Brandenberger, Jill M.

DOE OSTI2024

Editorial: Innovations in climate resilience

At Battelle’s second Annual Innovations in Climate Resilience Conference, we were inspired this year by the phrase “Bold Leaps and Action.” Climate resilience is a goal and a field that requires boldness. Achieving a state in which societies, countries, continents, and even the globe is robust to changes in climate is often met with doubt, speculation, and indifference. Boldness is required to overcome the gap between an individual’s personal experience and the sheer scale of climate interactions that span from microbes to planets and from nanoseconds to millennia. Moreover, individual scientists, government leaders, and industrialists might each make a small impact and never directly see a measurable effect in Earth’s climate resilience. Is the scientific work in this area futile? Are we as a community on the right path or are we on the right track? Or are we collectively leveraging our potential and contributions toward scalable and more impactful climate resilience solutions that create compounding effects for cities, regions, countries? That leads us to the second part of the phrase that inspired us. It is not just bold leaps but it is action too. The scientific community must put into effect the discoveries that come along with our work in climate resilience. This includes the processes or activities that translate foundational science into real products that society can use. Never before in the history of this country have we had such a commitment to the Research Topic of climate resilience. The White House has made it a major part of their platform. Congress has appropriated and authorized billions of dollars in support. The part we need next is real action. Through our efforts we can discover the interconnected scientific breakthroughs at many spatial scales from city/regional/state to global that were not possible without those government programs.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

DOE OSTI2021

The Effect of Antagonistic Behavior in Reinforcement Learning

The significant achievements of deep reinforcement learning (RL) have motivated researchers to also investigate its shortcomings. Such work has shown that typical methods in deep RL tend to produce brittle policies that overfit to the training environment. In this paper, we introduce the notion of purely antagonistic behavior in value-based agents, where the objective is not to maximize reward but to minimize the victim’s value over time. This notion is motivated by the scenario in which an antagonistic human architect, without access to the environment’s reward function, wants to build an RL agent that can impede another well-trained RL victim agent. First, we formalize a notion of antagonistic behavior in RL. Then, we provide experiments that show how a purely antagonistic agent performs compared to a well-trained victim that learns directly from the game’s rewards. Our results suggest that if one’s goal is to find vulnerabilities in well-trained agents, direct access to the environment’s rewards is not necessary, and antagonistic behavior can be measured independently from environment wins and losses.

Fujimoto, Ted C.↗