About me

Hi! I am a PhD student at Georgia Institute of Technology in the School of Computational Science and Engineering where I am advised by Polo Chau.

My research focuses on adversarial machine learning, and interpretability and explainability of deep convolutional neural networks. I am currently investigating strategies to secure deep architectures from causative and exploratory attacks.

I am also an active member of the Polo Club of Data Science where we conduct research that spans across the domains of data mining, machine learning, human-computer interaction and information visualization.

Publications

Keeping the Bad Guys Out: Protecting and Vaccinating Deep Learning with JPEG Compression. Nilaksh Das, Madhuri Shanbhogue, Shang-Tse Chen, Fred Hohman, Li Chen, Michael E. Kounavis, Duen Horng Chau. arXiv:1705.02900 [cs.CV]. [arXiv]

PASSAGE: A Travel Safety Assistant with Safe Path Recommendations for Pedestrians. Matthew Garvey, Nilaksh Das, Jiaxing Su, Meghna Natraj, Bhanu Verma. Companion Publication of the 21st International Conference on Intelligent User Interfaces. ACM, 2016. [ACM]

Projects

Classification of satellite imagery to determine land cover type is a challenging task, primarily owing to the high intra-class variability in the classes of land cover under consideration. In this paper, we explore different pre-trained deep neural networks for this purpose. By fine-tuning and adapting the models to train on a labeled satellite imagery dataset, we find that these redesigned deep architectures outperform other proposed methods which involve complex representations of the imagery, simply by training on raw images without any kind of elaborate trans- formations. We finally determine the best model thus obtained, and use it to then quantify the urban tree canopy coverage in the city of Atlanta. This was my capstone project for the CSE 8803: Computational Sustainability class at Georgia Tech in Spring '16.