Machine Learning Data Science

SURYA DUTTA

919-***-**** ad5q6l@r.postjobfree.com linkedin.com/in/suryadutta2000/ Raleigh, NC (Open to Relocate) EDUCATION

NORTH CAROLINA STATE UNIVERSITY Raleigh, NC, USA

Master of Science in Electrical Engineering – Machine Learning & Data Science GPA: 3.77/4.0 August 2022 - May 2024 Relevant Coursework: Machine Learning, Computer Vision, Pattern Recognition in Machine Learning, Data Science, Accelerating Deep Learning, Software Engineering for Robotics, Digital Imaging Systems, Cloud Computing Technology PES UNIVERSITY Bangalore, India

Bachelor of Technology in Electronics & Communication Engineering GPA: 8.03/10.0 August 2018 - June 2022 TECHNICAL SKILLS

Languages: Python, MATLAB, C++, SQL.

Tools/Libraries: OpenCV, NumPy, Matplotlib, Pandas, Scikit-learn, PyTorch, TensorFlow, Keras, AWS, Seaborn. Platforms/Frameworks: IoT Connectivity Platforms, ROS, Gazebo, RViz, MoveIt, Docker, Kubernetes, Plotly. PROJECTS

DeepFake Detection using MesoNet Deep Learning, Image Analysis, TensorFlow, Keras

• Built a deep learning model using the MesoNet CNN architecture along with incorporating residual blocks and an attention mechanism to identify deepfake images of human faces accurately.

• Evaluated the model on a custom generated dataset to focus on the most important features of the images, utilized pruning with a sparsity of 20% to improve generalization, and achieved a detection rate of 97%. Combatting Class Imbalance in Classification Machine Learning, Data Science, PyTorch, MySQL

• Executed and compared techniques like Oversampling, SMOTE, Penalized SVM, and ensemble methods to handle class imbalance in datasets, leveraging MySQL for efficient data storage, management, and preprocessing.

• Applied F-Measure, G-Mean, & Area under ROC Curve to evaluate the performance of the techniques and obtained a mean accuracy of 85.54% (G-Mean) in SMOTE with under sampling across all the datasets. Perception-Aware Robot for Pick-and-Place Tasks ROS, Gazebo, RViz, MoveIt, ML, C++

• Simulated an autonomous pick-and-place robot in Gazebo using the UR5 Robotic Arm and a Kinect 3D Camera to perform the task of picking up objects and placing them at desired locations.

• Adopted the Moveit interface for motion planning & collision avoidance of the arm, fetched camera data, applied the YOLOv5 edge detection algorithm for extracting object positions & boundaries, interfaced object positions and then executed the pick-and-place and sorting tasks with an accuracy of 83.33%. Image Manipulation with CVAE & GANs Machine Learning, Data Analysis, Computer Vision, PyTorch

• Employed interpolation and latent space exploration through conditional GANs and conditional variational auto- encoders, synthesizing and classifying faces with varied expressions from unpaired images in the CelebA dataset.

• Trained the model for 200 epochs & boosted the visual quality of the generated faces with an accuracy of 92%. EXPERIENCE

Construction Automation & Robotics Lab, NC State University Raleigh, NC, USA Summer Intern May 2023 - July 2023

• Worked on 3D mesh reconstruction from video data, where I engineered a custom Vision Transformer (ViT) model to extract spatial features from 360 construction videos, enabling precise depth estimation and spatial analysis crucial for accurate model reconstruction.

• Collaborated on integrating ViT-extracted features with Visual SLAM systems (Stella VSLAM) to improve real-time spatial mapping accuracy and ensure precise alignment of the 3D models with the physical sites. Center for Signal Processing & Advanced Machine Learning, PES University Bangalore, India Research Intern January 2022 - May 2022

• Developed an ISP pipeline using a novel Channel-Prior based Retinex algorithm for underwater image enhancement, significantly reducing detail loss, and improving visual quality by roughly 40% compared to other standard methods.

• Published findings as a conference paper in the 2022 IEEE ICAECT conference, India. [PDF/HTML] Ericsson Bangalore, India

Summer Intern May 2021 - July 2021

• Implemented a convolutional encoder & Viterbi decoder in an AWGN communication channel for different constraint lengths and code rates in MATLAB, and optimized BER, SNR & other parameters, reducing the error rate by 20%.

• Designed an end-to-end comm. channel using the UDM portfolio and achieved faultless transmission and reception of data.

Contact this candidate