James E. Kostas

College of Information and Computer Sciences, University of Massachusetts Amherst

jekostas [at] cs [dot] umass [dot] edu

I am a late-stage PhD student working with Dr. Philip Thomas in the Autonomous Learning Lab (ALL). My research interests lie at the intersection of reinforcement learning (RL), stochastic neural networks (coagent networks), deep learning, and robotics. I am particularly interested in "sim-to-real" transfer learning for RL with robotics applications. I am also interested in the intersection of RL and evolutionary algorithms/AutoML.

Publications

Highly competitive peer-reviewed publications:

• Austin Hoag, James Kostas, Bruno Castro da Silva, Philip Thomas, Yuriy Brun. Seldonian Toolkit: Building Software with Safe and Fair Machine Learning. In Proceedings of the 2023 International Conference on Software Engineering (ICSE 2023).
• Dhawal Gupta, Gabor Mihucz, Matthew Schlegel, James Kostas, Philip Thomas, Martha White. Structural Credit Assignment in Neural Networks using Reinforcement Learning. In Advances in Neural Information Processing Systems (NeurIPS 2021).
• James Kostas, Yash Chandak, Scott Jordan, Georgios Theocharous, Philip Thomas. High Confidence Generalization for Reinforcement Learning. In Proceedings of the 38th International Conference on Machine Learning (ICML 2021).
• James Kostas, Chris Nota, Philip Thomas. Asynchronous Coagent Networks. In Proceedings of the 37th International Conference on Machine Learning (ICML 2020).
• Yash Chandak, Georgios Theocharous, James Kostas, Scott Jordan, Philip Thomas. Learning Action Representations for Reinforcement Learning. In Proceedings of the 36th International Conference on Machine Learning (ICML 2019).

Other peer-reviewed or non-peer-reviewed publications:

• James Kostas, Cedric Zhao, Skanda Vasudevan, Tian Yu, Mike Berry. Estimating Displacement Costs for Individual Ads: An R-learner Approach. In the Amazon Machine Learning Conference.
• James Kostas, Scott Jordan, Yash Chandak, Georgios Theocharous, Dhawal Gupta, Philip Thomas. A Generalized Learning Rule for Asynchronous Coagent Networks. In the 5th Multidisciplinary Conference on Reinforcement Learning and Decision Making (RLDM 2022).
• Philip Thomas, Scott Jordan, Yash Chandak, Chris Nota, James Kostas. Classical Policy Gradient: Preserving Bellman’s Principle of Optimality. arXiv:1906.03063 (arXiv 2019).
• Yash Chandak, Georgios Theocharous, James Kostas, Scott Jordan, Philip Thomas. Improving Generalization over Large Action Sets. In the 4th Multidisciplinary Conference on Reinforcement Learning and Decision Making (RLDM 2019).
• Yash Chandak, Georgios Theocharous, James Kostas, Philip Thomas. Reinforcement Learning with a Dynamic Action Set. Continual Learning Workshop at the Thirty-second Conference on Neural Information Processing Systems (NeurIPS 2018).