When you "measure data", you quantify its characteristics to support dataset comparison & curation.
You also begin to know what systems will learn. Many ML systems don't reason with this, we posit you should.
Choices, Risks, and Reward Reports: Charting Public Policy for Reinforcement Learning Systems
Feb 8, 2022
Thomas Krendl Gilbert, Sarah Dean, Tom Zick, Nathan Lambert
Center for Long-Term Cybersecurity White paper Series
We detail why reinforcement learning systems pose a different type of (dynamic) risks to society. This paper outlines the different types of feedback present in RL systems, the risks they pose, and a path forward for policymakers.
In this paper we set out to understand the causes of compounding prediction errors in one-step learned models. With this, we hope a next generation of models can be used to improve model-based reinforcement learning.
AI Development for the Public Interest: From Abstraction Traps to Sociotechnical Risks
Feb 4, 2021
McKane Andrus, Sarah Dean, Thomas Krendl Gilbert, Nathan Lambert, Tom Zick
IEEE International Symposium on Technology and Society
We study three developing subfields of AI research and their growing relationship with the sociotechnical: AI Safety, Fair Machine Learning, and Human-in-the-loop Autonomy.
2020 Conference on Learning for Decision and Control
Studying the numerical effects of a dual-optimization problem in model-based reinforcement learning -- control and dynamics. When optimizing model accuracy, there is no guarantee on improving task performance!
2020 IEEE International Conference on Robotics and Automation (ICRA)
Learning how to walk with a real-world hexapod using a hierarchy of model-free RL for basic motion primitives with model-based RL for higher level planning.
A mixed talk discussing the research challenges of controlling microrobots and how model-learning can be used to synthesize highly specific controllers.