arXiv subjects

Michael Dennis

Publications and source records attributed to Michael Dennis.

28 records · Page 2Linked to original sources

arXiv2022

Grounding Aleatoric Uncertainty for Unsupervised Environment Design

Adaptive curricula in reinforcement learning (RL) have proven effective for producing policies robust to discrepancies between the train and test environment. Recently, the Unsupervised Environment Design (UED) framework generalized RL curricula to generating sequences of entire environments, leading to new methods with robust minimax regret properties. Problematically, in partially-observable or stochastic settings, optimal policies may depend on the ground-truth distribution over aleatoric parameters of the environment in the intended deployment setting, while curriculum learning necessarily shifts the training distribution. We formalize this phenomenon as curriculum-induced covariate shift (CICS), and describe how its occurrence in aleatoric parameters can lead to suboptimal policies. Directly sampling these parameters from the ground-truth distribution avoids the issue, but thwarts curriculum learning. We propose SAMPLR, a minimax regret UED method that optimizes the ground-truth utility function, even when the underlying training data is biased due to CICS. We prove, and validate on challenging domains, that our approach preserves optimality under the ground-truth distribution, while promoting robustness across the full range of environment settings.

cs.LG↗

arXiv2022

Cooperative and uncooperative institution designs: Surprises and problems in open-source game theory

It is increasingly possible for real-world agents, such as software-based agents or human institutions, to view the internal programming of other such agents that they interact with. For instance, a company can read the bylaws of another company, or one software system can read the source code of another. Game-theoretic equilibria between the designers of such agents are called \emph{program equilibria}, and we call this area \emph{open-source game theory}. In this work we demonstrate a series of counterintuitive results on open-source games, which are independent of the programming language in which agents are written. We show that certain formal institution designs that one might expect to defect against each other will instead turn out to cooperate, or conversely, cooperate when one might expect them to defect. The results hold in a setting where each institution has full visibility into the other institution's true operating procedures. We also exhibit examples and ten open problems for better understanding these phenomena. We argue that contemporary game theory remains ill-equipped to study program equilibria, given that even the outcomes of single games in open-source settings remain counterintuitive and poorly understood. Nonetheless, some of these open-source agents exhibit desirable characteristics -- e.g., they can unexploitably create incentives for cooperation and legibility from other agents -- such that analyzing them could yield considerable benefits.

cs.GT↗

arXiv2022

Sociotechnical Specification for the Broader Impacts of Autonomous Vehicles

Autonomous Vehicles (AVs) will have a transformative impact on society. Beyond the local safety and efficiency of individual vehicles, these effects will also change how people interact with the entire transportation system. This will generate a diverse range of large and foreseeable effects on social outcomes, as well as how those outcomes are distributed. However, the ability to control both the individual behavior of AVs and the overall flow of traffic also provides new affordances that permit AVs to control these effects. This comprises a problem of sociotechnical specification: the need to distinguish which essential features of the transportation system are in or out of scope for AV development. We present this problem space in terms of technical, sociotechnical, and social problems, and illustrate examples of each for the transport system components of social mobility, public infrastructure, and environmental impacts. The resulting research methodology sketches a path for developers to incorporate and evaluate more transportation system features within AV system components over time.

cs.CY↗

arXiv2022

Replay-Guided Adversarial Environment Design

Deep reinforcement learning (RL) agents may successfully generalize to new settings if trained on an appropriately diverse set of environment and task configurations. Unsupervised Environment Design (UED) is a promising self-supervised RL paradigm, wherein the free parameters of an underspecified environment are automatically adapted during training to the agent's capabilities, leading to the emergence of diverse training environments. Here, we cast Prioritized Level Replay (PLR), an empirically successful but theoretically unmotivated method that selectively samples randomly-generated training levels, as UED. We argue that by curating completely random levels, PLR, too, can generate novel and complex levels for effective training. This insight reveals a natural class of UED methods we call Dual Curriculum Design (DCD). Crucially, DCD includes both PLR and a popular UED algorithm, PAIRED, as special cases and inherits similar theoretical guarantees. This connection allows us to develop novel theory for PLR, providing a version with a robustness guarantee at Nash equilibria. Furthermore, our theory suggests a highly counterintuitive improvement to PLR: by stopping the agent from updating its policy on uncurated levels (training on less data), we can improve the convergence to Nash equilibria. Indeed, our experiments confirm that our new method, PLR$^{\perp}$, obtains better results on a suite of out-of-distribution, zero-shot transfer tasks, in addition to demonstrating that PLR$^{\perp}$ improves the performance of PAIRED, from which it inherited its theoretical framework.

cs.LG↗

arXiv2021

Improving Social Welfare While Preserving Autonomy via a Pareto Mediator

Machine learning algorithms often make decisions on behalf of agents with varied and sometimes conflicting interests. In domains where agents can choose to take their own action or delegate their action to a central mediator, an open question is how mediators should take actions on behalf of delegating agents. The main existing approach uses delegating agents to punish non-delegating agents in an attempt to get all agents to delegate, which tends to be costly for all. We introduce a Pareto Mediator which aims to improve outcomes for delegating agents without making any of them worse off. Our experiments in random normal form games, a restaurant recommendation game, and a reinforcement learning sequential social dilemma show that the Pareto Mediator greatly increases social welfare. Also, even when the Pareto Mediator is based on an incorrect model of agent utility, performance gracefully degrades to the pre-intervention level, due to the individual autonomy preserved by the voluntary mediator.

cs.GT↗

arXiv2021

Accumulating Risk Capital Through Investing in Cooperation

Recent work on promoting cooperation in multi-agent learning has resulted in many methods which successfully promote cooperation at the cost of becoming more vulnerable to exploitation by malicious actors. We show that this is an unavoidable trade-off and propose an objective which balances these concerns, promoting both safety and long-term cooperation. Moreover, the trade-off between safety and cooperation is not severe, and you can receive exponentially large returns through cooperation from a small amount of risk. We study both an exact solution method and propose a method for training policies that targets this objective, Accumulating Risk Capital Through Investing in Cooperation (ARCTIC), and evaluate them in iterated Prisoner's Dilemma and Stag Hunt.

cs.MA↗

arXiv2021

Quantifying Differences in Reward Functions

For many tasks, the reward function is inaccessible to introspection or too complex to be specified procedurally, and must instead be learned from user data. Prior work has evaluated learned reward functions by evaluating policies optimized for the learned reward. However, this method cannot distinguish between the learned reward function failing to reflect user preferences and the policy optimization process failing to optimize the learned reward. Moreover, this method can only tell us about behavior in the evaluation environment, but the reward may incentivize very different behavior in even a slightly different deployment environment. To address these problems, we introduce the Equivalent-Policy Invariant Comparison (EPIC) distance to quantify the difference between two reward functions directly, without a policy optimization step. We prove EPIC is invariant on an equivalence class of reward functions that always induce the same optimal policy. Furthermore, we find EPIC can be efficiently approximated and is more robust than baselines to the choice of coverage distribution. Finally, we show that EPIC distance bounds the regret of optimal policies even under different transition dynamics, and we confirm empirically that it predicts policy training success. Our source code is available at https://github.com/HumanCompatibleAI/evaluating-rewards.

cs.LG↗

arXiv2021

Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Design

A wide range of reinforcement learning (RL) problems - including robustness, transfer learning, unsupervised RL, and emergent complexity - require specifying a distribution of tasks or environments in which a policy will be trained. However, creating a useful distribution of environments is error prone, and takes a significant amount of developer time and effort. We propose Unsupervised Environment Design (UED) as an alternative paradigm, where developers provide environments with unknown parameters, and these parameters are used to automatically produce a distribution over valid, solvable environments. Existing approaches to automatically generating environments suffer from common failure modes: domain randomization cannot generate structure or adapt the difficulty of the environment to the agent's learning progress, and minimax adversarial training leads to worst-case environments that are often unsolvable. To generate structured, solvable environments for our protagonist agent, we introduce a second, antagonist agent that is allied with the environment-generating adversary. The adversary is motivated to generate environments which maximize regret, defined as the difference between the protagonist and antagonist agent's return. We call our technique Protagonist Antagonist Induced Regret Environment Design (PAIRED). Our experiments demonstrate that PAIRED produces a natural curriculum of increasingly complex environments, and PAIRED agents achieve higher zero-shot transfer performance when tested in highly novel environments.

cs.LG↗

arXiv2021

Adversarial Policies: Attacking Deep Reinforcement Learning

Deep reinforcement learning (RL) policies are known to be vulnerable to adversarial perturbations to their observations, similar to adversarial examples for classifiers. However, an attacker is not usually able to directly modify another agent's observations. This might lead one to wonder: is it possible to attack an RL agent simply by choosing an adversarial policy acting in a multi-agent environment so as to create natural observations that are adversarial? We demonstrate the existence of adversarial policies in zero-sum games between simulated humanoid robots with proprioceptive observations, against state-of-the-art victims trained via self-play to be robust to opponents. The adversarial policies reliably win against the victims but generate seemingly random and uncoordinated behavior. We find that these policies are more successful in high-dimensional environments, and induce substantially different activations in the victim policy network than when the victim plays against a normal opponent. Videos are available at https://adversarialpolicies.github.io/.

cs.LG↗

arXiv2020

The Stretch Factor of Hexagon-Delaunay Triangulations

The problem of computing the exact stretch factor (i.e., the tight bound on the worst case stretch factor) of a Delaunay triangulation is one of the longstanding open problems in computational geometry. Over the years, a series of upper and lower bounds on the exact stretch factor have been obtained but the gap between them is still large. An alternative approach to solving the problem is to develop techniques for computing the exact stretch factor of ``easier'' types of Delaunay triangulations, in particular those defined using regular-polygons instead of a circle. Tight bounds exist for Delaunay triangulations defined using an equilateral triangle and a square. In this paper, we determine the exact stretch factor of Delaunay triangulations defined using a regular hexagon: It is 2. We think that the main contribution of this paper are the two techniques we have developed to compute tight upper bounds for the stretch factor of Hexagon-Delaunay triangulations.

cs.CG↗