Projects

Following is a list of projects I am currently working on, or have worked on in the past. Where possible, I have included links to the Github repository, published/unpublished papers and the project website.

The project website can be consulted to get a broad overview of the project whilst the papers can be read if one wishes for more depth on the matter. I usually host my projects on Github, the README can be consulted to get familiar with the structure of the repository.

[2021] aocp.el

elisp

An emacs package to manage bibliographic information in org-mode (a major mode for Emacs).

[Github]

[2020] KM3NeT

Python, Pytorch, Pytorch Geometric

Neutrinos are highly elusive subatomic particles which can only be detected with the help of large particle detectors. The KM3NeT neutrino telescope is one such detector currently being constructed at the bottom on the Mediterranean Sea. Due to its large volume and the presence of background noise, event trigger algorithms are utilized by the data acquisition pipeline of the detector to sift through the noise. A GPU Pipeline was also developed to improve the quality of filtration of the event trigger algorithms without compromising their runtime performance. Despite these efforts, the quality of filtration require further improvements. The goal of this paper is to improve upon the GPU Pipeline using Artificial Neural Networks. The paper explores the possibility of replacing parts of the GPU Pipeline using Multi Layer Perceptrons and Graph Convolutional Neural Networks. The Multi Layer Perception performs better compared to the existing solution while the results of the Graph Convolutional Network are inconclusive in its existing form. Overall, the outcome is promising and new avenues of research are discovered through this work.

[Github][Paper]

[2020] Privacy Preserving Deep Learning for Medical Imaging

Latex

The success of Deep Neural Networks with image classification prompted researchers to explore the applications of Deep Learning in Medical Imaging and Medical Image Analysis (MIA). Deep Neural Networks have sufficiently demonstrated their capabilities of performing MIA tasks tirelessly and with fewer errors as opposed to their human counterpart. However the challenge of training neural networks using sensitive medical data, without violating the privacy of patients remains an active field of research. Many solutions exist to address this concern, however a systematic review and analysis of these techniques is yet to be conducted. This paper attempts to conduct the first systematic review of privacy-preserving techniques to train deep learning models. Emphasis is especially put on the performance and privacy analysis of the techniques. In addition, the communication and runtime costs, the ability of the solutions to scale, tolerance to faults and the level of security against threats and attacks are also studied.

[Paper]

[2020] vim-zettel

Vimscript

A (neo)vim plugin to manage plain text notes, inspired by Niklas Luhmann’s Zettelkasten technique.

[Github]

[2020] vim-text-lists

Vimscript

A (neo)vim plugin that provides utility functions for working with plain text lists.

[Github]

[2019] ACE: Art, Color and Emotions

JavaScript, D3.js

Art has been the cornerstone of human expression and social progress through time. It is no wonder that art historians and daily enthusiasts alike have spent countless hours trying to understand more than what meets the eye. What an individual can draw from a painting is very subjective, but we now know that human mind has a high sensibility for well-defined subset of traits - one of them being colour. This paper describes the process of bringing a new light on how colours and emotional tone (sentiment) can be interlaced with one another. This is achieved with the use of widespread artificial intelligence techniques, a vast dataset of art meta-data(OmniArt) and state-of-the-art visual interaction tools.

[Paper][Publication]

[2018] 3D Kadaster

Python, Apache Spark, Three.js

We created a 3D model of all the buildings in The Netherlands using point cloud dataset.

[Paper][Demo]

[2018] Knowledge Acquisition from CommonCrawl

Python, NLTK, Stanford NLP

We applied a complete knowledge acquisition pipeline to WARC datasets using Natural Language Processing, Part of Speech tagging, Named Entity Recognition and Entity Linking. We also proposed a novel idea to improve entity retrieval using machine learning.

[Paper]

[2017] Elevate

JavaScript, Phaser.io

Specialized educational resources for individuals with Down Syndrome are lacking. This problem space was explored in detail through surveys and interviews with both the primary and secondary users. Several problems were discovered in this space of which, the lack of an affordable, easy to use and engaging cognitive test was deemed critical. This problem was further explored and an improved form of this assessment using web based games was proposed. The design process was broken down into three iterative phases. The first was defining the problem, followed by validating the solutions and finally iterating on the final solution. Throughout the phases, five main approaches were used to help with the analysis and iterative process. The approaches included user surveys, user Interviews, Wizard of Oz testing, usability testing, and user testing. Engagement, key usability issues and scoring correlation with standard methods were the primary testing protocols used for validation of the final designed solution. The results and limitations of the designed solution are touched upon and a few reasonable next steps are laid upon for the future.

[Paper][Github][Website]