An interview with Snehasis Banerjee, Scientist at TCS Research and Innovation
This is a part of the interview series I have started on data science and machine learning. These interviews are from the people that have inspired me, that have taught me (read are teaching me) about these beautiful subjects. The purpose of doing this is to mainly get insights about the real-world project experiences, perspectives on learning new things, some fun facts and thereby enriching the communities in the process.
This is where you can find all the interviews done so far.
Today, I have Snehasis Banerjee with me. Snehasis is a Scientist at TCS Research and Innovation. His research interests include Web Reasoning, Machine Learning, Information Retrieval and Recommendation Systems. He got awarded as Senior Member by ACM (Association for Computing Machinery) in 2019. He is also the Chief Moderator and a Technology Expert of TCS Campus Commune’s Artificial Intelligence Channel. Off the work, Snehasis is an ardent painter too (Bhushan in Fine Arts). You can know more about Snehasis and his projects from here.
I would like to wholeheartedly thank Snehasis for taking the time to do this interview. I hope this interview serves a purpose towards the betterment of data science and machine learning communities in general :)
An interview with Snehasis Banerjee, Scientist at TCS Research and Innovation
Sayak: Hi Snehasis! Thank you for doing this interview. It’s a pleasure to have you here today.
Snehasis: Thanks Sayak, it’s good to be with you. Your passion for ML is an inspiration for many. Let’s get started!
Sayak: Thank you for the kind words :) Maybe you could start by introducing yourself — what is your current job and what are your responsibilities over there?
Snehasis: I am currently a Scientist at TCS R&I, Kolkata involved in AI work since I joined the lab. I am professionally hyperactive (ACM, IEEE, CSI, …). As an example, I am a CSI Resource Person in AI. Currently, in TCS Research, we are trying to build a cognitive platform for robots. The main theme is cognitive robotics — an interdisciplinary field — and accordingly, we do have an interdisciplinary & varied team. I look after the planning, reasoning & reinforcement learning in the current work. The final target is productizing the solution. Earlier, I had worked on core data science & ML for a considerable time, but now it’s time for related subjects to synergize and go one level up.
Sayak: Tremendous engagements, Snehasis. Also, your will to broaden your horizons is really dauntless. I am curious about how did you become interested in data science and machine learning.
Snehasis: This may sound a bit strange! I was and am extremely curious about everything — one of the many questions I wanted to explore is what is intelligence? That lead me to AI — you know how something works when you can build it. So interest in ML (stats with AI touch), Data Science (stats with optional AI algorithms) followed accordingly, apart from many other AI disciplines. Also, the current industry focus is targeted towards predictive analysis with data — hence focus shift is inevitable. Earlier it was: Pose a problem -> Get relevant data -> Solve. Now: Data -> Pose a relevant problem -> Get a solution. Data has become abundant and it will grow more…
Sayak: I absolutely second the inversion in the first two steps. When you were starting what kind of challenges did you face? How did you overcome them?
Snehasis: Specifically for data science & ML related projects, the main challenge is data gathering, data cleaning and finding apt representation and features (transforms over data). Finding an apt model algorithm (ML) part is well researched. So engineering effort is the most time-consuming. Also, domain knowledge-based related to data is crucial — this is a challenge and we had to speak with experts (say doctors or manufacturing plant engineers) to get a sense of data and features that are meaningful. For healthcare, data is sensitive — so ‘proper’ anonymization without losing valuable info needs must be done (look up ‘k-anonymity’ for this).
Sayak: That is good homework I shall do, for sure. What were some of the capstone projects you did during your formative years?
Snehasis: I started with recommendation systems for TV programs based on EPG dataset. Looking up formulas in research papers and actually implementing them needs both research as well as engineering aptitude — this effort can be thought of sub-projects. In industry, without demo (working systems end-to-end), nothing is valued — so that rigor was rooted in formative years.
Sayak: I am also a firm believer in getting hands-on with a concept as quickly as possible. Good to know about how you shaped your mindset with the project experience. Tell us about your NeurIPS gig — the paper you presented over there and how was the experience overall?
Snehasis: Due to the justified craze in Deep Learning, the conference NeurIPS (changed from NIPS) is #1 CS conference in terms of attendance, where registrations sell off in a few hours if not minutes. You can meet and learn from the best people (including many Turing Award winners) who share the mission to understand intelligence by trying to replicate it. Our paper was related to automating machine learning task for sensor signals and showed that for small to moderate data, automation of feature engineering works better than black-box deep learning approaches. An extended work later also produced a Springer Nature journal publication (link to the extended paper).
Neural Information Processing Systems (NeurIPS)is a machine learning and computational neuroscience conference held every December. — Wikipedia
Sayak: An experience worth the while, really :) These fields data science and machine learning are rapidly evolving. How do you manage to keep track of the latest relevant happenings?
Snehasis: Due to the field’s vastness, you can’t be an expert in all. So choose a sub-topic (say Adversarial Networks) and aim to be a world leader in it. Keep track of other developments by setting Google Alerts and following conferences such as NeurIPS, ICML, IJCAI, AAAI, ICDM, KDD. Each conference has some tutorials & keynotes — go through them if available online.
Sayak: I am glad that we are on the same page regarding keeping track of the relevant happenings. Being a practitioner, one thing that I often find myself struggling with is learning a new concept. Would you like to share how do you approach that process?
Snehasis: When learning a new topic, try to get behind the philosophy first — why the guy wanted to do this — there were other solutions — he could have gone on a road trip in the time. Once philosophy & motivation is clear, other things can be related. You can listen to their lectures & interviews (say Youtube) to get clarity behind their thought process. Next, implement it. What you can’t build (even in mind), you do not understand. Limited time will not permit all things, hence focus on a specific sub-field, and in other fields, you will have a shallow concept — that’s ok, this must be accepted. Motto: Jack of all trades and master of some.
Sayak: This thought process is gold, I must admit! Apart from your work, you are associated with a number of professional and social activities. How do you manage your time so effectively?
Snehasis: Time is a limited resource. I assign an imaginary numeric cost to it — this interview also has a cost. I set goals with deadlines — with self rewards. I have buckets of time — project work (identity in AI and help TCS grow), learning (sharpen the saw), professional (expand network & propagate AI), social (feels good), family (mental support and sharing) and Others.
Sayak: Those are really nice distinctions. Thank you so much Snehasis for doing this interview and for sharing your valuable insights. I hope they will be immensely helpful for the community.
Snehasis: I always love to contribute to the AI community & to empower students. Thanks for giving the opportunity! See you soon!
Summary
It is so amazing to see Snehasis and his team’s effort to build a cognitive platform for robots. From being a well-rounded practitioner to a researcher to a life-long learner, Snehasis has it all with a generous amount of ownership. His work is so diverse — from starting with recommendation systems to doing work around reinforcement learning! “What you can’t build (even in mind), you do not understand” — these words could not be any truer.
I hope you enjoyed reading this interview. Watch out this space for the next one and I hope to see you soon.
If you want to know more about me, check out my website.
