Master's research project:
As part of my Master's research I worked on Adversarial Machine Learning. More specifically, I am focused improving existing defenses against adversarial attacks on image classification tasks. Recently, I presented a poster at IEEE Security and Privacy conference 2018 titled "Enhancing Adversarial Example Defenses using Internal Layers". I am also working on using ensemble of models to improve adversarial example detection rate.
Undergraduate thesis research on Session Hijacking:
As part of my undergraduate thesis, I developed a system for protecting end users from SSL Stripping based Session Hijacking attacks. The system assigns a security rating for each website through classifying web pages using a Naive Bayes classifier. Then appropriate warning levels are shown for each website to end users. The ratings are updated based on user behavior towards these warnings, using statistical techniques. Code for the project was written in C and Python.
Cricket player rating system:
I developed an innovative system for rating cricket player performances using statistical analysis of historical cricket data (used C++, Python, SQLite). For this, I created a ball-by-ball database of cricket matches by scraping data from ESPNcricinfo website. Then I applied a dynamic programming model on the data to figure out expected runs and win probability for each situation in a game. Finally based on the expected runs model, I created individual player ratings from the ball-by-ball database. This rating system is more indicative of a player's true contribution, compared to existing approaches.
Sign of Life: A post-disaster rescue escalation system
Motivated by a human tragedy in a Bangladeshi garments factory, along with a few classmates, I developed a system for escalating post-disaster rescue operations. The project won the top award at Microsoft Imagine Cup 2015, Bangladesh regionals at World Citizenship category. We later published a research paper based on work on this project in IEEE NSysS 2015, titled "Sign of life: A system for escalating post-disaster rescue missions in ready-made garment factories".
DLcalc: Reverse engineered Duckworth Lewis software
Duckworth Lewis method is a proprietary system for deciding rain-affected cricket matches. The theory behind the system has been published in multiple research papers, but the parameters of the system are not made public. I managed to reverse engineer the system to find out the hidden parameters, and published an open source software, DLcalc, freely available to the public at http://dlcalc.codeplex.com. It is written in C\# and was later ported to Python.