Recent Publications

Research

Multitask Imitation Learning

Study of the evolution of imitation learning techniques in which an agent learns to perform a variety of different tasks, in contrast to traditional techniques which aim to learn only a single behavior.

PDF

Resolving Causal Confusion via Robust Imitation Learning

Algorithm to alleviate the problem of distributional shift in imitation learning by imposing constraints on non-causal variables during optimization, resulting in a robust imitation agent.

PDF Code

Quora Insincere Questions Classification

LSTM-type RNN to classify question text as either sincere or insincere. The text was collected from ~1.3M questions posted on the popular community-based, question-and-answer website, Quora.

PDF Code

Virtual Device Control Using Object Detection and Tracking

CNN (Single-shot Multi-Box Detector) and other skin segmentation techniques detect a user’s hand(s) in a live webcam video feed. Used hand location within video feed to enable real-time cursor/keyboard control.

PDF Code

Fault Detection in Aerospace Systems Using Self-Organizing Maps (SOM)

Method to quantify extent of damage and forecast maintenance events using anonymized sensor data from a fleet of 24 aircraft.

PDF

Contact