Rajdeep K. Dash

Computational-mechanism design: a call to arms

Computational-mechanism design has an important role to play in developing complex distributed systems comprising multiple interacting agents. Game theory has developed powerful tools for analyzing, predicting, and controlling the behavior of self-interested agents and decision making in systems with multiple autonomous actors. These tools, when tailored to computational settings, provide a foundation for building multiagent software systems. This tailoring gives rise to the field of computational-mechanism design, which applies economic principles to computer systems design.

A utility-based sensing and communication model for a glacial sensor network

This paper reports on the development of a utility-based mechanism for managing sensing and communication in cooperative multi-sensor networks. The specific application considered is that of GLACSWEB, a deployed system that uses battery-powered sensors to collect environmental data related to glaciers which it transmits back to a base station so that it can be made available world-wide to researchers. In this context, we first develop a sensing protocol in which each sensor locally adjusts its sensing rate based on the value of the data it believes it will observe. Then, we detail a communication protocol that finds optimal routes for relaying this data back to the base station based on the cost of communicating it (derived from the opportunity cost of using the battery power for relaying data). Finally, we empirically evaluate our protocol by examining the impact on efficiency of the network topology, the size of the network, and the degree of dynamism of the environment. In so doing, we demonstrate that the efficiency gains of our new protocol, over the currently implemented method over a 6 month period, are 470%, 250% and 300% respectively.

Trust-Based Mechanism Design

We define trust-based mechanism design as an augmentation of traditional mechanism design in which agents take into account the degree of trust that they have in their counterparts when determining their allocations. To this end, we develop an efficient, individually rational, and incentive compatible mechanism based on trust. This mechanism is embedded in a task allocation scenario in which the trust in an agent is derived from the reported performance success of that agent by all the other agents in the system. We also empirically study the evolution of our mechanism when iterated and show that, in the long run, it always chooses the most successful and cheapest agents to fulfill an allocation and chooses better allocations than other comparable models when faced with biased reporting.

Trust-based mechanisms for robust and efficient task allocation in the presence of execution uncertainty

Vickrey-Clarke-Groves (VCG) mechanisms are often used to allocate tasks to selfish and rational agents. VCG mechanisms are incentive compatible, direct mechanisms that are efficient (i.e., maximise social utility) and individually rational (i.e., agents prefer to join rather than opt out). However, an important assumption of these mechanisms is that the agents will always successfully complete their allocated tasks. Clearly, this assumption is unrealistic in many real-world applications, where agents can, and often do, fail in their endeavours. Moreover, whether an agent is deemed to have failed may be perceived differently by different agents. Such subjective perceptions about an agents probability of succeeding at a given task are often captured and reasoned about using the notion of trust. Given this background, in this paper we investigate the design of novel mechanisms that take into account the trust between agents when allocating tasks. Specifically, we develop a new class of mechanisms, called trust-based mechanisms, that can take into account multiple subjective measures of the probability of an agent succeeding at a given task and produce allocations that maximise social utility, whilst ensuring that no agent obtains a negative utility. We then show that such mechanisms pose a challenging new combinatorial optimisation problem (that is NP-complete), devise a novel representation for solving the problem, and develop an effective integer programming solution (that can solve instances with about 2×105 possible allocations in 40 seconds).

Computational Mechanism Design for Information Fusion within Sensor Networks

Conventional centralised information fusion and control architectures will be challenged by developments in sensor networks that allow sophisticated autonomous sensors, owned by different stakeholders with individual goals, to interact and share information. Given this, we advocate the use of tools and techniques from computational mechanism design (CMD), a field at the intersection of computer science, game theory and economics, to address the challenges posed by these networks. In particular, CMD allows us to engineer networks with desirable system-wide properties, in which sensors act as rational selfish agents, each attempting to fulfil their own individuals goals through the exchange of observations and information. In this paper, we present our work developing such networks. Specifically, we discuss our development of a generic and principled information valuation metric for sensor networks and we report our experiences applying it within a real world information fusion sensor network scenario

Trusted kernel-based coalition formation

We define Trusted Kernel-based Coalition Formation as a novel extension to the traditional kernel-based coalition formation process which ensures agents choose the most reliable coalition partners and are guaranteed to obtain the payment they deserve. To this end, we develop an encryption-based communication protocol and a payment scheme which ensure that agents cannot manipulate the mechanism to their own benefit. Moreover, we integrate a generic trust model in the coalition formation process that permits the selection of the most reliable agents over repeated coalition games. We empirically evaluate our mechanism when iterated and show that, in the long run, it always chooses the coalition structure that has the maximum expected value and determines the payoffs that match their level of reliability.

A utility-based adaptive sensing and multihop communication protocol for wireless sensor networks

This article reports on the development of a utility-based mechanism for managing sensing and communication in cooperative multisensor networks. The specific application on which we illustrate our mechanism is that of GlacsWeb. This is a deployed system that uses battery-powered sensors to collect environmental data related to glaciers which it transmits back to a base station so that it can be made available world-wide to researchers. In this context, we first develop a sensing protocol in which each sensor locally adjusts its sensing rate based on the value of the data it believes it will observe. The sensors employ a Bayesian linear model to decide their sampling rate and exploit the properties of the Kullback-Leibler divergence to place an appropriate value on the data. Then, we detail a communication protocol that finds optimal routes for relaying this data back to the base station based on the cost of communicating it (derived from the opportunity cost of using the battery power for relaying data). Finally, we empirically evaluate our protocol by examining the impact on efficiency of a static network topology, a dynamic network topology, the size of the network, the degree of dynamism of the environment, and the mobility of the nodes. In so doing, we demonstrate that the efficiency gains of our new protocol, over the currently implemented method over a 6 month period, are 78%, 133%, 100%, and 93%, respectively. Furthermore, we show that our system performs at 65%, 70%, 63%, and 70% of the theoretical optimal, respectively, despite being a distributed protocol that operates with incomplete knowledge of the environment.

Constrained bandwidth allocation in multi-sensor information fusion: a mechanism design approach

Sensor networks are increasingly seen as a solution for a large number of environmental, security and military monitoring tasks. Typically, in these networks, noisy data from a number of local sensors is fused to reduce the uncertainty in the global picture. A central issue in this information fusion is the decision of what data should be shared between sensors, in order to maximize the global gain in information, when the bandwidth of the communication network is limited. In this paper, we study the problem from a selfish agent perspective. We show how the uncertainty in the measurement of an event can be cast as a utility function derived from the Kalman filter. We then use the tools of mechanism design to engineer an incentive-compatible mechanism that allows rational selfish agents to individually maximize their own utility, whilst ensuring that the overall utility of the system is also maximized. We apply the mechanism to multi-sensor target detection and consider the complexity of finding an efficient solution with broadcast communication protocols.

On similarities between inference in game theory and machine learning

In this paper, we elucidate the equivalence between inference in game theory and machine learning. Our aim in so doing is to establish an equivalent vocabulary between the two domains so as to facilitate developments at the intersection of both fields, and as proof of the usefulness of this approach, we use recent developments in each field to make useful improvements to the other. More specifically, we consider the analogies between smooth best responses in fictitious play and Bayesian inference methods. Initially, we use these insights to develop and demonstrate an improved algorithm for learning in games based on probabilistic moderation. That is, by integrating over the distribution of opponent strategies (a Bayesian approach within machine learning) rather than taking a simple empirical average (the approach used in standard fictitious play) we derive a novel moderated fictitious play algorithm and show that it is more likely than standard fictitious play to converge to a payoff-dominant but risk-dominated Nash equilibrium in a simple coordination game. Furthermore we consider the converse case, and show how insights from game theory can be used to derive two improved mean field variational learning algorithms. We first show that the standard update rule of mean field variational learning is analogous to a Cournot adjustment within game theory. By analogy with fictitious play, we then suggest an improved update rule, and show that this results in fictitious variational play, an improved mean field variational learning algorithm that exhibits better convergence in highly or strongly connected graphical models. Second, we use a recent advance in fictitious play, namely dynamic fictitious play, to derive a derivative action variational learning algorithm, that exhibits superior convergence properties on a canonical machine learning problem (clustering a mixture distribution).

Optimal strategies for bidding agents participating in simultaneous Vickrey auctions with perfect substitutes

We derive optimal strategies for a bidding agent that participates in multiple, simultaneous second-price auctions with perfect substitutes. We prove that, if everyone else bids locally in a single auction, the global bidder should always place non-zero bids in all available auctions, provided there are no budget constraints. With a budget, however, the optimal strategy is to bid locally if this budget is equal or less than the valuation. Furthermore, for a wide range of valuation distributions, we prove that the problem of finding the optimal bids reduces to two dimensions if all auctions are identical. Finally, we address markets with both sequential and simultaneous auctions, non-identical auctions, and the allocative efficiency of the market.