This research proposes the use of imitation based learning to build collaborative strategies for a team of agents. Imitation based learning involves learning from an expert by observing her demonstrating a task and then replicating it. This mechanism makes it extremely easy for a knowledge engineer to transfer knowledge to a software agent via human demonstrations. This research aims to apply imitation to learn not only the strategy of an individual agent but also the collaborative strategy of a team of agents to achieve a common goal. The effectiveness of the proposed methodology is being assessed in the domain of RoboCup Soccer Simulation 3D which is a promising platform to address many of the complex real-world problems and offers a truly dynamic, stochastic, and partially-observable environment.

My thesis work aims to study change operations for argumentation systems, especially for abstract argumentation systems à la Dung. This paper presents a first study of the AGM revision adapted to the case of argumentation. We also sketch future research works planned to complete the one already achieved.

Gutierrez and Meseguer show how to enforce consistency during distributed search in the BnB-ADOPT+ algorithm for distributed constraint optimization, but they consider only unconditional deletions. However, during search, more values can be pruned conditionally according to variable instantiations that define subproblems. Enforcing consistency in these subproblems can cause further search space reduction. Here we introduce methods to maintain soft arc consistencies in every subproblem during search. Difficulties lie in the asynchronicity of the algorithm and on the overheads induced by backtracking and undoing. After a careful implementation, experimental results show substantial benefits on several benchmarks.

In reputation systems for multiagent-based e-marketplaces, buying agents model the reputation of selling agents based on ratings shared by other buyers (called advisors). With the existence of unfair rating attacks from dishonest advisors, the effectiveness of reputation systems thus heavily relies on whether buyers can accurately determine which advisors to include in trust networks and their trustworthiness. In this paper, we propose two approaches to deal with unfair rating attacks. The first method is to combine the advantages of different categorical trust models. Secondly, we propose a novel multiagent evolutionary trust model (MET) where each buyer constructs its trust network (information about which advisors should be include in the network and their trustworthiness) by the evolutionary model. Experimental results demonstrate the proposed algorithms are more robust than the state-of-the-art trust models against various unfair rating attacks.

After a disaster, human rescuers may have to wait for better conditions before beginning to search for survivors. A team of robots could enter long before the humans and scope out the environment to gather information that could help to prioritize tasks for the rescue operation. We have developed an algorithm to allow a small group of robots to progressively explore an unknown environment, moving as a group until full exploration is achieved. The novel concept behind this algorithm comes from the way in which the team stays together as a group, maintaining communication, in order to ensure full exploration as well as a path to the exit. We demonstrate in simulation that the algorithm works in multiple environments under varying conditions.

Interactions among agents are complicated since in order to make the best decisions, each agent has to take into account not only the strategy used by other agents but also how those strategies might change in the future (and what causes these changes). The objective of my work will be to develop a framework for learning agent models (opponent or teammate) more accurately and with less interactions, with a special focus on fast learning non-stationary strategies. As preliminary work we have proposed an initial approach for learning non-stationary strategies in repeated games. We use decision trees to learn a model of the agent, and we transform the learned trees into a MDP and solve it to obtain the optimal policy.

Collaborative filtering is a general technique for recommender systems, aiming to provide users with personalized recommendations. However, it suffers from two severe issues known as data sparsity and cold start. In this research, we present two different solutions to ameliorate these issues. First, we propose a trust-aware recommender system to incorporate the ratings of trusted neighbors and to form a more complete rating profile for the active users. Second, we also propose a novel Bayesian similarity measure by taking into account both the direction and length of rating profiles. Both research work shows promising empirical results based on real-world data sets. Finally, we outline the future research on trust-based clustering methods to further alleviate the concerned problems.

The effectiveness of machine learning models can often be improved by feature selection as a pre-processing step. Often this is a data driven process only and can result in models that may not correspond to true relationships present in the data set due to overfitting. In this work, we propose leveraging known relationships between variables to constrain and guide feature selection. Using commonalities across domains, we provide a framework for the user to express model constraints while still making the feature selection process data driven and sensitive to actual relationships in the data.

In this paper, we state the challenges of high-level program execution in multi-agent settings. We first introduce high-level program execution and the related work. Then we describe the completed work, the future work and its approaches. We conclude with the expected contributions of our research.

Our research is within the subfield of modeling trust and reputation in multi-agent systems for online communities. Specifically, in an online community involving users and entities, users provide opinions (ratings) to entities. For each user, we are interested in addressing two problems: (1) how to accurately model the reputation of entities by aggregating opinions from all the users (advisors); and (2) how to cope with the dishonesty of an advisor in providing opinions as well as her subjectivity difference with the user.