Stackelberg security games have been widely deployed to protect real-word assets. The main solution concept there is the Strong Equilibrium (SSE), which optimizes defender's random allocation of limited resources. However, solely deploying SSE mixed strategy has limitations. In extreme case, are where defender able defend all assets ``almost perfectly" at SSE, but she still sustains significant loss. this paper, we propose an approach for improving utility in such scenarios. Perhaps...

In this paper we present a novel Stackelberg-type model of security domains: Security Assets aSsignment with Information disclosure (SASI). The combines both the features Stackelberg Games (SSGs) and Bayesian Persuasion (BP) model. More specifically, SASI includes: a) an uncontrolled, exogenous state that serves as Defender's private information; b) multiple assets non-accumulating, targetlocal defence capability; c) pro-active, verifiable public, unidirectional information channel from...

We study convergence properties of iterative voting procedures. Such procedures are defined by a rule and (restricted) process, where at each step one agent can modify his vote towards better outcome for himself. It is already known that if the iteration dynamics (the manner in which voters allowed to their votes) unrestricted, then process may not converge. For most common rules this be observed even under best response limitation. therefore important investigate whether natural...

Following recent studies of iterative voting and its effects on plurality vote outcomes, we provide characterisations complexity results for three models under the rule. Our focus is providing a better understanding regarding set equilibria attainable by processes. We start with basic model voting. first establish some useful properties equilibria, reachable voting, which enable us to show that deciding whether given profile an iteratively equilibrium NP-complete. then proceed combine...

We consider the problem of limited-bandwidth communication for multi-agent reinforcement learning, where agents cooperate with assistance a protocol and scheduler. The scheduler jointly determine which agent is communicating what message to whom. Under limited bandwidth constraint, required generate informative messages. Meanwhile, an unnecessary connection should not be established because it occupies resources in vain. In this paper, we develop Informative Multi-Agent Communication (IMAC)...

In this work, we suggest representing multiagent systems using computational models, choosing, specifically, Multi-Prover Interactive Protocols to represent agent and the interactions occurring within them. This approach enables us analyze complexity issues related systems. We focus here on of coordination study possible sources complexity. show that there are bounds cannot be lowered even when approximation techniques applied.

This paper studies how automated agents can persuade humans to behave in certain ways. The motivation behind such agent's behavior resides the utility function that designer wants maximize and which may be different from user's function. Specifically, strategic settings studied, agent provides correct yet partial information about a state of world is unknown user but relevant his decision. Persuasion games were designed study interactions between players where one player sends other it way....

We present a novel computational method for advice-generation in path selection problems which are difficult people to solve. The advisor agent's interests may conflict with the of who receive advice. Such optimization settings arise many human-computer applications agents and self-interested but also share certain goals, such as automatic route-selection systems that reason about environmental costs. This paper presents an agent clusters into one several types, based on how their behavior...

In this article, we study automated agents that are designed to encourage humans take some actions over others by strategically disclosing key pieces of information. To end, utilize the framework persuasion games—a branch game theory deals with asymmetric interactions where one player (Sender) possesses more information about world, but it is only other (Receiver) who can an action. particular, use extended model, Sender’s imperfect and Receiver has than two alternative available. We design...

Information uncertainty is one of the major challenges facing applications game theory. In context Stackelberg games, various approaches have been proposed to deal with leader's incomplete knowledge about follower's payoffs, typically by gathering information from interaction follower. Unfortunately, these rely crucially on assumption that follower will not strategically exploit this asymmetry, i.e., behaves truthfully during according their actual payoffs. As we show in paper, may strong...

Current value-based multi-agent reinforcement learning methods optimize individual Q values to guide individuals' behaviours via centralized training with decentralized execution (CTDE). However, such expected, i.e., risk-neutral, value is not sufficient even CTDE due the randomness of rewards and uncertainty in environments, which causes failure these train coordinating agents complex environments. To address issues, we propose RMIX, a novel cooperative MARL method Conditional Value at Risk...

This paper addresses the problem of automated advice provision in settings that involve repeated interactions between people and computer agents. arises many real world applications such as route selection systems office assistants. To succeed agents must reason about how their actions present influence people's future actions. The describes several possible models human behavior were inspired by behavioral economic theories play interactions. These incorporated into agent designs to...

In Doodle polls, each voter approves a subset of the available alternatives according to his preferences. While such polls can be captured by standard models Approval voting, Zou et al. (2015) analyse real-life poll data and conclude that participants' behaviour seems affected considerations other than their intrinsic preferences over alternatives. To capture this phenomenon, they propose model social where voters approve top as well additional `safe' choices so appear cooperative. The...

In this paper we study, for the first time explicitly, implications of endowing an interested party (i.e. a teacher) with ability to modify underlying dynamics environment, in order encourage agent learn follow specific policy. We introduce cost function which can be used by teacher balance modifications it makes environment dynamics, learner's performance compared some ideal, desired, formulate teacher's problem determining optimal changes as planning and control problem, empirically...

Hierarchical reinforcement learning (HRL) is a promising approach to solve tasks with long time horizons and sparse rewards. It often implemented as high-level policy assigning subgoals low-level policy. However, it suffers the non-stationarity problem since constantly changing. The also leads data efficiency problem: policies need more at non-stationary states stabilize training. To address these issues, we propose novel HRL method: Interactive Influence-based Reinforcement Learning...