Research

This page presents a selection of my published research. You can find a full record of my research publications at my Google Scholar here.

My thesis defense presentation provides an overview of my work and my later, yet unpublished, research. It can be found here.

Selected Works

Bayesian Optimized Monte Carlo Planning (2020)

John Mern, Anil Yildiz, Zachary Sunberg, Tapan Mukerji, Mykel J Kochenderfer
Published in the AAAI 2020 Proceedings. 
This work presents a method to select actions from a large set in Monte Carlo tree search. The proposed method uses a Gaussian process as a belief over the action-value space and Bayesian optimization to select the best action to explore each step. 

Exchangeable Input Representations for Reinforcement Learning

John Mern, Dorsa Sadigh, Mykel J. Kochenderfer
Published in the American Controls Conference 2020 Proceedings and in the AAMAS 2019 Proceedings as an extended abstract.
This work presents a neural network architecture that uses neural attention to better scale deep reinforcement learning to problems with many or varying numbers of objects. The result is a reduction in problem input size that is factorial in the number of objects considered.

Visual Depth Mapping from Monocular Images using Recurrent Convolutional Neural Networks

John Mern, Kyle Julian, Rachael Tompa, Mykel J. Kochenderfer
Published in the AIAA SciTech Forum 2018 Proceedings. 
This work presents a method to extract dense depth field maps from sequences of images from a monocular camera. The proposed method uses a deep convolutional neural network with training based on neural style transfer. The images were generated using Microsoft's AirSim suite.