AI helps systems do more with less

Shubham Vaishnav recently finished his doctoral thesis at the Department of Computer and Systems Sciences (DSV). We had a brief chat to hear what he found, and what he’s up to next.

Hello Shubham, please tell us about your studies!

“Sure! My research is about enabling devices in distributed environments, like wireless networks and the Internet of Things, to take AI-driven smart decisions. These devices often have to juggle several conflicting goals at once. Imagine a smartwatch: it has to save battery, but it also wants to send data reliably to your phone and react quickly to its surroundings. The priorities amongst these goals keep changing, and so does the environment. Building on existing AI paradigms like Federated Learning and Reinforcement Learning, I designed methods that empower these devices to adapt their choices quickly, with less human supervision. What I learnt is that AI works much better on these devices when it’s designed from the start to be aware of its own limits: how much it can communicate, how much energy it has, and how its goals might shift over time,” explains Shubham Vaishnav.

What are the implications of your research?

“The kind of AI I work on runs quietly in the background of things that people already use: smart homes, fitness trackers, environmental sensors, and networks of small devices that learn together without sending all their raw data to one central server. My methods aim to help these systems do more with less: less battery use, less network traffic, less need for human supervision when conditions change. That matters because the number of connected devices in the world keeps growing, and energy and bandwidth aren’t free. The proposed AI-aided decision making is most helpful when the goals are well-defined, for example, when you can specify how much energy a device is allowed to use on average, or how much weight to give to one objective versus another. The limitation follows from the same point: AI can make a system more efficient once the objectives are clear – but it can’t decide the objectives themselves when they are unclear or involve value judgements. Those still need people.”

How did you become interested in this?

“Before my PhD, I was working as an iOS developer at Capgemini in India, building apps for clients like the Swedish company Husqvarna. It was good work, but I gradually realised that I was more drawn to research and teaching than to building products. That’s when I decided to switch my career to academia. I already had an appreciation for Swedish working culture, so when an opportunity opened at DSV, I was happy to apply, and lucky to get in. The specific topic, AI for wireless, connected devices, came naturally from my master’s work at IIT (ISM) Dhanbad in India, where my co-supervisor Praveen Kumar Donta first introduced me to the world of wireless networks. I found it fascinating that even tiny, resource-limited devices could be made to ‘learn’ if you design the right algorithms, and that’s the thread I’ve been pulling at ever since.”

What’s it been like to do a PhD at DSV?

“My PhD at DSV has been a rewarding journey. It was not a cakewalk, but the environment made it much smoother than I had expected. I remember one particular time when I was very stressed and went into a meeting with my supervisor, Sindri Magnússon, expecting a tough conversation. Instead, he calmly told me I had enough time and would make it through. That moment stayed with me and made me realise the value of a supportive supervisor. The same goes for my colleagues at DSV, who were competent and kind, and from whom I learnt a great deal.”

“Beyond research, one of the most fulfilling parts of these years was founding the Bhakti Yoga Society at Stockholm University, and seeing it grow into a thriving community. It gave me space to develop as a more wholesome person, with skills in management, organisation, and community building.”

What happens next for you?

“I plan to continue in academia. I have been offered a postdoctoral position at Uppsala University, which I plan to take up after my defence. The topic of the work, distributed learning and optimisation, builds on the fundamentals and expertise I gained during my PhD. Being based so close to Stockholm, I will be happy to stay in touch with my colleagues at DSV,” says Shubham Vaishnav.

Artikeln finns också i en svensk version

Text: Åse Karlén