The First of Written Interviews: Invaluable Advice by Abhishek Das
A GaTech PhD Student, Facebook Graduate Fellow & Reviewer at PAMI, CVPR, NIPS, ACL, ECCV and ICCV
Abhishek’s achievements in the field of CS are jaw-dropping; they make you believe that intellectual curiosity, dreams and hard-work have no bounds. A CS PhD student at Georgia Tech, Abhishek has bagged 5 renown fellowships and interned at places like Facebook AI Research and DeepMind (London). He actively publishes and serves as a reviewer at some of the most prestigious AI and CV conferences like CVPR, NIPS, ACL, ECCV, PAMI and ICCV. He has also achieved many laurels like the Outstanding Reviewer Award for CVPR (2017), Best Student Paper Award (ICML 2016 Workshop on Visualization for Deep Learning) and the University of Queensland (UQ) Research Scholarship. During his undergrad years at IIT Roorkee, he was a 2-time Google Summer of Code Scholar under the Department of Biomedical Informatics (Emory University) and the Open Web Application Security Project.
About Abhishek 👨💻
- Currently pursuing a PhD at GaTech, under the guidance of Prof. Dhruv Batra.
- Former intern at Facebook AI Research (Montréal), Facebook AI Research (Menlo Park), Virginia Tech (Blacksburg),Queensland Brain Institute, GSoC’13 and GSoC’14.
- Selected for the Facebook Graduate Fellowship, Microsoft Research Ph.D. Fellowship, NVIDIA Graduate Fellowship, Adobe Research Fellowship and Snap Inc. Research Fellowship.
- Over 17 Journal Articles, Preprints, Conference Papers and Workshop Papers.
- Among top 30% reviewers, NeurIPS 2018.
- Speaker and panelist at the Allen Institute for Artificial Intelligence, Montréal Institute for Learning Algorithms, IIT Kanpur, Toyota Technological Institute at Chicago, CVPR, ICCV, ICML, NVIDIA GTC, SIGDIAL Special Session and CoRL,Spotlight.
- Won the 1st prize in the Deloitte Collegiate Cyber Threat Competition, Microsoft Code.Fun.Do. (IIT Roorkee), Yahoo! HackU! (IIT Delhi) and Adobe Express Apps (IIT Roorkee).
By no means is this list exhaustive. It’s the tip of the ice-berg, actually.Make sure to visit his website, to read his projects. His CV will make you go wow! :)
The Written Interview ✍️
Q1: Bagging multiple fellowships is a remarkable feat. According to you, which aspects of your profile stood out while applying to such research fellowships?
A: Given that I don’t have transparency into the process and that there is possibly a fair amount of stochasticity in these decisions, it’s quite a fortuitous turn of events to be selected for so many fellowships! :) I imagine recommendation letters have a huge role to play. Other than that, the application typically consists of a research statement. I tried answering the following questions in my statement:
- My long-term research vision / goal,
- Why should others care about it?
- Why am I the right person to pursue it?
These questions (or some paraphrasing of these) initially came from Dhruv (my PhD advisor) in a different context, but I think they are quite broadly applicable and provide a good framework to think when drafting a research statement (or just about any research proposal). It took many iterations to get it just right. So yeah, recommendation letters + research statement + prior work + lots of luck!
Q2: As a reviewer of top-tier conferences, you have certainly come across a diverse range of submissions. Based on your experience, what advice would you give to students aiming at submitting papers to conferences like CVPR ( or to individuals aspiring to be reviewers at such conferences)?
A: Before I answer the primary question, I just want to say that reviewing is an amazing exercise to get exposed to good / bad papers and new ideas early on. The former helps with writing — you’ll soon start noticing patterns of what makes a paper clear and well-written, and conversely elements that make it hard to follow (lack of context before diving into details, unclear notation, etc.), and the latter broadens the space of ideas to explore when thinking of what to work on next.
Here are (what I believe are) some good pointers:
- Reviewing advice: https://www.dropbox.com/s/725p60wcajbb8xh/How%20to%20Review%20for%20CVPR.pptx?dl=0
- Writing advice: https://billf.mit.edu/sites/default/files/documents/cvprPapers.pdf)
- The CVPR 2018 Good Citizen workshop also had tons of useful advice: https://www.cc.gatech.edu/~parikh/citizenofcvpr/.
Q3: An article on the Facebook Research website states, “His research goal is to build intelligent agents that possess the ability to perceive the rich visual environment around us, reason and infer from perception in an interpretable and actionable manner, communicate this understanding to humans and other agents, and act on this understanding in physical worlds.” With regard to this, can you elaborate on any related project that you enjoyed working on?
Yeah, of course. One idea I currently find exciting is Embodied Question Answering, where the goal is to develop artificial agents / robots that are situated in real-world environments (such as our homes) and capable of answering natural language questions (such as “Where are my keys?’’) by navigating to explore the environment and gathering the necessary visual information. So it combines perception and language understanding with motor control. We presented some work on this in virtual environments at CVPR and CoRL last year (https://embodiedqa.org/). These were internship projects at Facebook AI Research, where I had an amazing time working on these.
Q4: Can you give us some insight (a brief overview) of your current work?
A: Sure. The two main thrusts during my PhD so far have been Visual Dialog (https://visualdialog.org/) and Embodied Question Answering (https://embodiedqa.org/). I’m continuing to push on those, in building better visual dialog agents — adapting to novel domains, reinforcement learning-based training algorithms, better evaluation, etc. — and better embodied question answering and more generally grounded language learning agents — in more realistic environments, for answering more complex questions, etc. — as well as exploring extensions to multi-agent setups, where these agents can interact with other agents (and humans) and learn to cooperate with each other.
Tidbits 🌟
We absolutely love his detailed and informative responses, especially on fellowships and reviewing. The resources shared by Abhishek are truly invaluable (check out the embedded links)! It is an honour to have someone of his caliber interview with us :)
