PFedMA
AcceptedPruned Federated Learning with Matched Averaging — extending neuron matching to more complex architectures
NeurIPS 2026 · Workshop
2026
Independent research contractor at Perssonify, where I build stock and social media correlation analysis used in large class action lawsuits. Alongside that — paper-driven, experiment-heavy work across federated learning, computer vision, interpretability, and AI safety, with publications at AAAI 2026 and research programs at IIT Gandhinagar and IIT Kharagpur.
Independent Research Contractor
Present
Remote · Bangalore
Training across decentralized clients without centralizing sensitive data — neuron matching, matched averaging, and multimodal domain generalization.
Camouflaged and salient object detection, image artifact correction, and frequency-domain architectures.
Understanding what models learn and why — topological methods and analysis of adversarial behavior.
Adversarial robustness, purification defenses, and reliable behavior under distribution shift.
Laplacian-Based Frequency Injection Network for Camouflage Object Detection (Student Abstract)
AAAI 2026 · Student Abstract
2026
Mitigating Adversarial Attacks Using Topological Purification in OCR Models
AAAI 2026 · AICS Workshop
2026
Adaptive Phishing Detection System for Blockchains Using Machine Learning
Springer LNNS · ISBM 2024
2024
Prediction of high-flow events. Two co-authored papers from the SRIP program at IIT Gandhinagar.
Water Resources Research
2025
Pruned Federated Learning with Matched Averaging — extending neuron matching to more complex architectures
NeurIPS 2026 · Workshop
2026
Evaluating whether agreement with human sentiment labels predicts market signal in financial social media
EMNLP 2026 · FinNLP Workshop
2026
Linked entries open the published paper
Undergraduate Researcher
Camouflaged object detection, adversarial robustness, and federated learning.
SRIP Research Intern
Selected among 1800+ applicants — prediction of high-flow hydrological events.
Research Intern
Multimodal domain generalization for federated models.
Research Head
Leads a 10-member student research team publishing in top-tier venues.