Deep Neural Networks
I am mainly interested in understanding why neural networks generalize well, what kind of features they learn, how much data is required for generalization, as well as their applications such as object detection and image classification.
I want to understand what agents learn by reinforcement, and how to scale up learning to real environments such as complex robot systems.
A model needs to know what is does not know. I want to develop uncertainty quantification methods for neural networks that are usable in resource constrained platforms, and validate that they represent the appropriate features.
Robot vision works under the premise of a near real-time constraint with small training sets in dynamic environments. I am interested in learning with small datasets, object detection, pose estimation, and network architectures for resource constrained platforms.
Applications using multi-beam and forward-looking sonar imagery, specially related to detecting marine debris in real-world environments.