Dr. Oates’ recent work has explored deep neural networks for weakly supervised EEG denoising for brain-machine interfaces, human-in-the-loop deep reinforcement learning to train robots using video demonstrations, and learning declarative representations of the functionality of “found” hardware using black box methods. Ongoing efforts include novel methods for learning semantically rich compositional sentence embeddings, learning policies for monitoring and updating deployed deep learning models to maintain performance in the face of domain shifts, and unsupervised methods for learning grounded, relational policies for understanding and control of real and simulated environments.