Goal recognition over POMDPs: inferring the intention of a POMDP agent

Plan recognition is the problem of inferring the goals and plans of an agent from partial observations of her behavior. Recently, it has been shown that the problem can be formulated and solved usingplanners, reducing plan recognition to plan generation.In this work, we extend this model-basedapproach to plan recognition to the POMDP setting, where actions are stochastic and states are partially observable. The task is to infer a probability distribution over the possible goals of an agent whose behavior results from a POMDP model. The POMDP model is shared between agent and observer except for the true goal of the agent that is hidden to the observer. The observations are action sequences O that may contain gaps as some or even most of the actions done by the agent may not be observed. We show that the posterior goal distribution P(GjO) can be computed from the value function VG(b) over beliefs b generated by the POMDPplanner for each possible goal G. Some extensionsof the basic framework are discussed, and a numberof experiments are reported.

Relying on the agent in charge of production for project evaluation

I study the optimal project choice when the principal relies on the agent in charge of production for project evaluation. The principal has to choose between a safe project generating a fixed revenue and a risky project generating an uncertain revenue. The agent has private information about the production cost under each project but also about the signal regarding the profitability of the risky project. If the signal favoring the adoption of the risky project is goods news to the agent, integrating production and project evaluation tasks does not generate any loss compared to the benchmark in which the principal herself receives the signal. By contrast, if it is bad news, task integration creates an endogenous reservation utility which is type-dependent and thereby generates countervailing incentives, which can make a bias toward either project optimal. Our results can offer an explanation for why good firms can go bad and a rationale for the separation of day-to-day operating decisions from long-term strategic decisions stressed by Williamson.

Are agent-based simulations robust? The wholesale electricity trading case

Agent-based computational economics is becoming widely used in practice. This paperexplores the consistency of some of its standard techniques. We focus in particular on prevailingwholesale electricity trading simulation methods. We include different supply and demandrepresentations and propose the Experience-Weighted Attractions method to include severalbehavioural algorithms. We compare the results across assumptions and to economic theorypredictions. The match is good under best-response and reinforcement learning but not underfictitious play. The simulations perform well under flat and upward-slopping supply bidding,and also for plausible demand elasticity assumptions. Learning is influenced by the number ofbids per plant and the initial conditions. The overall conclusion is that agent-based simulationassumptions are far from innocuous. We link their performance to underlying features, andidentify those that are better suited to model wholesale electricity markets.

Solving heterogeneous-agent models by projection and perturbation

The paper proposes a numerical solution method for general equilibrium models with a continuum of heterogeneous agents, which combines elements of projection and of perturbation methods. The basic idea is to solve first for the stationary solutionof the model, without aggregate shocks but with fully specified idiosyncratic shocks. Afterwards one computes a first-order perturbation of the solution in the aggregate shocks. This approach allows to include a high-dimensional representation of the cross-sectional distribution in the state vector. The method is applied to a model of household saving with uninsurable income risk and liquidity constraints. The model includes not only productivity shocks, but also shocks to redistributive taxation, which cause substantial short-run variation in the cross-sectional distribution of wealth. If those shocks are operative, it is shown that a solution method based on very few statistics of the distribution is not suitable, while the proposed method can solve the model with high accuracy, at least for the case of small aggregate shocks. Techniques are discussed to reduce the dimension of the state space such that higher order perturbations are feasible.Matlab programs to solve the model can be downloaded.

Organisational adaptation of multi-agent systems in a peer-to-peer scenario

Organisations in Multi-Agent Systems (MAS) have proven to be successful in regulating agent societies. Nevertheless, changes in agents' behaviour or in the dynamics of the environment may lead to a poor fulfilment of the system's purposes, and so the entire organisation needs to be adapted. In this paper we focus on endowing the organisation with adaptation capabilities, instead of expecting agents to be capable of adapting the organisation by themselves. We regard this organisational adaptation as an assisting service provided by what we call the Assistance Layer. Our generic Two Level Assisted MAS Architecture (2-LAMA) incorporates such a layer. We empirically evaluate this approach by means of an agent-based simulator we have developed for the P2P sharing network domain. This simulator implements 2-LAMA architecture and supports the comparison between different adaptation methods, as well as, with the standard BitTorrent protocol. In particular, we present two alternatives to perform norm adaptation and one method to adapt agents'relationships. The results show improved performance and demonstrate that the cost of introducing an additional layer in charge of the system's adaptation is lower than its benefits.

Spatial behavior in groups: an agent-based approach

We present an agent-based model with the aim of studying how macro-level dynamics of spatial distances among interacting individuals in a closed space emerge from micro-level dyadic and local interactions. Our agents moved on a lattice (referred to as a room) using a model implemented in a computer program called P-Space in order to minimize their dissatisfaction, defined as a function of the discrepancy between the real distance and the ideal, or desired, distance between agents. Ideal distances evolved in accordance with the agent's personal and social space, which changed throughout the dynamics of the interactions among the agents. In the first set of simulations we studied the effects of the parameters of the function that generated ideal distances, and in a second set we explored how group macrolevel behavior depended on model parameters and other variables. We learned that certain parameter values yielded consistent patterns in the agents' personal and social spaces, which in turn led to avoidance and approaching behaviors in the agents. We also found that the spatial behavior of the group of agents as a whole was influenced by the values of the model parameters, as well as by other variables such as the number of agents. Our work demonstrates that the bottom-up approach is a useful way of explaining macro-level spatial behavior. The proposed model is also shown to be a powerful tool for simulating the spatial behavior of groups of interacting individuals.

Promoting the development of secure mobile agent applications

Peer-reviewed

Securing dynamic itineraries for mobile agent applications

In this paper we present a novel mechanism for the protection of dynamic itineraries for mobile agent applications. Itineraries that are decided as the agent goes are essential in complex applications based on mobile agents, but no approach has been presented until now to protect them. We have conceived a cryptographic scheme for shielding dynamic itineraries from tampering, impersonation and disclosure. By using trust strategically, our scheme provides a balanced trade-off between flexibility and security. Our protection scheme has been thought always bearing in mind a feasible implementation, and thus facilitates the development of applications that make use of it. An example application based on a real healthcare scenario is also presented to show its operation.

ACACIA: an agent-based program for simulating behavior to reach long-term goals

We present ACACIA, an agent-based program implemented in Java StarLogo 2.0 that simulates a two-dimensional microworld populated by agents, obstacles and goals. Our program simulates how agents can reach long-term goals by following sensorial-motor couplings (SMCs) that control how the agents interact with their environment and other agents through a process of local categorization. Thus, while acting in accordance with this set of SMCs, the agents reach their goals through the emergence of global behaviors. This agent-based simulation program would allow us to understand some psychological processes such as planning behavior from the point of view that the complexity of these processes is the result of agent-environment interaction.

Flocking behaviour: Agent-based simulation and hierarchical leadership.

We have studied how leaders emerge in a group as a consequence of interactions among its members. We propose that leaders can emerge as a consequence of a self-organized process based on local rules of dyadic interactions among individuals. Flocks are an example of self-organized behaviour in a group and properties similar to those observed in flocks might also explain some of the dynamics and organization of human groups. We developed an agent-based model that generated flocks in a virtual world and implemented it in a multi-agent simulation computer program that computed indices at each time step of the simulation to quantify the degree to which a group moved in a coordinated way (index of flocking behaviour) and the degree to which specific individuals led the group (index of hierarchical leadership). We ran several series of simulations in order to test our model and determine how these indices behaved under specific agent and world conditions. We identified the agent, world property, and model parameters that made stable, compact flocks emerge, and explored possible environmental properties that predicted the probability of becoming a leader.

Emergent patterns of social organization in captive Cercocebus torquatus: Testing the GrooFiWorld agent-based model

We empirically applied the GrooFiWorld agent-based model (Puga-González et al. 2009) in a group of captive mangabeys (Cercocebus torquatus). We analysed several measurements related to aggression and affiliative patterns. The group adopted a combination of despotic and egalitarian behaviours resulting from the behavioural flexibility observed in the Cercopithecinae subfamily. Our study also demonstrates that the GrooFiWorld agent-based model can be extended to other members of the Cercopithecinae subfamily generating parsimonious hypotheses related to the social organization.

Towards the Construction of a Multi-agent Approach for Discovering the Meaning of Natural Language Collaborative Conversations

Peer-reviewed

Social Presence Approach Within the Question and Answering eLearning Model: An Experiment with a Multi-Agent System

The model of Questions Answering (Q&A) for eLearning is based on collaborative learning through questions that are posed by students and their answers to that questions which are given by peers, in contrast with the classical model in which students ask questions to the teacher only. In this proposal we extend the Q&A model including the social presence concept and a quantitative measure of it is proposed; besides it is considered the evolution of the resulting Q&A social network after the inclusion of the social presence and taking into account the feedback on questions posed by students and answered by peers. The social network behaviorwas simulated using a Multi-Agent System to compare the proposed social presence model with the classical and the Q&A models