Information Agents Research Articles

Lying to Speak the Truth: Selective Manipulation and Improved Information Transmission

ABSTRACTWe analyze a principal‐agent model in which an effort‐averse agent can manipulate a publicly observable performance report. The principal cannot observe the agent's cost of effort, her effort choice, and whether she manipulated the report. An optimal contract links compensation to the realized output and the (possibly manipulated) report. Manipulation can be beneficial to the principal because it can make the report more informative about the agent's effort choice, thereby reducing the agent's information rent. This is achieved through a contract that incentivizes the agent to selectively engage in manipulation based on her effort choice.

Learning in Online Principal-Agent Interactions: The Power of Menus

We study a ubiquitous learning challenge in online principal-agent problems during which the principal learns the agent's private information from the agent's revealed preferences in historical interactions. This paradigm includes important special cases such as pricing and contract design, which have been widely studied in recent literature. However, existing work considers the case where the principal can only choose a single strategy at every round to interact with the agent and then observe the agent's revealed preference through their actions. In this paper, we extend this line of study to allow the principal to offer a menu of strategies to the agent and learn additionally from observing the agent's selection from the menu. We provide a thorough investigation of several online principal-agent problem settings and characterize their sample complexities, accompanied by the corresponding algorithms we have developed. We instantiate this paradigm to several important design problems — including Stackelberg (security) games, contract design, and information design. Finally, we also explore the connection between our findings and existing results about online learning in Stackelberg games, and we offer a solution that can overcome a key hard instance of previous work.

Sliding-mode control for multi-agent systems under the event-triggering WTOD protocol

This paper considers the sliding-mode consensus control problem of the multi-agent systems (MASs), in which an event-triggering weighted try-once-discard (ET-WTOD) protocol is proposed to schedule the transmission of each agent's state information. The key feature of the ET-WTOD protocol is to flexibly adjust the number of transmitted state components at each triggering instant based on the variation of each component and the set thresholds. Then, a token-dependent sliding-mode controller under the ET-WTOD protocol is designed to achieve the desirable consensus of MASs, meanwhile, an iterative optimisation algorithm is used to search the sliding matrix and reduce the sliding domain. Eventually, a numerical simulation example demonstrates the efficacy of the proposed sliding-mode control approach.

Privacy-Preserving Adaptive Resilient Consensus for Multiagent Systems Under Cyberattacks

This article investigates the secure and privacy-preserving consensus problem of multi-agent systems (MASs) with directed interaction topologies under multiple cyber attacks, which contain deception attacks and DoS attacks. First, a unified attack model is introduced to characterize such a multiple attack phenomenon. Besides, considering the existence of eavesdroppers who can intercept the data transmitted on the links, a fully distributed agent value reconstruction method based on the idea of state decomposition is designed to prevent the leakage of the agent's initial information. Then, a novel privacy-preserving adaptive resilient consensus algorithm (PPARCA) with certain graph robustness condition for MASs under the multiple cyber attacks is proposed. The algorithm adaptively takes different countermeasures in the face of different cyber attacks. PPARCA uses the reconstructed agents' states and combines with the modified Secure Acceptance and Broadcast Algorithm (SABA). Theoretical analysis shows that the proposed algorithm can effectively protect the privacy of the initial state of the agents, and reach resilient consensus in the face of cyber attacks. Finally, numerical simulations and Raspberry Pi MASs practical application experiments demonstrate the effectiveness of the proposed results.

Open‐loop equilibrium seeking in two‐agent noncooperative linear‐quadratic dynamic game with asymmetric horizon length

AbstractThis article proposes an iterative‐based algorithm for open‐loop equilibrium seeking in a two‐agent noncooperative dynamic game. The finite‐horizon dynamic games handled in almost all existing articles assume the common prediction horizon length. From the viewpoint of divergence and differences in personal values, it is socially rational to solve the equilibrium of the games with asymmetric prediction horizon length. We thus propose an open‐loop equilibrium‐seeking algorithm without the private knowledge of the other agent's horizon information through a receding‐horizon linear‐quadratic game. Our proposed algorithm is based on iterative optimization. Each agent obtains its own best response control strategy while estimating the other agent's state feedback gain from the state information and guaranteeing the stability of the whole system. We also discuss the performance of the proposed equilibrium‐seeking algorithm through numerical examples.

Understanding how heterogeneous agents affect Principal's returns: Perspectives from short-termism and Bayesian learning

We consider a general framework of optimal contract design under the heterogeneity and short-termism of agents. Our research shows that the optimal contract must weigh the agent's information rent, incentive cost, and benefit to overcome the contract's adverse selection and moral hazards. Agents with higher moral levels were more likely to choose higher effort and lower manipulation. Simultaneously, the principal offers lower incentives and receives more significant payoff. We also extend our model to investigate the benefits of Bayesian learning. Furthermore, we compare the principal's returns in general and learning models and find that the learning contract can bring more profit to the principal.

Consensus control for multi-rate multi-agent systems with fading measurements: the dynamic event-triggered case

This paper studies the distributed -consensus control problem for a class of multi-rate multi-agent systems with fading measurements. A multi-rate sampling strategy is adopted to be more in line with actual need and the channels between each agent and its sensor always fade non-identically. The multi-rate system is transformed into a single-rate system via the lifting technique. For the purpose of reducing the transmission burden, a dynamic event-triggered mechanism is utilized to determine whether the agent's information is allowed to transmit to its neighbours. This paper aims to design an observer-based event-triggered controller for each agent to achieve the -consensus control performance constraint. With the help of the Lyapuonv stability theory, sufficient conditions are obtained that can ensure the desired control performance for the resulting closed-loop systems, and then the desired gain matrices are calculated by solving the linear matrix inequality. Finally, a numerical simulation example is given to demonstrate the effectiveness of the distributed event-triggered consensus control scheme.

Optimal assignment mechanisms with imperfect verification

Objects of different quality are to be allocated to agents. Agents can receive at most one object, and there are not enough high‐quality objects for every agent. The value to the social planner from allocating objects to any given agent depends on that agent's private information. The social planner is unable to use transfers to give incentives for agents to convey their private information. Instead, she is able to imperfectly verify their reports through signals that are positively affiliated with each agent's type. We characterize mechanisms that maximize the social planner's expected payoff. In the optimal mechanism, each agent chooses one of various tracks, which are characterized by two thresholds. If the agent's signal exceeds the upper threshold of the chosen track, the agent receives a high‐quality object, if it is between the two thresholds, he receives a low‐quality object, and if it is below the lower threshold, he receives no object.

Privacy Preservation of Optimization Algorithm Over Unbalanced Directed Graph

This paper proposes an algorithm with the property of privacy preservation to solve the multi-agent optimization problem, where the communication topology is a fixed and strongly connected directed graph with a row stochastic weight matrix. Here, we consider the external eavesdropper attack model and take the gradient information of the objective function as the agent's privacy information. In this algorithm, each agent adds an additional state variable that interacts with it with a time-varying weight. And this new variable performs gradient iterative calculation. The original state variable interacts with the new variable and the original neighbors. It is proved that the algorithm converges to the optimal solution of the problem and at the same time achieves the purpose of privacy preservation. In addition, the algorithm neither requires additional hidden signals, nor does it increase a large amount of calculation. Finally, a simulation example is given to verify the performance of the algorithm.

Addressing Strategic Uncertainty with Incentives and Information

A principal privately contracts with a set of agents who then simultaneously make a binary decision. Each contract specifies an individual allocation and the information the agent is given about a fundamental state and other agents' contracts. We study the principal's optimal scheme that induces a desired action profile as the unique rationalizable outcome. Our main result reduces this multiagent problem to a two-step procedure where information is designed agent-by-agent: the principal chooses a fundamental-state-contingent distribution over agent rankings and, separately for each agent, the agent's information about the realized ranking and fundamental states. We illustrate with a team-production application.

Stochastic choice and rational inattention

We consider a decision maker who first chooses a costly information structure, and then chooses an action after receiving a signal. The choice of action is observed, but the choice of information is not. Moreover, the state of nature is considered unobservable as well. In such a situation, the choice of action appears random and cannot be conditioned on the states from an outside analyst's point of view. We show that, given only stochastic choice from menus of actions, an analyst can identify the agent's taste (risk attitude), prior belief, and information cost function. In addition, we discuss the behavioral implications of our model that are weaker than some key properties of random expected utility models because of the endogeneity and hence menu-dependence of private information. Finally, we provide necessary and sufficient conditions for stochastic choice to be rationalized by our model.

Optimizing Payments in Dominant-Strategy Mechanisms for Multi-Parameter Domains

In AI research, mechanism design is typically used to allocate tasks and resources to agents holding private information about their values for possible allocations. In this context, optimizing payments within the Groves class has recently received much attention, mostly under the assumption that agent's private information is single-dimensional. Our work tackles this problem in multi-parameter domains. Specifically, we develop a generic technique to look for a best Groves mechanism for any given mechanism design problem. Our method is based on partitioning the spaces of agent values and payment functions into regions, on each of which we are able to define a feasible linear payment function. Under certain geometric conditions on partitions of the two spaces this function is optimal. We illustrate our method by applying it to the problem of allocating heterogeneous items.

Game Theory Without Partitions, and Applications to Speculation and Consensus

Abstract Decision theory and game theory are extended to allow for information processing errors. This extended theory is then used to reexamine market speculation and consensus, both when all actions (opinions) are common knowledge and when they may not be. Five axioms of information processing are shown to be especially important to speculation and consensus. They are called nondelusion, knowing that you know (KTYK), nested, balanced, and positively balanced. We show that it is necessary and sufficient that each agent's information processing errors be (1) nondeluded and balanced so that the agents cannot agree to disagree, (2) nondeluded and positively balanced so that it cannot be common knowledge that they are speculating, and (3) nondeluded and KTYK and nested so that agents cannot speculate in equilibrium. Each condition is strictly weaker than the next one, and the last is strictly weaker than partition information.

On the precision of information

AbstractThis paper explores a model in which the agent's effort affects (solely) the precision (variance) of a performance measure (signal) of the outcome. Both the principal and the agent are risk averse. The contract the principal offers is composed of a fixed payment plus variable compensation, depending on the outcome and based on a (linear) risk‐sharing rule between the principal and the agent. Moral hazard alone leads to an upward distortion (above the first‐best) of the risk‐sharing rule. This serves to induce more effort to increase information precision. Adverse selection alone introduces two new features. First, to reduce informational rents, the risk‐sharing rule is distorted downward below the first‐best. Second, to induce truthful information revelation, the risk‐sharing rule becomes increasing with the expected outcome—the agent's private information. When moral hazard and adverse selection are considered together, and there is enough uncertainty related to adverse selection, the risk‐sharing rule is above the first‐best for more efficient types, but below the first‐best for less efficient types. Further, the precision of the signal increases with the expected outcome.

Decentralized Control of Large Collaborative Swarms Using Random Finite Set Theory

Controlling large swarms of robotic agents presents many challenges including but not limited to computational complexity due to a large number of agents, uncertainty in the functionality of each agent in the swarm, and uncertainty in the swarm's configuration. The contribution of this work is to decentralize random finite set (RFS)-based control of large collaborative swarms for controlling individual agents. The RFS-based control formulation assumes a Gaussian mixture probability hypothesis density approximation and a complete topology for centralized swarm control. To generalize the control topology in a localized or decentralized manner, a sparse linear quadratic regulator (LQR) is used to sparsify the RFS-based control gain matrix obtained using iterative LQR. This allows agents to use information of agents near each other (localized topology) or only the agent's own information (decentralized topology) to make a control decision. Sparsity and performance for decentralized RFS-based control are compared for different degrees of localization in feedback control gains which show that the stability and performance compared to centralized control do not degrade significantly in providing RFS-based control for large collaborative swarms.

Collusion and delegation under information control

This paper studies how information control affects incentives for collusion and optimal organizational structures in principal-supervisor-agent relationships. I consider a model in which the principal designs the supervisor's signal on the productive agent's private information and the supervisor and agent may collude. I show that the principal optimally delegates the interaction with the agent to the supervisor if either the supervisor's budget is large or the value of production is small. The principal prefers direct communication with the supervisor and agent if the supervisor's budget is sufficiently small and the value of production is high.

Robust $\mathscr{H}_{\infty }$ Consensus for Markov Jump Multiagent Systems Under Mode-Dependent Observer and Quantizer

This article investigates the robust \mathscr H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">∞</sub> consensus issue for Markov jump multiagent systems with external disturbances. The agent's information is quantized via a mode-dependent logarithmic quantizer before being transmitted to its neighbors. For each agent, a mode-dependent observer is designed for that only agent's measurement output can be directly transmitted to the neighbors. The original closed-loop system is converted to a reduced-order decomposed system. Then, a mode-dependent and decomposed-system-related Lyapunov function is constructed. According to such a Lyapunov function, some criteria in term of linear matrix inequality on stochastic stability and \mathscr H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">∞</sub> -based condition analysis for the reduced-order decomposed system are derived. The observer and consensus controller gains are explicitly deduced with the aid of the singular value decomposition method. In the end, a numerical example and a practical example of LC oscillator network show the feasibility of the developed consensus protocol.

Learning to Communicate Implicitly by Actions

In situations where explicit communication is limited, human collaborators act by learning to: (i) infer meaning behind their partner's actions, and (ii) convey private information about the state to their partner implicitly through actions. The first component of this learning process has been well-studied in multi-agent systems, whereas the second — which is equally crucial for successful collaboration — has not. To mimic both components mentioned above, thereby completing the learning process, we introduce a novel algorithm: Policy Belief Learning (PBL). PBL uses a belief module to model the other agent's private information and a policy module to form a distribution over actions informed by the belief module. Furthermore, to encourage communication by actions, we propose a novel auxiliary reward which incentivizes one agent to help its partner to make correct inferences about its private information. The auxiliary reward for communication is integrated into the learning of the policy module. We evaluate our approach on a set of environments including a matrix game, particle environment and the non-competitive bidding problem from contract bridge. We show empirically that this auxiliary reward is effective and easy to generalize. These results demonstrate that our PBL algorithm can produce strong pairs of agents in collaborative games where explicit communication is disabled.

Distributed Linear Estimation Via a Roaming Token

We present an algorithm for the problem of linear distributed estimation of a parameter in a network where a set of agents are successively taking measurements. The approach considers a roaming token in a network that carries the estimate, and jumps from one agent to another in its vicinity according to the probabilities of a Markov chain. When the token is at an agent it records the agent's local information. We analyze the proposed algorithm and show that it is consistent and asymptotically optimal, in the sense that its mean-square-error (MSE) rate of decay approaches the centralized one as the number of iterations increases. We show these results for a scenario where the network changes over time, and we consider two different sets of assumptions on the network instantiations: (I) they are i.i.d. and connected on the average, or (II) that they are deterministic and strongly connected for every finite time window of a fixed size. Simulations show our algorithm is competitive with consensus+innovations and diffusion type of algorithms, achieving a smaller MSE at each iteration in all considered scenarios.

Collusion and delegation under information control

This paper studies how information control affects incentives for collusion and optimal organizational structures in principal–supervisor–agent relationships. I consider a model in which the principal designs the supervisor's signal on the productive agent's private information, and the supervisor and agent may collude. I show that the principal optimally delegates the interaction with the agent to the supervisor if either the supervisor's budget is large or the value of production is small. The principal prefers direct communication with the supervisor and agent if the supervisor's budget is sufficiently small and the value of production is high.