Close

Dharma KC

Machine Learning

kcdharma@email.arizona.edu

Download Resume

About Me

Hi, I am currently a PhD student at the University of Arizona. My research is mainly on applied machine learning.

Publications

Raj, K. D., Chairat, A., Timtong, V., Dailey, M. N., & Ekpanyapong, M. (2018). Helmet violation processing using deep learning, In 2018 international workshop on advanced image technology (iwait). IEEE

Chairat, A., Dailey, M., Limsoonthrakul, S., Ekpanyapong, M., & KC, D. R. (2020). Low cost, high performance automatic motorcycle helmet violation detection, In The ieee winter conference on applications of computer vision

Experience

University of Arizona

Research Assistant

  • Continual learning of heart disease prediction from ECG signals using feedback from doctors
  • Active learning with privileged information
  • Building interpretable and adversarially robust models with privileged information

AIT AI Center

Research Engineer

  • Helmet violation processing: Increased helmet violation classification accuracy from 80% to 97% on 1000 test images
  • Automatic ticketing system: Ruby on Rails web application for traffic police officers to issue tickets based on automated violation analytics
  • License plate recognition: Increased motorcycle license plate segmentation and recognition accuracy from 50% to 91.8% on a dataset of 1000 blurry images
  • Deblurring: Developed a system for license plate image deblurring. Made it work on blurry license plates
  • Tree height estimation from monocular quadcopter video stream: Built custom quadcopter, implemented SLAM algorithms for autonomous navigation, and developed analytics to compute tree height from video
  • Human analytics: Worked on face detection, face recognition, age estimation, gender classification
  • Haptic system for teaching kids how to write. It is the proof of concept with a metal ball and android application with arduino and bluetooth module such that it will apply a break if ball goes out of a region
    demo

Education

University of Arizona

2019 -- 2024

PhD in Computer Science

GPA: 4.0
Related courses and projects:
Principles of machine learning, online learning and multi-armed bandits, machine learning theory, probabilistic graphical model, Multi label object classification with convolutional neural networks and graph convolutional neural networks, best arm identification with monte carlo tree search

Asian Institute of Technology

2015 -- 2017

M.Sc. in Computer Science

GPA: 3.67

Thesis:Helmet violation processing using deep learning

Related courses and projects:
Data structures and algorithms, web programming, computer graphics, computer vision, Big data visualization (Ruby on Rails), mesh based routing (Python), compiler (Java), car simulation (C++, OpenGL)

Tribhuvan University

2009 -- 2013

Bachelor in Electronics

GPA: 3.92
Related courses and projects:
C/C++ programming, linear algebra, calculus, probability, digital signal processing, Android app for best trekking route determination, tank war game in C++, RFID based student attendance system

Projects

Helmet violation processing using deep learning

This is a project I worked on when I was at AIT AI center. It includes detection and tracking of motorbikes without helmets, their license plate detection and using them for violation tickets.

View Project

Human Analytics

This is a project I worked on when I was at AIT AI center. It includes face detection, age, gender and emotion classification.

View Project

Monte carlo tree search using best arm identification

This is my open source implementation of the paper Monte carlo tree search with best arm identification Please note that this is not official implementation. This also includes my own algorithm which performed better than the paper in simpler cases [I have not tried complicated ones].

View Project

Multi label object classification with Graph convolutional neural networks

This project uses Yolov3 for object detection on the coco dataset and then refines the labels using graph convolutional neural network. I have my own way of generating similarity matrix. We can also extract feature vectors from CNN to train these GCNs.

View Project

Skills