Shamya Karumbaiah

Relevant Ed+ML Projects

Emotion Detection Using Video Logs

The real time video stream from the tutoring system is used to detect a student's engagement levels. Patterns in facial action unit intensities is exploited to identify the causual emotional state in a student. The research studies the interactions between student emotion, engagement, learning and system design. This work was funded by US Department of Defense and is under NDA.

RL as a Service for Smarter Tutoring

Reinforcement Learning is used to pick the best next pedagogical action that is personalized to a student's state in a tutoring system. This is built as a service in a generalized library, available for consumption by an tutoring system. (SuperGLU). This work is a property of US Army Research Lab (ARL).
Git | Documentation (In Progress)

Emotion Detection Using Interaction Logs

Towards building a cost effective emotional state detector for tutoring systems, the zero dollar solution is to exploit the interaction logs of a student. The best classifiers performed 38% better than chance. Needs more data and feature engineering. No third party tool was used.
Git | Report | Data on request

Tutors worked with




Other ML Projects


Professor Gerald Downes at UMass biology department studies zebrafish locomotive behavior by making mutations to genes and then observing changes in how they swim away from a tactile stimulus. This needs tracking the angle between the body axis and the tail tip throughout a high-speed video. ZebraSwish automates this task by using deep learning techniques.

Task: Object Recognition [Computer Vision]
Models: CNN, Image Processing
Git | Report | Data on request

Deep Subreddit Simulator

Reddit is a social news aggregation website with user-created sections called subreddits that are themed around a particular topic. Deep Subreddit Simulator is a bot that trains deep neural network models on individual subreddits' posts, and then generates human-like new posts.

Task: Text Generation [Natural Language Processing]
Models: RNN, Markov Chains
Git | Report | Data on request