Sandip Sen

Evolving Beharioral Strategies in Predators and Prey

The predator/prey domain is utilized to conduct research in Distributed Artificial Intelligence. Genetic Programming is used to evolve behavioral strategies for the predator agents. To further the utility of the predator strategies, the prey population is allowed to evolve at the same time. The expected competitive learning cycle did not surface. This failing is investigated, and a simple prey algorithm surfaces, which is consistently able to evade capture from the predator algorithms.

Searching for optimal coalition structures

Coalition formation has been a very active area of research in multiagent systems. Most of this research has concentrated on decentralized procedures that allow self-interested agents to negotiate the formation of coalitions and division of coalition payoffs. A different line of research has addressed the problem of finding the optimal division of agents into coalitions such that the sum total of the the payoffs to all the coalitions is maximized (Larson and Sandholm, 1999). This is the optimal coalition structure identification problem. Deterministic search algorithms have been proposed and evaluated under the assumption that the performance of a coalition is independent of other coalitions. We use an order-based genetic algorithm (OBGA) as a stochastic search process to identify the optimal coalition structure. We compare the performance of the OBGA with a representative deterministic algorithm presented in the literature. Though the OBGA has no performance guarantees, it is found to dominate the deterministic algorithm in a significant number of problem settings. An additional advantage of the OBGA is its scalability to larger problem sizes and to problems where performance of a coalition depends on other coalitions in the environment.

Voting for movies: the anatomy of a recommender system

Personal assistant agents embody a clearly beneficial application of intelligent agent technology. A particular kind of assistant agents, recommender systems (RSs), can be used to recommend items of interest to users [l]. To be successful, such systems should be able to model and reason with user preferences for items in the application domain. We are developing a movie recommender system that caters to the interests of a user. Our primary concern is to utilize a reasoning procedure that can meaningfully and systematically tradeoff between conflicting user preferences. We have adapted mechanisms from voting theory that have desirable guarantees regarding the recommendations generated from stored preferences. We provide multiple query modalities by which the user can pose unconstrained, constrained, or instance-based queries. Typically a domain has several features or dimensions. Each dimension consists of a collection of elements, and the preferences of a user are given by his/her ratings of those elements on some ordinal or cardinal scale. To obtain a recommendation rating for a given item, an RS considers the feature values of that item, obtains ratings for these values from corresponding dimensions, and then combines these ratings by some evaluation scheme.

Believing others: Pros and cons

Abstract In open environments there is no central control over agent behaviors. On the contrary, agents in such systems can be assumed to be primarily driven by self interests. Under the assumption that agents remain in the system for significant time periods, or that the agent composition changes only slowly, we have previously presented a prescriptive strategy for promoting and sustaining cooperation among self-interested agents. The adaptive, probabilistic policy we have prescribed promotes reciprocative cooperation that improves both individual and group performance in the long run. In the short run, however, selfish agents could still exploit reciprocative agents. In this paper, we evaluate the hypothesis that the exploitative tendencies of selfish agents can be effectively curbed if reciprocative agents share their “opinions” of other agents. Since the true nature of agents is not known a priori and is learned from experience, believing others can also pose its own hazards. We provide a learned trust-based evaluation function that is shown to resist both individual and concerted deception on the part of selfish agents in a package delivery domain.

Satisfying user preferences while negotiating meetings

Our research agenda focuses on building software agents that can facilitate and streamline group problem solving in organizations. We are particularly interested in developing intelligent agents that can partially automate routine information processing tasks by representing and reasoning with the preferences and biases of associated users. The distributed meeting scheduler is a collection of agents, responsible for scheduling meetings for their respective users. Users have preferences on when they like to meet, e.g. time of day, day of week, status of other invitees, topic of the meeting, etc. The agent must balance such concerns, proposing and accepting meeting times that satisfy as many of these criteria as possible. For example, a user might prefer not to meet at lunchtime unless the president of the company is hosting the meeting. We apply techniques from voting theory to arrive at consensus choices for meeting times while balancing different preferences.

Developing an automated distributed meeting scheduler

Automated scheduling agents allow users to concentrate on productive tasks and to improve the quality of information processing. The author uses a distributed approach with intelligent agents to design and develop efficient meeting scheduling. The purpose of the project is to design and implement a software system that uses intelligent meeting-scheduling agents that can negotiate with other agents without compromising their user-specified constraints. A small group of researchers in our department runs the prototype system, and we are using feedback from them to improve the interface and to expand the functionality of the system.

Social instruments for robust convention emergence

We present the notion of Social Instruments as mechanisms that facilitate the emergence of conventions from repeated interactions between members of a society. Specifically, we focus on two social instruments: rewiring and observation. Our main goal is to provide agents with tools that allow them to leverage their social network of interactions when effectively addressing coordination and learning problems, paying special attention to dissolving metastable subconventions. Our initial experiments throw some light on how Self-Reinforcing Substructures (SRS) in the network prevent full convergence to society-wide conventions, resulting in reduced convergence rates. The use of an effective composed social instrument, observation + rewiring, allow agents to achieve convergence by eliminating the subconventions that otherwise remained meta-stable.

On the design of an adaptive meeting scheduler

We present design considerations for an automated meeting scheduling agent that processes meeting requests on behalf of its associated user. In our formulation of the meeting scheduling problem, distributed meeting scheduling agents, one per user, intelligently exchange information with each other to schedule meetings without compromising user-specified constraints. In this paper, we first enumerate various strategies we have investigated to focus distributed negotiation between scheduling agents. Next, we demonstrate the necessity for such a scheduler to be adaptive in its choice of options for the various strategy dimensions, so that it can perform effectively over time. In order to build an adaptive scheduler that can effectively choose from available strategy options, we develop quantitative performance estimates of these options using detailed probabilistic analysis. Results from these analyses are used to provide guidelines to choose the most appropriate strategy combination given current environmental conditions and local problem-solving states.<<ETX>>

Topology and Memory Effect on Convention Emergence

Social conventions are useful self-sustaining protocols for groups to coordinate behavior without a centralized entity enforcing coordination. We perform an in-depth study of different network structures, to compare and evaluate the effects of different network topologies on the success and rate of emergence of social conventions. While others have investigated memory for learning algorithms, the effects of memory or history of past activities on the reward received by interacting agents have not been adequately investigated. We propose a reward metric that takes into consideration the past action choices of the interacting agents. The research question to be answered is what effect does the history based reward function and the learning approach have on convergence time to conventions in different topologies. We experimentally investigate the effects of history size, agent population size and neighborhood size the emergence of social conventions.

Individual learning of coordination knowledge

Abstract. Social agents, both human and computational, inhabiting a world containing multiple active agents, need to coordinate their activities. This is because agents share resources, and without proper coordination or ‘rules of the road’, everybody will be interfering with the plans of others. As such, we need coordination schemes that allow agents to effectively achieve local goals without adversely affecting the problem-solving capabilities of other agents. Researchers in the field of Distributed Artificial Intelligence (DAI) have developed a variety of coordination schemes under different assumptions about agent capabilities and relationships. Whereas some of these researchers have been motivated by human cognitive biases, others have approached it as an engineering problem of designing the most effective coordination architecture or protocol. We evaluate individual and concurrent learning by multiple, autonomous agents as a means for acquiring coordination knowledge. We show that a uniform reinforc...