May 2021 – Jul 2021 · 2 mos · Hyderabad, Telangana, India

• As part of AdQuality team, worked on detection of fraud domains using WHOIS data and C2C logs with different feature sets and models to capture the common characteristics/properties of fraud domains.
• After training multiple LightGBM models, proposed and implemented a novel unifying Graph Neural Network model incorporating both types of features (fraud rates and 0-1 features), with domains represented as nodes, common characteristics explicitly represented through edges, and using specially constructed node and edge feature vectors.
• Extensively worked with these frameworks, libraries and software: Python, Pandas, NumPy, SciPy, matplotlib, Scikit-learn, LightGBM, XGBoost, PyTorch, PyTorch Geometric, DGL, SCOPE

May 2021 – Jul 2021 · 2 mos

• Prof. Debmalya Panigrahi
• Designing online and dynamic algorithms with low recourse for set cover/hitting set problem when VC-dimension is small.

Apr 2020 – Oct 2020 · 6 mos · Bengaluru, Karnataka

• Prof. Arindam Khan
• Improved Approximation Algorithms for Weighted Edge Coloring of Graphs
• https://arxiv.org/abs/2012.15056

Apr 2020 – Jul 2020 · 3 mos · Singapore

• Supervised by Prof. Thambipillai Srikanthan.
• Research focused on finding ways to detect malware/suspicious codes (vulnerabilities listed in CWE-Common Weakness Enumeration) in a binary package. Binary packages use a lot of open source code and it becomes easy for an adversary to exploit the known vulnerabilities in these packages.
• The main aim of the project is to automate the above using only the source code of different libraries (listed in CWE) as they are mostly easily available.