ISMRM 2026AdaptCMR is a parameter-efficient, detail-preserving deep learning framework for cardiac MRI reconstruction. It uses a spectrally guided mixture of experts to generalize across different views, contrasts, and acceleration factors, preserving fine anatomical detail with fewer parameters and faster inference.
IEEE ICCKE 2024Integrated FinBERT, a financial sentiment analysis LLM, to extract sentiment from news and combined it with stock prices in a PPO-LSTM model for stock trading. Achieved 134.39% cumulative return (2.34× initial budget) and a 1.46 Sharpe ratio, significantly outperforming DJIA, Ensemble, and PPO models.
Under Review — Journal of Ambient Intelligence and Humanized ComputingApplied SincNet layers to classify Parkinson’s Disease from gait data, achieving 98.7% accuracy. Pruned filters after initial training using K-means clustering and silhouette scores to identify key frequency bands that explain the model's decisions.
RoboCup Proc.