I am an enthusiastic Machine Learning Researcher with a passion for Deep Learning and Natural Language Processing.
I love to dissect and break down complex problems into their fundamental elements, and solve them ground-up. Hailing from an Electronics background, I am a self-learnt coder and have acquired most of the knowledge through own interest and self-study.
My long term goal is to create a intelligent system which directs human beings and helps establish world order and peace, by providing perspective on solving problems with social consequences.
M.S in Computer Science
University of Massachusetts Amherst, USA
Courses: Machine Learning, Advanced Natural Language Processing, Intelligent Visual Computing, Systems for Data Science, Advanced Algorithms, Methods of Applied Statistics
B.E (Electronics and Communication Engineering)
SSN College of Engineering, (Autonomous, Affiliated to Anna University)
Courses: Machine Learning, Deep Learning, Computer Vision, Speech Technology, OOPS and Data Structures, Web Technology, Programming in Python, Linear Algebra, Probability and Stochastic Processes, Digital Signal Processing
2018 - 2022
May 2021 – July 2021
(Advisor: Prof. Boyu Wang, CS Dept)
MITACS Globalink Research Intern under Dr. Boyu Wang from the Department of Computer Science, Western University, Ontario, Canada.
Research focused on:
Researched Deep Learning models for Sensor-Based Human Activity Recognition (HAR) on accelerometer data of elderly in-patients.
Created a novel Luong Self-Attention + CNN model using TensorFlow and Keras to improve the accuracy on HAR by 8% in Pooling Task Learning and Meta Learning settings.
Research Assistant, Machine Learning
Solarillion Foundation, Chennai, India
Jan 2020 - Jun 2022
(Advisor: Mr. Vineeth Vijayaraghavan, Director)
Collaborated with the NLP-DL team that developed the Multi-Context Transformer for Sign Language Translation using PyTorch and FastText Embeddings. (Won 3rd Best Paper at MICAI 2021)
Built a custom, end-to-end Transformer-based model that learned from video segment representations, captured temporal dependencies between distant frames and eliminated the need for sign language intermediaries (glosses).
Achieved 98.19% ROUGE-L and 86.65% BLEU-4 score retention with a 30.88%
Research Assistant, Deep Learning
Jan 2020 - Jun 2022
(Advisor: Prof. Venkateswaran N, ECE Dept)
Researched and developed novel Deep Learning algorithms using PyTorch and OpenCV for two Satellite Image Fusion projects guided by the Indian Space Research Organization (ISRO).
Created a 2D CNN + Channel-attention model for Pansharpening. Improved run-time on large satellite images using Multi-Patch Attention, a novel method that computes self-attention on smaller, non-overlapping image tiles.
Adapted and modified the U2Fusion (unsupervised fusion) algorithm for SAR-Optical Image Fusion to produce Colorized-SAR Images for my undergraduate research project.
'Sign Language Translation Using Multi-Context Transformer'
Badri Narayanan M, Mahesh Bharadwaj K, Nithin G R, Dhiganth Rao Padamnoor, Vineeth Vijayaraghavan
Accepted and Presented at Mexican International Conference for Artificial Intelligence (MICAI) 2021.
'Sensor-Based Human Activity Recognition for Elderly In-Patients using Luong Self-Attention Networks'
Nithin G R, Mihika Chhabra, Yujiao Hao, Boyu Wang, Rong Zheng
Accepted at IEEE/ACM Conference on Connected Health: Applications, Systems, and Engineering Technologies (CHASE) 2021
To be published in the IEEE Proceedings.
3rd Best Paper Award for 'Sign Language Translation Using Multi-Context Transformer' at the Mexican International Conference for Artificial Intelligence (MICAI) 2021.
Won 2nd Place for 'for building a Restaurant Management System' at Vashisht IIITDM Kanchipuram Hackathon 2019, India.
Modelling and Predicting On-Time Performance of Flights:
Modelled and developed a Two-Stage Pipelined Machine Learning Engine to forecast the Delay Time of Flights. Experimented with linear, tree-based and ensemble classifier and regressor models. Performed dataset preprocessing and merging and oversampling to eliminate imbalance in the dataset. Performed regression testing along with general evaluation of the model with common metrics. (Tools used: Python, Sci-Kit Learn, Pandas and NumPy)
Clickbait Detection using Deep Learning:
Built a task-specific model to classify news headlines as clickbait or nonclickbait by adapting the Transformer encoder for classification task. (Tools used: Python and TensorFlow)
Wearable Device Design for Monitoring People with ASD (ongoing,IFP):
Conducting a study to measure physiological signals - Pulse Rate, Skin Temperature, Skin Conductivity and Heart Rate - of individuals having Autism Spectrum Disorder (ASD). Future directions involve performing analysis on cloud using Deep Learning on collected data and designing the resource constrained DL architecture for a smart wearable device.
Smart Road Traffic Clearance for Emergency Vehicles – Smart India Hackathon 2020:
Proposed an IoT framework for traffic clearance at road signals for ambulances and other emergency vehicles using Google Geolocation APIs, Computer Vision and Edge CNNs. (Tools used: NodeMCU, Arduino, ThinkSpeak and Python)
Restaurant Management System - Vashisht IIITDM Hackathon 2019:
Bagged 2nd place in the hackathon. Developed a prototype of a basic Restaurant Management System. (Tools used: Python, Tkinter GUI)
Teaching Assistant, Solarillion Foundation
Main responsibilities include designing assignments for orientees to the foundation, tracking and reviewing their progress in research and providing technical assistance when required. Other responsibilities include writing detailed weekly reports on research progress of orientees.
Volunteer - U&I
Teaching beginner-level English for underprivileged youth at learning centres