In an era when the boundaries between security and machine learning grow ever more indistinct, one young researcher at the Indraprastha Institute of Information Technology has produced a body of work that demands the attention of the field.
Mr. Bikrant Bikram Pratap Maurya, a second-year M.Tech candidate, has this year seen four papers accepted at international venues of consequence — AsiaCCS, ICME, ICASSP, and AACL-IJCNLP — while submitting three further manuscripts to Interspeech and the ACM Conference on Computer and Communications Security.
His research spans four principal territories: the reconstruction of keystrokes from acoustic emanations propagated across heterogeneous hardware; the susceptibility of large-language-model agents to psychology-informed jailbreak strategies; the classification of malware via hyperbolic cross-attention in curved geometric space; and the detection of conversational deception through positive-unlabelled learning.
A unifying thread runs through each body of work. Rather than treating threat models as an afterthought, Mr. Maurya insists on designing the adversary before designing the model — an approach that has yielded results robust to conditions that defeat contemporary baselines.
His DECKER framework, accepted at AsiaCCS 2026, achieves zero-shot transfer across unseen keyboard hardware — a milestone that prior domain-adaptation methods could not reach. His hyperbolic fusion work, appearing at ICASSP and ICME, demonstrates that the geometry of representation space is not a mere convenience but a principled choice with measurable consequences.