I want to contribute to a future where AI systems are safe, understandable, and reliable. To pursue this mission, I focus on fundamental questions with the goal of identifying foundational concepts and laws that will underpin the science of AI. My primary interests are reinforcement learning, goal specification, methods for achieving generality in AI, and computatonal manifestations of the “4Es” (extension, embodiment, enaction, embedment).

Publications

* denotes joint first authorship.

Meta-Gradient Search Control: A Method for Improving the Efficiency of Dyna-style Planning,
Brad Burega*, John D. Martin*, Luke Kapeluck, Michael Bowling,
ArXiv (2024) [pdf]

Settling the Reward Hypothesis,
Michael Bowling*, John D. Martin*, David Abel, Will Dabney,
ICML (2023) (Oral) [pdf]

MOTO: Offline to Online Fine-tuning for Model-Based Reinforcement Learning,
Rafael Rafailov, Kyle Beltran Hatch, Victor Kolev, John D Martin, Mariano Phielipp, Chelsea Finn,
ICLR Reincarnating Reinforcement Learning Workshop (2023) [pdf]

Model-based Adversarial Imitation Learning as Online Fine-tuning,
Rafael Rafailov, Victor Kolev, Kyle Beltran Hatch, John D Martin, Mariano Phielipp, Jiajun Wu, Chelsea Finn,
ICLR Reincarnating Reinforcement Learning Workshop (2023) [pdf]

Learning to Prioritize Planning Updates in Model-based Reinforcement Learning,
Brad Burega, John D. Martin, Michael Bowling,
NeurIPS Workshop on Meta Learning (2022) [pdf]

Time to Take Embodiment Seriously,
John D. Martin,
The Multi-disciplinary Conference on Reinforcement Learning and Decision Making (2022) [pdf]

Should Models Be Accurate?,
Esra’a Saleh, John D. Martin, Anna Koop, Arash Pourzarabi, Michael Bowling,
The Multi-disciplinary Conference on Reinforcement Learning and Decision Making (2022) [pdf]

The Stochastic Road Network Environment for Robust Reinforcement Learning,
John D. Martin, Paul Szenher, Xi Lin, Brendan Englot,
ICRA Workshop on Releasing Robots into the Wild (2022) [pdf]

Adapting the Function Approximation Architecture in Online Reinforcement Learning,
John D. Martin*, Joseph Modayil*,
ArXiv (2021) [pdf] [code]

Reinforcement Learning Algorithms for Representing and Managing Uncertainty in Robotics,
John D. Martin,
Ph.D. Thesis, Stevens Institute of Technology, (2021) [pdf]

Stochastically Dominant Distributional Reinforcement Learning,
John D. Martin, Michal Lyskawinski, Xiaohu Li, Brendan Englot,
37th International Conference on Machine Learning (2020) [pdf]

Variational Filtering with Copula Models for SLAM,
John D. Martin*, Kevin Doherty*, Caralyn Cyr, Brendan Englot, John Leonard
International Conference on Intelligent Robots and Systems (IROS) (2020) [pdf]

Autonomous Exploration Under Uncertainty via Deep Reinforcement Learning on Graphs,
Fanfei Chen, John D. Martin, Yewei Huang, Jinkun Wang, Brendan Englot,
International Conference on Intelligent Robots and Systems (IROS) (2020) [pdf]

Fusing Concurrent Orthogonal Wide-aperture Sonar Images for Dense Underwater 3D Reconstruction,
John McConnell, John D. Martin, Brendan Englot,
International Conference on Intelligent Robots and Systems (IROS) (2020) [pdf]

On Catastrophic Interference in Atari 2600 Games,
William Fedus and Dibya Ghosh, John D. Martin, Marc G. Bellemare, Yoshua Bengio, Hugo Larochelle,
ArXiv (2020) [pdf], [code]

Sparse Gaussian Process Temporal Difference Learning for Marine Robot Navigation,
John D. Martin, Jinkun Wang, Brendan Englot,
Conference on Robot Learning (2018) [pdf]

Distributed Gaussian Process Temporal Differences for Actor-critic Learning,
John D. Martin, Zheng Xing, Zhiyuan Yao, Ionut Florescu, Brendan Englot,
New York Academy of Sciences Machine Learning Symposium (2018)

Extending Model-based Policy Gradients for Robots in Heteroscedastic Environments,
John D. Martin, Brendan Englot,
Conference on Robot Learning (2017) [pdf]