A research internship differs from an industrial internship in a number of ways. The line between them is blurred to an outsider, which is why there often exists a lot of confusion about what exactly does one do as a research intern. One significant difference between the two is skill set requirement.
Industrial internships are a brilliant way for companies to hire talented individuals on a temporary basis, and use that time to educate and instruct them; mold them into the company culture. If deemed a good fit after the duration of the internship, the company extends their initial offer to a co-op and potentially, a full time.
Research internships however, require you to bring to the table, a more or less complete skill set. You are expected to already have a clear picture of what you will be working on and what your course of actions are going to be. Unlike industry, your efforts won’t be immediately directed towards designing or adding value to a new product to be released in market, but to contribute to the vast existing scientific pool to further expertise in a field. I chose research over the summer because I identified what my area of interest was, and the experience I needed to land myself a job in industry after I graduate.
That being said, research, on paper, is difficult. You are required to add some value to a field teeming with the existing work of brilliant scientists and experts. I had taken up the Machine Learning course, ENTS 669D, during my second semester. The course, jointly instructed by Dr. Rama Chellappa and Dr. Behtash Babadi, quickly became my favorite. I knew what I wanted to do ahead, but was also aware that one course will not be enough to land me a role as a machine learning engineer.
 |
| Nihal Paul, during a break from his otherwise hectic time at research |
I needed to gain more experience in this domain and that is when I had the opportunity to work under Dr. Chellappa. One of the best at what he does, working under him was inspiring, to say the least. He had no prior experience in using machine learning to further telecommunications, which is what I, as a research intern, brought to the table. We discussed several potential applications of machine learning techniques to telecommunications. There were QoS predictions, traffic pattern analysis and network security, to name a few.
My work was in network anomaly detection where I intended using powerful machine learning techniques to predict any malicious activity over a computer network. Not getting into too much of gory technical detail, deep learning (a machine learning technique) is where the AI and machine learning community is heading toward. It works better when fed with more data, which is why fields like face recognition and speech translation have grown tremendously over the past few years.
Computer networks are the single biggest carriers of data today. You may have, for example, the best and most sophisticated water tanks in the world, state of the art water heaters in your house, taps that are brilliantly motion sensitive, but without pipes to carry any of that water around, your whole set up is pretty useless. And if I am going to use the same analogy, if I could have a system that could predict how good/contaminated the water flowing into my house is even before that water actually flows through my tap, I can take a lot of preventive and counter measures, thereby saving myself considerable time and resources. ‘Prevention is better than cure’ is the general tagline of prediction using machine learning anyway.
 |
| The deep network model results, run on training and testing data of network packets, both malicious and normal |
All in all, the summer has been a terrific learning experience for me in many ways. Primarily, I have gained valuable exposure in the machine learning world, without completely shutting the door to opportunities in the networking field as well. Research requires a lot of patience. Just when you think that you have it all ready, a silly bug in your code will give you countless sleepless nights. But consistency is the key. A vision for what you want to do, even if it not perfect, is more than enough to get up and running. And it WILL eventually get you over that hill slowly, albeit steadily.
About the Author: Nihal Paul is a second year graduate student at University of Maryland, pursuing research on employing machine learning techniques in telecommunications.