Machine Learning Deep

AMAN KR. SHARMA

® linktr.ee/amankrsharma* R *************@*****.*** amansharma-oo7 Ó+1-240-***-**** College Park, MD EDUCATION

Masters in Robotics - University of Maryland, College Park, CGPA - 3.66/4 Aug 2022 - May 2024 Relevant Courses - ML for Cybersecurity using LLMs, Deep Learning, Soft Dev for Robotics, Computer Vision. Bachelor of Technology - Dehradun Institute of Technology Aug 2016 - May 2020 SKILLS AND INTERESTS

Languages and Methodologies Python, C++, Matlab, Agile, Test Driven Development. Libraries/ Frameworks Pytorch, Tensorflow, Jax, OpenCV, Numpy, Linux, Docker, ROS, Git, CI/CD, W&B. Relevant Courses Deep Learning Specialisation, Machine Learning, Introduction to Mlops. FULL TIME WORK EXPERIENCE

Independent Researcher, Efficient ML August 2023 - Present University of Maryland, College Park, MD, USA Supervisor - Prof. Abhinav Shrivastava

· Trained a hardware-friendly base Transformer Model for Multimodal Large-Language Models (LLMs) reducing cost to half.

· Employed efficient DL model Optimization techniques such as Block Sparsity and Token Pruning, saving runtime and cost.

· Trained a lottery ticket (sub-network) with a 2X speedup and a non-substantial accuracy drop, achieving practical speed-ups. Perception Research Engineer Dec 2020 - June 2022

Vision Labs, Autonomous Vehicle Project(ALIVE), IIIT Delhi Supervisor - Dr. Saket Anand

· Developed and Deployed a Multi-Obstacles Tracking system for the Perception Module using Camera and LIDAR.

· Achieved Real-time performance (30 Fps) and high mAP (41%) on an Nvidia AGX using TorchScript & TensorRT.

· Investigated model failure modes using interpretability techniques (GradCam/++) for robustness on rare classes.

· Designed the simulation bed for Evaluating the AV’s Software Stack and improving it using Domain Adaptation.

· Utilized self-training methods to transfer knowledge from a simulated dataset to real-world road conditions.

· Implemented data augmentation using Generative Models for diverse weather conditions saving resources for data collection and annotation.

· Led the team responsible for designing and developing the Human-Vehicle Interface (HVI) for the car using ROS. Artificial Intelligence Research Engineer Jan 2020 - Dec 2020 Secure Meters Pvt. Ltd. Supervisor - Mr. Ranjit Nair

· Developed and deployed a Safety (PPE) Kit Detection system for Onsite workers using Computer Vision based on DeTR

(Detectron) Network effectively reducing accidental damage.

· Created a Safety Detection and Hazard Warning System based on Virtual Geo-fencing for On-site workers in real-time on Jetson Nano using Yolov3 mitigating fatal risks at the factory’s floor.

· Spearheaded the development of a system for automated rejection of defective products for Quality Control using Anomaly Detection Methods (Isolation Forest) reducing the cycle time by 12% and saving material cost. Deep Learning/Machine Learning Engineer (Part-Time) May 2020 - Aug 2020 Human Machine Interface Labs Supervisor - Dr. Jainendra Shukla

· Prototyped a Drowsiness Detection & Warning System for Drivers at night using Near-Infrared (NIR) camera.

· Created a Custom light-weighted CNN network to accurately work on NIR images optimised for Jetson Nano.

· Led a team to develop a prototype android app (Demo app) for Driver’s Distraction prevention using MLkit. INTERNSHIPS & PERSONAL PROJECTS

Machine Learning/Computer Vision Intern May 2023 - Aug 2023 VechTech Inc., Baltimore, USA Supervisor - Dr. Adam Goodwin

· Boosted F1-score by 9.10% through a hybrid ensemble of Classical and Deep Learning techniques for Bio-Medical Data.

· Slashed image annotation time to 10%, thanks to an innovative semi-automatic segmentation labeling pipeline.

· Utilized interpretability methods to refine the accuracy of disease-causing mosquito classification. 3D Scene Reconstruction Sept 2022 - Present

Personal Project Supervisor - Prof. Jia-Bin Huang

· Used Linear Triangulation, PnP and Bundle Adjustment to reconstruct a scene captured from a camera in 3D.Link

· Quantifying ReprojectionError for different techniques such as Linear and non-linear Triangulation, PnP, Bundle-Adjustment.

· Implemented Neural Radiance Field (NeRF) from scratch to reconstruct a 3D Scene and generate novel scene views. Link Graduate Research Assistant, Computer Vision Oct 2022 - Mar 2023 Machine Vision Lab, University of Maryland, USA Supervisor - Dr. Yang Tao

· Benchmarked current SOTA transformer-based models (DeTR, MaskFormer, Mask2Former) for Instance Segmentation using Pytorch’s Detectron2 on small RGBD Dataset (Crab and Meat) with 80% AP (48% increment).

· Developed a Semi-Automated pipeline for live Image Segmenatation Annoatation using Eye Tracking on AR/VR.

Contact this candidate