Avatar

Shinjan Dutta

PhD in Electrical Engineering, Northwestern University

Evanston, IL

shinjandutta2021@u.northwestern.edu

(702)-505-1979

LinkedIn:// shinjan-dutta-1012

GitHub:// shinjandutta


Skills

Languages: Python, R, C++

Libraries: PyTorch, Tensorflow, Keras, Tflearn, NLTK, Numpy, OpenCV,

Databases: MySQL, Oracle11g, Postgresql, Firebase

Tools: Tableau, Trifacta, D3.js, Databricks, Matlab




I, Shinjan Dutta, am a PhD candidate in the Electrical Engineering program at Northwestern University. I am curious and driven to innovate in the field of Artificial Intellegince.

The coursework from my current and past programs like Advanced Computer Vision, Machine Learning, Neural Networks and Fuzzy Logic, Image Processing, Data Analytics, Statistical Pattern Recognition etc, helped in developing my expertise in areas of programming and will only strengthen it with the upcoming courses like, Deep Reinforcement Learning and Advanced Topics in Deep Learning.

My internships and projects helped me gain practical exposure to the current world problems that can be solved using machine learning. For my final year undergrad project, titled Audio Assistance for the Visually Impaired, I worked on making a tool to help the blind navigate and walk without difficulty. It acted as a guidance system with an audio output telling the blind about his surroundings using the input from the cameras attached to a cap. This project was based on object detection using the YOLO Algorithm. The distance calculation was done using monocular vision in order to give the visually impaired approximate distances of all the detected objects. For my advanced computer vision course project, I’m working on Image inpainting where I using the k nearest pixels to any pixel that needs to be predicted, using a LSTM network and then using position blending to preserve the intensities.

I was lucky to be a part of a start-up named Home Drone as an intern where I worked on making drones fully automated and smarter in order to develop a marketable product for specific applications. I was also part of a ML research internship under Professor J H Nirmal which involved emotion detection using facial images as well as speech.

Education

Northwestern University, Illinois, USA
2019 - Present

Master of Science in Electrical Engineering.

Manipal Institute of Technology, Karnataka, India
2015 - 2019

Bachelor of Technology in Mechanical Engineering

Work Experience

<
Graduate Research Assistant | Northwestern University, Evanston, IL, USA
January 2020 - Present

I am working at the Image and Video Processing Lab under the guidance of Professor Aggelos Katsaggelos on projects using AI in the field of medical imaging. I have worked on detetecting Cardiac Amyloidosis on nuclear imaging data of the heart, developing state of the art for Covid-19 detection on lung x-ray images using deep learning, worked on predicting areas of the brain associated with personality traits using Graph Convolutional Neural Networks on structural MRI data.


Research Intern | Indian Institute of Space, Science and Technology (IIST),Kerala, India
January 2019 - June, 2019

I worked on a team developing a humanoid robot to carry out to be used in test flights for the Human Spaceflight program by ISRO. This internship was my capstone project for the degree of Bachelor of Technology. I worked on designing the knee and ankle mechanism of the robot. This was my first exposure to artificial intelligence in practice as I was able to work with Computer Vision engineers and learn from them. I worked on object detection for the robot using YOLO.


Summer Research Intern | Indian Institute of Technology- Madras (IIT-M), Chennai, India
May 2018 - June 2018

Worked on mathematical modelling of the motion a three wheeled vehicle over a speed hump. The goal is to employ the use of advanced suspension systems and mitigate vibbrations. I used calculus, algebra and physics for the modelling and mitigation of vibrations


-->
-->
Research Intern | Ford Motor Company, Sanand Vehicle Assembly Plant, Gujarat, India
Nov 2017 - Jan 2018

Maintained and managed official websites of the organization. Added new webpages and additional features like webinars and live chat. Made the website secure and also easy to use and handle for visitors as well as company employees.


Projects

Predicting Brain Area associated with Personality Traits using Graph Neural Networks
June, 2020 - Present

I am working on using Structural MRI data of the brain to construct a graph of the brain with coordinates of different points of the brain as nodes and data on personality trait of the patient is used as a label. Using Graph Convolutional Neural Networks we can train the algorithm to identify which part of the brain is associated with that personlaity trait. We use Class Activation Maps to know which area of the brain the network is "looking" at to make the decision. This provides transparency and the neural network is not a black box. The ultimat aim is to be able to learn features of the brain and reconstruct them using Variational Auto-Encoders and help neuoscientists in early detetction of degenrative diseases.


DeepCovidXr: Detecting Covid-19 on Lung X-ray images
March 2020 - June,2020

I was part of a team of radiologists and students from Northwestern University that developed a deep learning model to detect Covid-19 on lung x-ray images using ensemble CNNs. We pretarained our model on 200,000 lung x-ray images from the NIH dataset and taught the model to differentiate between different lung diseases. Then we trained on our dataset of 15000 images from 7 different hospitals of Covid positive and negative images. We had a hold test set of 2000 images from another hospital. Our model achived 85% test accuracy and performed better than 3 radiologists from Northwestern Hospital. Our work is soon to be published in the journal Radiology.


Detecting Cardiac Amyloidosis in nuclear imaging data of the heart
Jan,2020 - March 2020

Cardiac Amyloidosis refers to a protein buildup in the heart which can prove to be fatal leading to cardiac arrest. Earlier, the only known way to diagnose Cardiac Amyloidosis was through a biposy. With the advent of Nuclear imaging, cardiologists can diagnose amyloidosis through 2D and 3D nuclear imaging data. Although, in almost 40% of the cases, the nuclear imaging data is not clear enough for cardiologists to diagnose the patient and hence a biopsy is needed. The aim of our project was to use deep learning to diagnose amyloidosis. We used both 2D and 3D data for the model to learn from since bothn modalities are used by cardiologists to diagnose a patient.


NLP tasks on Wikipedia movie data corpus | Course Project: Statistical Language Modeling
Jan - Feb 2020

NLP tasks like text generation, machine translation etc on Wikipedia movie data corpus. Implemented networks like Bengio model and LSTM.


Data Analytics on Chicago Police dataset | Course Project: Data Science Seminar
Sept - Dec 2019

Analysed the Chicago Police Dataset by the Invisible Institute to find out patterns in the misconducts by the police using queries, Machine Learning techniques and Natural Language Processing. The theme of the project was to analyse the trends before and after this data went public. Used tools like Tableau, Trifacta, Databricks, D3.js.


Audio Assistance for Blind | B.Tech Final Year Project
July 2018 - March 2019

Built a system which would provide walking assistance to the visually impaired. Used YOLO for Object Detection and Monocular Vision for Depth Estimation. The system consisted of a camera attached to the visually impaired's hat or walking stick which would provide assistance in form of audio. Used a deep learning approach to generate the second image of the pair of camera caliberated images need to estimate depth using binocular vision.


Snakes using Deep Reinforcement Learning
June 2019

Training an AI agent to play the game of Snakes using Deep Reinforcement Learning. Used Q learning for the agent and PyGame for the game.


Facial Expression Detection
Dec 2017 - June 2018

Classification of facial emotions using three techniques- CNN, Transfer Learning and Feature extraction using Haar Cascades. Achieved an accuracy of 96.39 on Haar cascades approach.


Face Recognition using One-Shot learning
April 2018 - May 2018

Face recognition using just one image of the subject. Trained a Siamese network using Triplet Loss. Got an accuracy of over 90% for the one shot and two shot approach and an accuracy of over 97 for five shot.


Poetry Generator
Feb 2018

Poetry generator using a Recurrent Neural Network.