacToe Beginner’. Then soon I started improving on it, reading books on strategies for the same. I created two iterations ‘TicTacToe Novice’ and ‘TicTacToe Advanced’ over the next four days, working almost 24x7, and the codebase was messy with over 6k lines of code.
When I joined college, I wanted to get into research, since I loved algorithms, and would have loved to develop new ones. A senior was impressed by my work on modifying heap sort, but advised me not to pursue pure algorithms research as it’s not in demand. He was right, no one cares about a faster implementation of a sorting algorithm, today people are more interested in cool things (read AI, Block-chain etc). He suggested I could instead make better implementations of algorithms used in AI, which resonated well with me. At that time, I did not know much of Machine Learning, and decided to work more on my game to finish it. I was inspired reading about how BFS, DFS, A*, SSS* and Scout algorithms were developed, I read up on the lives of the Godfathers like Donald Knuth, Hans Berliner, Judea Pearl and Djikstra, and started listening to their lectures.
I got my code base to a manageable 1200 lines, it’s acceptable since it’s an expert system, and not an ML algorithm.I called it T3DT (Tic Tac Toe with Decision Trees). My focus was not just on speed though. I wanted to make it human like. This was a huge challenge as standard approaches like Minimax fail miserably at this, as they are non-randomized, plus they are slow, more on this later. To validate that my game was indeed no-loss, I had my algorithm play against minimax, both when the AI player (T3DT) and the opponent (Minimax) starts first. My algorithm never lost in all the possible games; elated by this, I started doing speed tests, and found that my algorithm was in fact upto 23k times, without JIT compiler optimization. In the real world, against a human player, this is not a big deal as humans won’t notice the difference between millisecond and nanosecond response times, though it was a huge jump. Now I faced an enormous problem. As that senior had rightly pointed out, my algorithm could indeed be SOTA, but it was a field no one cared about. Hence it became difficult for me to publish my work, and somewhat demotivated I kind of gave up on all this. However it made me think about how we could use ML to generate explainable Expert systems, it’s in my repertoire of ideas to work on.
That year, I won Research Day in SRM Univ amongst 5000+ students and research scholars with my algorithm. The judges were really impressed by the blue-sky nature of my research, refreshingly different from anything else that was presented. That got me back on my feet and I sent it to a few very high impact factor journals, though I got rejected first.
After that, due to Robocon, another team I was working in, research took a back seat, and I also started pursuing ML. Fast forward a few years, I established Cognitive Applications Research Lab, a student-run research lab of my University, to educate students about the Do’s and Don’t of Research, how to write and format papers, and to help them avoid the mistakes I made. I also got into Image Encryption Research working with a Professor here.
Then, I got an opportunity to work on a SPARC funded research project on Detection of Non-Proliferative Diabetic Retinopathy, in collaboration with UC Davis and in association with IIT Kgp, and MHRD, Gov of India. That was a really cool experience. I did not wish to pursue Image Encryption forever and wanted to switch to ML, and this gave me a nice boost. I also got an opportunity to present a similar project related to Diabetic Retinopathy to experts at Google, and got a chance to be mentored by them, including GDEs, which was a unique experience. I won research day again the following year, making it twice in a row. Currently I continue working on a few image encryption papers, and ML.
Q) Coming from a tier 2-3 university, the university places far lesser focus on research and quality publications. How did you overcome this barrier and kick start your research journey?
Actually, I never thought of my college in any tier, I feel it’s a shackle people put on themselves, believing that since they are in a certain place they have to behave in a certain way, this mindset is restrictive and in my humble opinion, not very productive. I took every opportunity I got, it wasn’t always easy, but looking back, it was completely worth it. There was a quote along the lines of, If you do something you love, you never have to work a single day of your life, it’s apt for describing why I do research, I just don’t want to work :P
I thank Aditya for all the efforts he's put into answering my questions. Also, I really appreciate your efforts into teaching tech and spreading awareness about research to a wider audience:) Really hope you keep inspiring your juniors in future, (and post more free resources that everyone can use xD)