User details
SD
SD
Subashree Dinesh
@subashree_dinesh_jx0lw13y3
Graduate Student
Deora, CO, United States
1Follower0Following0Mutual0Post
Member since: Jan 21, 2025• Led a visual perception study involving 2,500 U.S. participants, utilizing AWS and Amazon Mechanical Turk to assess cognitive responses to flood risk information from NOAA, FEMA, and Climate Central.
• Developed a graph-based database using Neo4j to model participant data, facilitating advanced analysis of cognitive responses through networked relationships based on psychological distance.
• Conducted in-depth statistical analysis using Python (Pandas, NumPy) and hypothesis testing methods (ANOVA, chi square, regression models) to uncover significant differences in risk perceptions across various demographic groups.
• Developed an end-to-end real-time Multi Object Tracking (MOT) framework leveraging YOLOv6 and FairMOT for user prompted target segmentation and tracking for surveillance applications, achieving a 78% MOTA score. . Utilized CUDA and TensorFlow with NVIDIA GPU to implement multi-GPU training, reducing training time by 75%.
• Designed dashboards in AWS SageMaker for efficient hyperparameter optimization, cutting model tuning time by 30%.
• Conducted quantization and pruning experiments, reducing model size and inference time by 60% on the GrAI chip.
• Led the seamless migration of YOLO-v6, MobileNet-v2 and FairMOT from PyTorch to TensorFlow and performed cross platform analysis on various CPU and GPU hardware.
• Built a DBSCAN-based clustering algorithm to group radar point cloud data, achieving real-time clustering of vehicles and pedestrians, with a 95% classification accuracy and a processing speed of 60 point-cloud frames per second (FPS).
• Engineered a high-throughput application to optimize point cloud data retrieval, boosting processing efficiency by 70%.
• Awarded one of only 5 coveted Research Internship positions nationally from a pool of 800 applicants. Performed research focused on detecting defects in additive manufacturing.
• Developed an object classification model using transfer learning to detect defects in 3D printed products.