Tuhin Das

I’m a Research Associate in the Human Technology Interaction (HTI) Lab at USC, building interpretable machine learning pipelines for multimodal clinical decision support. In collaboration with J&J MedTech (Ethicon), I designed supervised learning models to estimate DISH complication risk (Dehiscence, Infection, Seroma, Hematoma) and validated performance using model evaluation metrics.

I also focus on turning messy, heterogeneous data into usable signals—doing data preprocessing, feature engineering (Pandas, NumPy), and similarity analysis with unsupervised learning. I’ve built human-in-the-loop feedback systems to refine clinical models, with the goal of improving interpretability and workflow integration.

Before this, I worked as a Research Intern across robotics and computer vision: developing reinforcement learning (PPO) algorithms in TensorFlow for robotic arm control, and implementing Arduino motor control and Python drivers for real-time hardware interfacing. I’ve led research projects from super-resolution (SRCNN/SRGAN in PyTorch) to ML-assisted continuous authentication with PPG signals, and I’m listed as an author on “A2IPPG: A Machine Learning Assisted Real-time Standalone Continuous Authorization to Identification Using Photoplethysmogram Features” (IEEE Sensors Letters, 2025).