In 2006, a mathematician named Cynthia Dwork proposed differential privacy, a mathematical method for protecting individual data now embedded in the systems of Amazon, Google, and Apple. She was working early, before the demand was obvious. Srishti Yadav thinks about her often, noting a similarity with herself. Her own aspiration is stated simply, without performance, “I want to propose something as foundational, as consequential, in a field that does not yet know it needs protecting,” she says.
Srishti is a fourth-year TCS Research Fellow and PhD scholar at IIT Roorkee’s Mehta Family School of Data Science and Artificial Intelligence, working under Professor Balasubramanian Raman and Professor Sanjeev Kumar on privacy preservation in spiking neural networks, AI systems considered more energy-efficient and closer to how the brain actually processes information than conventional deep learning.
From Etawah to the IIT Frontier
Srishti grew up in Etawah, a small district in Uttar Pradesh between Agra and Kanpur. She describes it as a conservative environment, and her parents as an exception within it, consistently supportive, never pushing her toward the paths that were easier to explain. Her mother, who has an MA in Sociology, taught Srishti at home through middle school. Even though mathematics ran through the household almost by coincidence, Srishti is the first PhD in her family.
She did her BSc in Mathematics with distinction at Dayalbagh Educational Institute in Agra, where she was selected for a summer training programme funded by the National Board for Higher Mathematics at IIT Guwahati. It was her first visit to an IIT, her first experience of serious mathematics culture beyond her own classroom. “At that time,” she says, “I had not imagined this far.”
After completing her MSc at Banaras Hindu University and during a year she spent preparing for GATE, she watched the job market shift toward AI and data science. Scrolling through a brochure, she came across the newly established Mehta Family School of Data Science & Artificial Intelligence at IIT Roorkee, accepting PhD students from science and mathematics backgrounds. She got in, as part of the first batch of five full-time PhD students the school admitted.
Asking the Unasked Questions
Spiking neural networks transfer information differently from conventional deep neural networks. Rather than passing data continuously between layers, they pass signals only when a predefined threshold is crossed, making them significantly more energy-efficient and, in their functioning, closer to how the brain actually works. They are still new and the research community still small. The question of whether these models can be manipulated to expose the private data they were trained on has generated years of painstaking work in conventional deep learning. In spiking networks though, it has barely been asked and Srishti is one among the few who are beginning to ask this question.
Srishti has implemented adversarial attacks to test the robustness of spiking networks against manipulation and proposed a model more resistant to those attacks than anything currently in the literature. She has also developed a privacy-preserving image generation model using differential privacy, the technique that Cynthia Dwork formalised nearly two decades ago.
Having published two papers at the International Joint Conference on Neural Networks (IJCNN) and International Conference on Pattern Recognition (ICPR), Srishti is now working on her third submission. The topic was suggested by her supervisors and shaped through early discussions with her seniors, but the direction since then has been her own.
The Fellowship That Bought Her Time
There was pressure for Srishti to complete her PhD as quickly as possible as her father was the sole earner in the family and she wanted to contribute. The family breathed a sigh of relief when she received the TCS Research Fellowship in her third year, changing the texture of the research work. “After I got TCS, the urgency lifted and I now have time to go deeper,” she reflects.
Without apology, Srishti describes herself as an introvert. What she is good at is focusing on her work, reading further into her subject, finding out what a simple thing contains when you look at it from varied angles. “I like reading and I guess the PhD culture suits my orientation,” she says. She did have a hard time initially as the pivot from pure mathematics to computer science felt like starting from zero. “In mathematics, we don’t study coding or deep neural networks. I had to start from scratch and learn the absolute basics.” Ultimately, it was the steady, regular support of her supervisors and colleagues that helped her navigate through this crossover.
Srishti likes to trek when she can, through the Himalayan Explorers Club that takes students from IIT Roorkee into the mountains every month. She reads fiction, and is hooked on Japanese fiction at the moment, “I am reading Sayaka Murata’s ‘Convenience Store Woman’ currently. My hobbies keep changing,” she shares with a smile.
Chasing consequential research
After the PhD, Srishti will apply for postdocs first, industry second, academia as a distant third. She prefers to keep moving, to keep finding new problems. She suspects that after five or six years in one field, the need for change is real.
She hopes that her current research work becomes something foundational just like Cynthia Dwork’s work on differential privacy that became a method rather than just a finding. “I don’t know if it will but that is still something that I hope,” she shares.
Srishti Yadav, a fourth-year TCS Research Fellow and PhD scholar at IIT Roorkee, works on privacy preservation in spiking neural networks, an emerging AI architecture inspired by the human brain. Combining a background in mathematics with AI research, she develops adversarial attack and differential privacy models for systems whose security implications remain largely unexplored.