Shweta Jain: At the intersection of game theory and machine learning
July 31, 2024 | Bharti Dharapuram
Shweta Jain is an Assistant Professor at the Department of Computer Science and Engineering, Indian Institute of Technology (IIT) Ropar. She completed her Masters and PhD from the Department of Computer Science and Automation at the Indian Institute of Science (IISc), Bangalore. Bharti Dharapuram caught up with her at the ACM (Association for Computing Machinery) school on ‘An Invitation to Algorithmic Game Theory’ hosted at IMSc, Chennai where she taught several lectures.
What are your broad research interests?
I work at the intersection of game theory and machine learning. A lot of interesting problems involve solution techniques from both these fields. The field of ML-AI (Machine Learning and Artificial Intelligence) is evolving at a rapid pace. What excites me is how theoretical analysis can help towards finding efficient solutions to its practical applications.
An aspect of game theory is that it is a perfect information model. Players understand valuation, utility model, choices made by others, and based on these one can run analyses and derive equilibrium strategies. Game theory says that if people are rational and intelligent, they will behave in a certain way. However, if you consider a situation where the world is not perfect, one wouldn’t know this information beforehand, then my research intends to find out how agents behave in such situations.
In my talk at the summer school, I spoke about the multi-armed bandit mechanism, which combines multi-armed bandits from reinforcement learning with auction theory. In this problem, a social planner and agents realize the reward of an action only after that action is taken. The question that we are trying to ask is, eventually, when everyone starts gathering more information about the rewards while exploring the actions, how does the strategic choice of the agents change over time?
What are some of the applications of your research?
One application is the sponsor search auction – Google gets paid whenever somebody clicks on an advertisement, and they want to learn the click probability of these ads. At the same time, they also want to learn how much advertisers value a particular advertising slot on the website, which is an auction problem. Another interesting application is the combination of congestion games with imperfect information. There could be multiple routes from my office to home, and I can only get to know which route is better when I actually take that route. This is a multi-agent problem because the cost of that route depends upon what routes others are taking.
Another example that I talked about in my lecture is demand response. Imagine that you want to reduce the peak load in a smart grid scenario. One practical way to do that is to charge people based on how much power they consume and also based on the peak time. A different perspective is to formulate the problem from a mechanism design perspective and design a game to incentivize consumers to reduce their load during peak hours. This is an exciting application because the demands are stochastic, and you are learning over a period of time while also trying to come up with a game to shift the peak demand.
What are the challenges that you have faced in your academic journey?
A primary challenge was that I had a two-body problem – my husband did his PhD in mechanical engineering at IISc and was also looking for a job. We were not particularly in favor of working at two different places, so our focus was to find a job at the same place and support our daughter. Every institute has its own requirements and expectations, so this was particularly challenging and an issue that a lot of people are facing. Academic institutes should start looking into this if they want good people to join them. I also had to do a lot of interviews, and sometimes, the process was really slow. I enjoyed doing interviews, talking to people, and traveling, though. It helped me make many friends across different IITs.
What is your advice for PhD students entering the job market?
I wanted to be an academician right from the beginning, not just because of research but because I have a huge passion for teaching. We wanted to be in India to be closer to family, so we never considered going abroad. I gave some interviews for industry jobs after my PhD, but I don’t think it was meant for me. The only downside of academia is its salary structure compared to that of industry. Otherwise, you have the freedom to choose your problems and shape your research. It is great if you are in academia and working in collaboration with industry – you have the freedom to choose among challenging problems and decide how to solve them. This is my personal perspective because a lot of people do enjoy an industry job. There are pros and cons to both, and I would say it is a personal choice. And a student should list his/her priorities before entering the job market.
What are your thoughts on representation of women in ML-AI research?
It is low, but I have seen an increasing trend over the years. Even among the computer science faculty at my institute, there are only four female faculty out of twenty-three. During my Masters from IISc in 2008-2010, there were only three women in an entire batch of about fifty. The representation is increasing but not at the pace it should. The drop-off is significant if you consider a family's expectations of marriage for women. I could only do my PhD after marriage because I was fortunate to find someone who was also interested in research and was supportive. That doesn't happen to everybody. However, these days, I am seeing a good representation of women in PhD programs, which is encouraging. Every conference has women-specific workshops or committees and we are discussing women in science. Women are coming together, collaborating and talking, which is nice.
