Worked with Amazon Active Defense security telemetry data, collected through the internal service called Chronicle, to to detect when a host in a homogenous class of hosts is suddenly exhibiting anomalous behavior.
Used PySpark to parse though daily AWS Chronicle data with 12-14 billion events per day, and benchmarked unsupervised machine learning algorithms to discover which worked best with the dataset’s string based, command line argument features.
Mapped out daily/weekly host execution behavior across AWS accounts.
Discovered significant seasonality within AWS execution events and mapped out host execution behavior drift, leading to the discovery of potential malicious actors.
Used OpenCv and deep learning to teach object recognition, primarily humans
Utilized a raspberry pi and common USB camera as a platform for the recognition program
Future works will include building a behavior tree, inspired by the AI from the video game Alien Isolation, to have a mobile robot seek and follow humans
Currently in the process of building the robot and developing the AI
Used Time series analysis and regression tactics to model east coast energy use over the last decade
Plotted seasonality of energy use to find biweekly regional consumption spikes
Trained models using black friday consumer purchase data from a retail chain to predict future consumer purchasing based on consumer traits
Utilized Random Forest and the CART algorithm
Trained regression models to predict park usage using twitter and social media data from the Twin Cities Metropolitan Area in Minnesota. Used ANOVA to select for important features among parks and local neighborhoods.