Malware Classifier
Research Project in Static Malware Detection (IIT Dharwad)
Motive: To analyze patterns that render a program as malware; to build a classifier that detects such patterns.
Explored feature extraction methods like EXE header info, control and data flow graphs; developed a pipeline involving logistic regression and naive Bayes classifier with well-known feature selection methods.
Achieved a
3% improvement
in accuracy compared to a
previously renowned
paper by modeling opcodes better.
Collected a dataset of 50k benign samples
by innovatively tackling the copyright hinderance; working towards publishing the novelty and the dataset
Soft skills learned: planning research, software documentation
Mar — June 2021; Dec — Apr 2022
Retro-game-playing Bots
Research Project in Reinforcement Learning (IIT Dharwad)
Extensively used PyTorch to develop RL bots for Tetris, Pong, Snake, and a few more Atari games that achieved admirable performance
Implemented various evolutionary and deep reinforcement algorithms like CEM, DQN, and A2C
Analysed state-of-the-art algorithms like AlphaGo, Rainbow DQN, and MuZero
(Open-source) Published all implementations and the report with a summary of each key family of algorithms to aid in quick bootstrapping of fellow RL followers
Soft skills learned: planning research
Code
Report
Demo
Aug 2020 — Dec 2020
Automated Essay Grader
Natural Language Processing
Analyzed ML and DL based approaches to NLP; algorithms like Naive Bayes, SVM and LSTM coupled with strategies like CBOW, Skip-gram, and TF-IDF
Analyzed and used BERT to obtain encodings that were fed to a shallow Neural Net; achieved a
Quadratic Weighted Kappa of 0.77
on the Kaggle AES Dataset
(highest in the competition being 0.81)
Mar 2020 — Jun 2020
Robust Facial Recognition for Real-world Applications
C-DAC Nvidia AI Hackathon (2019)
Innovated upon various DL pipeline architectures to overcome natural constraints like rotation, occlusion, and blurring while processing real-life surveillance data
Analyzed popular literature like YOLOv3, De-blur GAN, Facenet, and DREAM layer to understand their behaviour and ultimately integrate them together
Achieved a test accuracy of
73%
on the
Indian Movie Face Dataset
to finish in the
Top 30 teams in India
(sidenote: the model performed better on higher-res datasets)
Demo
Aug 2019 — Oct 2019
Other Notable Projects
Now We See You (Jan 2019)
Implemented an automated attendance tracking system based on landmark facial recognition using OpenCV Python and served it with a django backend
Our team finished 2nd in the hackathon and was selected to present the idea to Mr. N. R. Narayana Murthy (founder, Infosys)
Code
EasyPrint (Oct 2018 — Jan 2019)
Developed an online portal using django to facilitate on-the-move wireless printing in college. It was the only course project to be actually proposed to be put in production in college (out of 20)
Code