Machine Learning / Artificial Intelligence

Sparse-view 3D Reconstruction | website
Jan 2022 – Present

  • Working on the reconstruction of novel views given a few (3-5) context views with approximate cameras.
  • Developed a Transformer-based model that outperforms PixelNeRF on generalization and compute overhead.

Pytorch Boilergithub

  • A simple user-friendly boilerplate code for training PyTorch models
  • Enables training classifier models in 40-lines of code (incl. imports and function calls!)

Multi-GAN Distillation | website | github
Jan 2022 – May 2022

  • Developed a framework for data-free knowledge transfer across StyleGANs to cater to single-source and multi-source knowledge distillation scenarios

Background Removal / Blurring | github

  • Implemented a simple background removal / blurring application using OpenCV with GrabCut algorithm

Real-Time Face Spoof Detection using a Single RGB Camera | github | report
Aug 2018 – Dec 2018 | Advisor: Prof. Santanu Chaudhury

  • Implemented a stable face tracking using Haar Cascades & ROI thresholding
  • Trained MobileNet on self-collected face spoof data which achieved less than 6% false acceptance rate on imposter detection in NUAA dataset
  • Extracted motion-robust remote PPG signals from single RGB webcam for living skin detection
  • Recreated RGB images from Hyperspectral images with applications to PPG signal generation

Ethnicity Detection using Deep CNNs | github | report
Dec 2018 | Advisor: Prof Navneet Goyal

  • Trained several CNN models for detecting ethnicity of a person
  • Compared the effects of size and shape of the network on training and inference
  • Obtained 80% accuracy of prediction on a subset of UTKFace dataset

Disaster Related Tweet Detection using Machine Learning | github | report
Nov 2018 | Advisor: Dr. Poonam Goyal

  • Applied logistic regression to classify tweets as disaster-related or spam
  • Developed a matching-based algorithm inspired by Hien To et al. to retrieve disaster-related tweets
  • Demonstrated the effectiveness of learning-based method when used with effective feature extraction techniques and compared it with the matching method

Predicting Final Grades using Machine Learning | github | report
Nov 2018 | Advisor: Prof Navneet Goyal

  • Studied the effectiveness of various machine learning models in predicting final grades based on a continuous assessment of undergraduate student performance in a course.

Investigating the Curse of Dimensionality | github | paper
Sep 2018 | Advisor: Prof Navneet Goyal

  • Established the effects of high dimensional spaces through various experiments
  • Presented some simple solutions to the curse of dimensionality in various use cases

Text Guided Attention Model for Image Captioning | Presentation
Apr 2018 | Advisor: Prof. Surekha Bhanot

  • Implemented the paper by Mun et. al. as a proof of concept
  • Used CIDEr score to extract guidance captions to train the Attention Model on the MS-COCO 2014 Dataset
  • Built the decoder LSTM model to generate captions

Connect – 3 Game Playing Bot | project page | github
Sep 2017 – Dec 2017 | Advisor: Dr. Vandana Agarwal

  • Implemented a 2 player Connect-3 game using minimax algorithm
  • Obtained a 14 times speedup using alpha-beta pruning during the search

Probabilistic Inference using Bayesian Networks | project page | github
Sep 2017 – Dec 2017 | Advisor: Dr. Vandana Agarwal

  • Modeled a real world scenario as a bayesian network to answer queries regarding various events (health problems, accidents etc.) and their factors (air pollution, bad road conditions etc.)
  • Used markov blanket of each Random Variable to reduce the number of computations needed to answer the probability query

Improving Sales of a Late Night Restaurant using Data Mining Techniques | github
Oct 2017 – Nov 2017 | Advisor: Dr. Poonam Goyal

  • Used Association Rule Mining to find patterns in the sales of the All Night Canteen of BITS Pilani
  • Proposed a dynamic pricing scheme to adjust the price of an item based on the hour of the day and the customer type
  • Obtained a 5% rise in profits while keeping a low penalty
  • Suggested combo offers of low rated items with higher rated items to increase the sales of the low rated items

Jarvis Patrick Clustering in Racket | github
Nov 2017 | Advisor: Dr. Lavika Goel

  • Explored functional programming in Racket by implementing Jarvis Patrick Clustering algorithm

Visualizing Perceptron Learning | project page | github
Oct 2017

  • Implemented a visualization tool for visualizing the learning process of a perceptron in a 2 dimensional input space
  • Used Python’s TKinter graphics for displaying the curve learnt along with positively and negatively classified samples
Classical Computer Science

C-Template Library | github

  • Developed a library for generic programming in C using void* pointers, to implement standard data structures and algorithms in C

Compiler | github
Feb 2018 – Apr 2018 | Advisor: Prof. Vandana Agarwal

  • Built a partial compiler for a C-like language
  • Wrote a DFA based lexer to tokenize source file with error reporting
  • Wrote LL(1) compatible grammar for the language
  • Built the parse table for LL(1) parsing, along with syntactic checks (and appropriate error reporting)
  • Wrote semantic rules to create abstract syntax tree
  • Working Abstract Syntax Tree Construction

Logic Programming | github
Nov 2016 | Advisor: Prof. Sundar S. Balasubramaniam

  • Used prolog as a predicate logic system to implement a mathematical expression reducer
  • Implemented a Knowledge-base query-answering system, to answer questions about the Academic Regulations of BITS Pilani

TINY: An inline Turing Machine Simulator for C++ Language | report | github

  • Implemented an inline Turing Machine – a simple c/c++ library that has the functions to simulate a turing machine inline with c/c++ code
Other Projects

The following are some pet projects I am doing / have done in the past. The list is not exhaustive. Please visit my github profile for latest updates.

  • Simple-DB : A simple JSON-like data storage, retrival and presentation API
  • PACRYPT : A PolyAlphabet enCRYPTion program
  • ds : Scalable implementations of common data structures
  • spin-ui : A light-weight customizable browser based text editor
  • hello : A clean about-me website template
High School Projects

The following few apps are among the many that I made as fun projects in my high school. They are written in Visual Basic 6. More on github.

  • Writer MDI : A light-weight Multi Document Interface, word processor
  • Media Studio : A wrapper for the Windows Media Player control
  • Surf : Used the internet explorer control to build a web browser wrapper

Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s