Every year-end, with just a few days left in it, institutions, individuals, and various other entities typically undertake a review exercise – a reflective ‘looking back’ of sorts where they take stock of the highs and the lows that marked the year for them. Among students on campus, we found Vaibhav Gupta, a 5th year Dual Degree student in CS who has consistently made it to the Dean’s Merit List for all the academic years, for whom the year 2019 turned out to be quite eventful. Read on.
For someone who only knew that the dual degree programme was “longer than the conventional BTech programme”, and was initially a little skeptical about it, Vaibhav Gupta says that in retrospect it has been “a fairly decent journey so far”. He currently works in the Machine Learning Lab with Prof. Praveen Paruchuri in the area of Deep Reinforcement Learning (RL). For the uninitiated, RL refers to the training of machine learning models to make a sequence of decisions to maximise reward in a particular situation. As a third year researcher, Vaibhav chose to intern at IIT Madras under Prof Balaraman Ravindran, an expert in RL who heads the Robert Bosch Centre for Data Science and AI at IITM. It was based on his work at IITM that a paper “Advice Replay Approach For Teacher-Student Framework” was accepted at the International Conference on Autonomous Agents and Multiagent Systems (AAMAS) in May 2019 at Montreal, Canada.
Weak vs. Strong Autonomous Agents
Explaining the paper as a framework based on simple principles of nature, Vaibhav talks of a “weak” Reinforcement Learning (RL) agent trying to learn a complex task in the presence of a “strong” experienced agent who knows how to solve that task. “A natural approach is one where the experienced agent would help the new agent solve that task more efficiently and quickly. In this setting we propose different ways in which both the agents can exchange their knowledge,” he says. An immediate real-time application of this can be found in virtual assistants such as Cortana, and Siri where the weak agent can be in hand-held devices while the strong experienced agent can be likened to complex servers.
Believing in gaining a myriad of experiences, alongside the research internship at IITM, Vaibhav was simultaneously doing an internship project in the much coveted Google Summer of Code (GSoc) with the Red Hen Lab in 2018. He was also part of the Student Placement Council successfully coordinating not just internships but also full time placements for students in high-profile companies. Over the last three years, Vaibhav has been doubling as a Teaching Assistant for course topics such as Computer Programming, Computer Networks, Distributed Systems and Artificial Intelligence.
In 2019, just after the AAMAS conference, Vaibhav took off to Munich where he interned as a Site Reliability Engineer (SRE) at Google from May-August. Talking of his work at Google, he says, “My project comprised of designing and building a complete tool, including an administrative interface to annotate datasets (comprising of images, text, videos, audios)”. Even with the Google internship underway, he was constantly mentoring students who were interning at GSoc 2019. Not one to waste time, Vaibhav got back from Munich only to sign up immediately for yet another internship at Facebook, London from September-November. His project there had him design and implement a universal language that can port to languages like HackLang, Python and JS.
Back on campus after being out for the greater part of 2019, Vaibhav can be found plugged into music when not socializing with his peers over deep discussions and debates. “Music is my first love. I listen to it all the time,” he says. When he is not working himself into a sweat at the gym, he can be found unwinding over a game of pool or bowling. For someone who has been on a winning streak whether in high school (he won the 22nd International Rank in the International Science Olympiad), or as an engineering student (He qualified twice to the regional finals of the International Collegiate Programming Contest – ICPC), he’s extremely modest and explains it as having access to the right opportunities at the right time. Ask him if he has job offers lined up, and he cryptically replies that he has other plans for the near future, with a job ranking low on his priority list.
Made For Machine Learning
As a student graduating from school, Vaibhav originally intended to pursue a career in Chemistry when he was guided towards Computer Science by well-meaning teachers and other well-wishers. Now, when you hear Vaibhav speak passionately about his love for Reinforcement Learning, you can’t help but feel he made the right choice. He signs off saying, ”Observing patterns in real life, translating them into algorithms and making them usable for actual machines fascinates me”.