Dr. Aravind Srinivasan graduated in 1989 with a B.Tech in CSE from IITM, and is currently a professor of Computer Science and at the Institute for Advanced Computing Studies at the University of Maryland, College Park. He was awarded the Distinguished Alumnus award, 2016 at Alumni Night held on the 23rd July 2016. His areas of research include algorithms – especially randomized algorithms, networks, machine learning, algorithms and modeling uncertainty in computational biology, and network science, algorithms & technology for public health. We caught up with him the day before the ceremony, as he was in the campus teaching a GIAN course on Randomized Algorithms.
The department of Computer Science was founded in 1973, the B.Tech program started in 1982, and you started in 1985. What pushed you to the young field of CS? Did you have higher studies in mind?
My main interest was in math, and I wanted to get into a field that was significantly mathematical. One field that was suggested was electronics, and CS was an option. I also factored in what was popular then, and CS was quite popular at that time also.
So how has the popularity of CSE changed with time? Did it have the same demand it has now when you were studying?
Right now, the industry is booming and that has pushed up the popularity for computer science. Computers were not as ubiquitous in society those days as they are now, so interest has definitely increased over time.
There is this mentality here that someone from the CSE department is assured of a good placement and high package, and people from other departments don’t have the same level of placements. Is that thought process prevalent outside the country?
Yes, it is true that different fields have different distributions of salary. It is true that at the undergrad level, computer science probably has the highest pay packages, thanks to the software industry. But really, as we all know, what we earn later on has to do with the quality of our work. So this refers to just the initial salary. But if you do a masters, finance probably pays more. But this is probably true only in regard to starting salaries.
Do you think that this is related to the fact that engineering is such a common field of study in India? Has that perhaps been pushing computer science to the top of the preference order?
In India, I think it is due to the growth of the IT industry. That probably contributed more to the popularity of computer science.
In India we find a large number of engineering grads moving into finance or consulting. Is this common in the US as well?
I believe so. And I think it’s a good idea for society to give freedom to move around and explore. People think of IIT as a safe option, and we cannot blame them. I believe that significant freedom to choose is the best thing we can do for students.
This was one of the reasons given behind the recent fee hike imposed on IITs- that not enough engineering graduates end up becoming engineers. Your take on this?
I don’t know the exact numbers- but I am grateful to India. My annual tuition free was 200 rupees, my monthly “mess” (food) fee was more than this annual fee! But seeing that education is an expensive path, I don’t see a problem with such fee hikes, provided that students who cannot afford it are being given scholarships. And this is still less than that of private institutes, where the standard of education and infrastructure is probably not as high.
You’ve done a lot of work in public health networks. Can you tell us about that?
If you look at an infectious disease, you can think of it percolating through a network of contacts. So now we’re sitting together – if we shake hands, you might get the disease from me, if I already had it. And you may meet someone else and the process continues. In other words, you can create a virtual network where people are constituents and you can create links between them- like shaking hands, or coughing, or being in the same lift. This is a social contact network as opposed to an online social network. And you can imagine the disease percolating through a network through an initial subset of the population. This is actually where the term “viral” comes from, with regard to social media- these phenomena spread exactly like viruses do. So it could be social networks, or human networks or networks in the brain- but the underlying mathematical formulation is the same- there is an initial subset that is “infected” (maybe with a disease, maybe with an idea) and they expose it to the larger population. Social media companies are very interested in this kind of growth, as they see it as a means of advertising. Though the phenomena are different, our algorithms are similar.
So is this how you navigate between fields like biology & public health and theoretical computer science? Though you are a trained CS man, you do work in computational biology.
Actually, there are three aspects common to my research- algorithms, randomness and networks. My core background is in (randomized) algorithms. How does probability play a role here? Back to our disease analogy. If I have a disease and sneeze around you, it isn’t certain that you will get it. There is incomplete information and you have to study the statistics to come up with a probability that you contract the disease when I have it. We come up with models like this. Similarly in viral ideas, we model complicated phenomena by observing their spread through a network. So randomness is the second aspect of my work. Third is the network. Network models are everywhere nowadays- from social networks to human-made networks (e.g., the Internet, the Web, power grids) to networks where diseases can spread. So I like the interplay of these three. But independently, I’m also interested in learning more about biology and sustainable growth. So I don’t just want to apply my knowledge to those fields, I want to see what the big questions are there, and see if I can make a contribution to them.
How hard is acquisition of data for your social media research?
It is quite challenging. Thankfully, I have some collaborators who work with the acquisition of real-world data. Some companies don’t give data. It is important, but we are fortunate enough to get access to what we need.
Is data easier or harder to come by in your biology work?
It depends on the domain. If you get funding from the government, which is what most scientists do, there is a push to make your data public so that others can benefit from your knowledge.
Machine learning is now a buzzword in society. What is about machine learning that makes it garner this much attention?
Machine learning is much more than pattern matching. Pattern matching is- say for example that I’m trying to read a foreign script without learning the language. Then if I see “Coca Cola” written a certain way I can assume that that must be its written notation. ML is a step beyond- it is about generalization. A computer can easily memorize things. But if I’m given many examples of elephants through pictures and I’m given pictures of white things and tiny things, I can come up with a simple hypotheses that “elephants are not white or tiny”. This kind of general conclusion that we can draw is what makes ML special. In several domains now we have so much information. For example, our cell phone carrier has so much detail about where you were, how long you spent there, who you spoke to, etc. Similarly websites also collect lots of information through cookies. Companies now collect so much data, that it allows you to come up with statistically significant hypotheses. You can get petabytes of information just sitting at your computer. So we can come up with superb hypotheses and inferences. So just consider the internet of things. More and more home appliances are going to collect data. You can control your thermostat or your fridge through the cloud. Companies have so much information that (hopefully) do not compromise your privacy. The sheer magnitude of data available that was not previously there, promises several insights that were just not possible in the past. Data from images, medical data and so many more sources lead the push. And these inferences are beyond statistics, in as much as we need much less data to draw conclusions. Statistics allows you to say things like IF condition A is true and condition B is false, then C will happen 60% of the time. But to draw conclusions like this, we need gigantic quantities of data. ML here can come up with much more accurate inferences.
Do we have to program what inference to look out for?
Not always. There was this lab at Cornell University, where the professor came up with a machine-learning approach to come up with known physical laws, like gravitation. So we just tell it some basic stuff like “the law we are looking for is spherically symmetric” and ask the machine to observe systems for long times — they actually came up with valid laws. We can come up with scientific inferences with minimal human interference. This is being driven by two things. Firstly, machine learning algorithms are becoming very powerful. Secondly, lots of significant data is available. If you want to find a disease that affects one in a million people, you are going to need to see at least 3-4 million people to have a decent chance of finding it. But now combining imagining data acquired around the world, combined with the powerful algorithms and supercomputing, we can infer some very new things.
What are neural networks?
Neural nets are things that people hypothesize that are present in the brain. The idea is based on the neurons present in the brain. Now these neurons can fire meaning it emits an electric signal. So we can imagine it as a simple electronic element that has 3 wires. For simplicity let’s say each wire can be on or off, 0 or 1. And if at least 2 of the inputs are on, it fires a pulse. If the inputs are A, B and C, we can say the output is 1, if and only if A+B+C >1. This is a simple threshold gate. A neural network is composed of elementary gates like this, and the elements could be more functional than just threshold identifying. For a long time, these have been useful in pattern matching. Now we are thinking of deep nets, that are composed of layers of elements and these have been able to do a lot of interesting face recognition and so on. So there is a lot of buzz about them.
Can you tell us about your work at Bell Labs? What are the differences between industrial research and academic research?
I worked on internet telephony, and also content distribution on the internet along with algorithms and theoretical computer science. In industrial research, your research has to be more or less oriented in the direction of the company, and in academic research, in principle, you can chose to formulate your own problems. Most funding in the US is government based, but there is a lot of private backing too.
Your advisor here was Professor Pandu Rangan. What role did he play in shaping you?
He was a motivating teacher and mentor – to an infectious level. He inspired us to read research papers, and he was always ready to discuss them with us.
You are here early teaching your GIAN course. How effective do you feel the GIAN program will be, and what should its objectives be?
It seems like a fantastic idea to me. As you know, there is a shortage of faculty in the country and students want quality education. So I feel this is a way of scaling up if these can be recorded and made available to students in the country. I think this can solve the scalability issue through electronic distribution. I believe NPTEL makes videos available to interested colleges like this.
Could you compare the outlook towards research in India and abroad? How are the student-teacher relationships different?
In India I know only of the IITs and a few such places (e.g., IISc, IMSc, TIFR, CMI). The US has several more universities that are established, so you feel you are part of a larger community. India has a strong research community, but the sheer numbers are smaller. The US perhaps offers more opportunity through conferences that are primary drivers of new learning and collaboration, especially in fast-evolving fields like Computer Science. And about the student-teacher relationship. I thought it was very good in the IITs. I would say there isn’t much of a difference, as we encourage students to speak up and to ask questions without fear of being wrong.
What do you think about the new startup culture that has permeated India? Do you see it as a drain in the number of people entering academics?
I think it is great, it will go a long way to solving job shortages here. I don’t see why everyone has to be an academic. People should do what they like, and if startups can scale up and offer jobs, it’s a great thing.
Have you gone around the campus? Can you tell some changes that you’ve observed here?
Yes I have. It’s wonderful. The campus has grown more green and more diverse, there are more people here from other parts of the country, and there are more women. The non-engineering programs have also grown.
How important is innovation in the modern world? Is it possible to make it a habit? Perhaps a way to make it more frequent in daily life?
I feel that if failure is not ridiculed, it would help. One reason why Silicon Valley is so successful is because failure is encouraged- one can talk freely about failure. Even in the classroom, asking “silly” questions should be encouraged. Let people fail 10 times, and then they will succeed. But if every failure is treated as a shameful thing, people won’t make it. I think innovation is great. In India for instance, our wired telephone network, our wired infrastructure, wasn’t that great. But we just leapfrogged it by going to wireless. We didn’t build a great wired network and then move to wireless. Similarly in Africa. Innovation allows you to leapfrog problems- progress 20 steps instead of one at a time.
Could you share some of your happiest memories here on campus?
My last year was very happy, when I was fully into my research. It was intensive, but I really enjoyed it. Besides that, I loved going for movies at night and coming back from Mount Road at 11 pm by cycle. We went out a lot like that.
How does IITM compare to leading research institutes abroad?
I would say the IITs are doing very well. However, if you look at a place like MIT, the sheer magnitude of resources they have make it an unfair comparison. The endowment for Harvard, for example, for a tiny sized university is perhaps around 40 billion dollars. With that kind of money, resources, and alumni support, they are able to attract students and faculty from anywhere in the world. One thing I am very happy about is that our institute involves alumni very actively, and I believe the network can play a very positive role. The resources that a few select institutions abroad have sets the bar too high to compare with. It’s not that money can solve all problems- but say, to fund a new field, they can pour the kind of money needed quickly, and we will take a lot more time. Money isn’t the only factor. Immigration is easy in the US, it’s simpler to get a work visa there. Rather than compare with them, I think we should set goals that that we can reach, like develop startups, mentor programs for students, involve alumni more, etc.
Why is India so good at economizing? When you consider ISRO, the cost per km to Mars was cheaper than a Chennai auto. They are the most economic form of putting satellites into orbit and they have an amazing success rate. Similarly, medical goods manufactured here are heads and shoulders cheaper than their competitors abroad, with the same quality. Labor cannot be the only factor here.
I think constraints can be a benefit. Constraints foster creativity. Paradoxically, it is true sometimes that the more constraints that are placed, the more creative the approach. At the same time, major constraints could lead to fall in quality. It’s a fine boundary, but creative constraints can be a significant step to birth innovation. Being inspired by your work also helps along the way. With ISRO, it was a matter of national pride. Being passionate about your goal combined with creative thinking is an amazing combination.
What is your take on privatizing research? Craig Venter made leaps in genome sequencing by taking research out of the realm of pure academia. Should this be encouraged in the future?
There is no silver bullet here. I think government should fund, and at the same time, we should involve private entities. Creative research needs latitude, so whatever funding it receives should not have many strings attached, but at the same time should have strong peer review and feedback.
What are your interests besides your research?
My dream is to be more helpful to society. Instead of hobbies, let me tell you what my ambitions are. Society has been very good to me, and I’m trying to do something in return, especially to those less fortunate. My work in public health is something I hope can make a contribution. Besides that, my efforts have been ad hoc and small, so I cannot really say too much.
Given your position, is it not fair to say that you can make larger contributions through your work?
That’s a good point. In fact, my own research has been veering more in the sense of what is in the public good these days. I wish to make a lasting contribution as a thanks to society. This animates me a lot.
If you could give one piece of advice to students of the institute, what would it be?
I feel that several people are factors in our success; remembering them and contributing to something that is bigger than yourself is a great thing. Be happy and make others happy. If you’re a successful entrepreneur, then support others, especially in a value-added manner. I don’t mean just charity, perhaps invest it in something that will transform people’s lives for the better. This will solve a lot of problems in the world. If your success and happiness can be aligned to also caring for many, I think it would be a truly wonderful thing.