Nishanth Marer Prabhu

Master's

Computer Engineering

Embedded System Laboratory Northeastern University

Research Assistant Advisor

- Prof. Gunar Schirner - PyTorch to TensorRT model conversion for improved inference speed on Jetson Orin Edge Device - Applied mixed-precision techniques (FP32 and FP16) for sensitive models, significantly boosting speed and reducing model - Optimized CNN, CNN-LSTM, Transformer models and observed a speedup of 20x and throughput across varying batch sizes - Terrain Augmented Channel Model (TACM) for Automatic Modulation Recognition (AMR - Implemented a custom CNN model in PyTorch for processing frames in parallel, creating resolution awareness for the model - Trained multiple models on the novel TACM dataset, creating spatially aware signal recognition and classification models - Developed a specialized novel optimization algorithm for placing receivers in a varying terrain environment - Employed the usage of Linear Sum Assignment, MIP, and Quadratic Least Square Solvers to provide better accuracy - Conducted a detailed CUDA timing analysis, identifying inefficiencies due to multiple kernel launches and suboptimal operation - Automated model training/testing using SLURM scripts, streamlined log file loading, performance visualization with Python