Summer Diaries: University of Tennesse Knoxville

Source: Deep Learning on Medium

Summer Diaries: University of Tennesse Knoxville

My attempt to summarize the mind-boggling journey I have had, from being in for the long haul while applying for research internships, finally landing one at the Imaging, Robotics, and Intelligent Systems Laboratory at University of Tennesse.

Scenario Before August 2018:

I started learning about Machine Learning in my first-semester midterm break and got interested in mathematical modeling of the algorithms. So after my first-year summer break, I took deep learning seriously and took a research project under a professor at IIT-Bombay. It helped me develop an idea of how a research problem is formulated and approached. It’s that summer which turned around my internship plans. How?

The experience during the summers helped me realize that research in Deep Learning was the primary thing I wanted to do, specifically I wanted to improve certain aspects for a better application to real-world scenarios. Having a research intern made me believe that I should be getting a lot of opportunities to work in some of the big colleges.

Landing that Internship:

After consulting with seniors about their experiences, I just didn’t believe I would be able to do the work that I wanted to do and hence gave a shot at mailing the professors at some universities in the field of Computer Vision.

A tip I would give is to start early in emailing professors regarding internship opportunities. I kept on emailing professors at various labs, sending well-curated and specific emails. In March, I got an email that I had been selected as an intern at UTK.

Apply to places that actually interest you, else you might waste your whole summer intern hoping you had interned somewhere else. Be specific, to-the-point and polite in your emails, but don’t dwell too long in drafting one. A little blunt here, but if a professor is searching for an intern, they’ll show interest; if not, there won’t be a reply. It’s as simple as that.

Install Mailtracker to at least give you the satisfaction that your email has been read (or misery that you got rejected besides the prof reading it, half-full half-empty scenario). However, make sure that you leave no evidence that you’re using it. Professors do deem it as off-putting if they receive a mail with Mailtracker.

Don’t be hesitant or judgemental of applying to lower-ranked universities. You need to get rid of the mentality that “ranking of college > research lab”. It works the other way around in research.

Just hang in there. You’ll spend lots of hours sending emails. You will hardly get a response. I am talking about 5–6 replies from 45–50 well-curated emails. Even most of those will be negative, but the prof was kind enough to reply. Some people get lucky and have their summer plans sorted by November. Some don’t and have their plans sorted out by May. But if you persevere, you’ll get through. 🙂

Work Experience:

My advisor was Professor Mongi Abidi, whose current work was primarily to find a good object detection architecture for videos and images. Dr. Abidi has been conducting research in the field of three-dimensional imaging, specifically in the areas of scene building, scene description, and data visualization. Our work was involved in the areas of robotic multi-sensing, landmark tracking and sensor calibration, data fusion and probabilistic reasoning, and enhancement of medical images.

For training a single shot detector architecture, you need a large amount of dataset and with that a huge amount of computational power. I tried to increase the accuracy of the detection of video sampled images and improved on the architecture. I was given an ongoing project with another Ph.D. student there. We collaborated and discussed ideas to improve the model. The workload was exhausting. The cherry on top was the days when training the model for several days continuously turned out to give an increase in the MaP score(accuracy) of only 0.09%. (It was demotivating!!)

The atmosphere of the workplace, however, was quite relaxed, with extremely flexible work hours. People solely care about the work getting done by a certain deadline, although that was also flexible for me as I was working on something completely new. People respect your personal space and work, allowing you to shift deadlines in case of some blow up in the code. The American culture is very different, with trust and openness very important virtues.

Outcome

The major takeaway for me was that research isn’t always as sugar-coated and straight-forward as it was in my previous summer with IIT Bombay, that you have got to stumble and explore a lot before you make some progress, especially when you don’t know how to proceed. Having clarity and a record of explored directions helps a lot in these situations. But if you ask me now, if waiting for 4 months with my hope rising from each email notification just to see it crushed by a reject or unrelated matter, if I would go through it all again to experience what I did during the summers, I would say definitely, because hey,interacting with so many people and making new friends,getting to know a new culture,achieving a more complete sense of what research entails was worth every single moment.

So don’t be afraid to take that risk and go after what you want to do! Internships are for exploring your options after college, and unless you don’t put yourself out of your comfort zone, you’ll be robbing yourself of that chance to find what motivates you and what your limits are. Mind you, this doesn’t mean that you apply without any experience or work to show. 😛

PS: We were able to get a good MaP score(accuracy) until the end of the summers.

PPS: It was because of the research Internship that I landed a research intern at the National University of Singapore coming winter.

Cheers!