When Aditi was seven, her father pointed at the night sky and told her something that changed how she saw the world: the light from those stars took so long to reach Earth that what we see is already ancient history. He explained time travel and space travel in ways children could grasp – fascinating stories about girls living far away whom you would need thousands of years to meet, who would watch you only after your great-grandchildren were born.
“What stayed was not the astronomy,” she recalls, “but the feeling that some questions are so vast you could spend a lifetime approaching them and still not reach the bottom.”
That feeling led her from a state government college in Jhansi, a city carrying Rani Lakshmibai’s defiant legacy but not known for academic opportunity, through IIT JAM to IIT Mandi, and now to IIT Roorkee’s Mehta Family School of Data Science and AI. She is one of the handful researchers in India working on a problem most people have never heard of: restless bandits.
And she is convinced it might be one of the most practical mathematics problems there is.
The Problem That Won’t Stay Put
Imagine you are a doctor with fifteen patients and only five staff members. Or a defense analyst tracking ten submarines with resources to monitor just half. Here is the catch: situations keep changing whether you are watching them or not. Patient conditions worsen. Submarines move. The world won’t hold still.
How do you decide what to pay attention to when you cannot watch everything?
This is the restless bandit problem. Unlike simpler frameworks where options freeze until you interact with them, restless bandits reflect messy reality: everything evolves continuously, whether you are paying attention or not.
The standard approach uses the Whittle index, a score assigned to each option that tells you what to prioritize. Elegant, mathematically beautiful, and practically limited: it only works if the problem has a property called indexability.
Aditi is currently working under the supervision of Prof. Manu Kumar Gupta and collaborating with Nicolas Gast from Inria, France, focusing on sequential decision making in the field of restless bandits. An India AI Fellow, she is presently channeling all her time into learning about the index based heuristics used in solving the Restless Multi Armed Bandit Problem.
Where Mathematics Became Home
The appetite for problems like this began in Jhansi, at a state government college affiliated with Bundelkhand University. Mathematics had been Aditi’s favorite subject since third standard. “I never felt I was doing something tiring when I was solving mathematics,” she says simply.
Her physics teacher Nitin Gupta gave her a line she still carries close: “A day where you have not learned anything new is a day wasted,” an advice on deciding what kind of person to be.
Between her father’s wonder at distant starlight and her teacher’s philosophy about wasted days, Aditi built a foundation: some questions deserve a lifetime of approaching, and the approaching itself is the point.
The Conviction That AI Is Mathematics
IIT JAM brought her to IIT Mandi for applied mathematics. Deep learning courses, data science training, and self-taught reinforcement learning for her thesis, all pointed toward one conviction: “AI is not fundamentally a computer science problem. It is a mathematics problem. The deeper you go, the more mathematics matters.”
Near the end of her master’s, a LinkedIn post about a DST-Inria project at IIT Roorkee under Professor Manu Kumar Gupta caught her attention. Three months as a project researcher turned into full PhD admission at the Mehta Family School of Data Science and AI, a school built precisely for the intersection she had been trying to inhabit, where mathematicians and computer scientists work on AI together rather than in parallel.
The Poster That Changed Perspective
When Aditi presented at the National AI Summit in New Delhi, something shifted. “Reinforcement learning for restless bandits is not computer vision. It’s not the term people reach for when they think of AI,” she reflects. “But the crowd stopped at my poster. They asked questions.”
That was the first day she realized how impactful the field could be.
The experience also crystallized what she’d found in Professor Manu Kumar Gupta. People jokingly say it’s more important to choose a good supervisor than a life partner, she laughs. “He’s been really supportive, kind, and humble. The work and the person are not in competition. Having someone you can talk freely to becomes a bonus.”
It is the kind of environment that makes the hardest problems feel approachable rather than isolating.
Small Joys, Big Questions
Her first year brought small eureka moments every few weeks. The first joy was when she finalized the LP index mathematical model. Second, when she got good results comparing with existing policies. But the AI Summit stood out: “People applauding you, being interested in your work – that’s the biggest highlight.”
The hardest part? Time management without midterms or semester exams. Realizing there’s no one to put a track on you, she reflects that she sort of started liking it. “If there’s no check, no one to talk to about what we’re doing, at some point you’ll get demotivated. For myself, fascination is one thing, and a schedule is another.”
After her PhD, Aditi plans postdoc work wherever strong restless bandit research happens, then returning to academia with her dream application being healthcare. She feels she would be happy if she can apply her work to any healthcare problem in the next two or three years – providing medical aid, optimally utilizing medical resources. In a country where healthcare resources are perpetually stretched, where triage decisions happen constantly across overwhelmed systems, better frameworks for dynamic resource allocation could transform outcomes.
From the seven-year-old marveling at ancient starlight to the researcher working on problems that resist easy answers, Aditi’s path reflects particular ambition: not to solve everything, but to spend a lifetime approaching questions vast enough to matter.
Some questions are vast enough to spend a lifetime approaching. Some are practical enough that the approach itself helps millions. Aditi found both in the same problem.
Aditi Jha is a second-year PhD scholar at IIT Roorkee’s Mehta Family School of Data Science and AI, working on restless bandit frameworks for real-world decision-making under resource constraints.. As part of the first IndiaAI Fellow cohort, she collaborates with Inria Grenoble, France. Her work on LP indices for non-indexable restless bandits addresses real-world resource allocation problems in healthcare, infrastructure, and maintenance.