Ron Katz

JAAMAS Journal 2013 Journal Article

Efficient bidding strategies for Cliff-Edge problems

• Rina Azoulay
• Ron Katz
• Sarit Kraus

Abstract In this paper, we propose an efficient agent for competing in Cliff-Edge (CE) and simultaneous Cliff-Edge (SCE) situations. In CE interactions, which include common interactions such as sealed-bid auctions, dynamic pricing and the ultimatum game (UG), the probability of success decreases monotonically as the reward for success increases. This trade-off exists also in SCE interactions, which include simultaneous auctions and various multi-player ultimatum games, where the agent has to decide about more than one offer or bid simultaneously. Our agent competes repeatedly in one-shot interactions, each time against different human opponents. The agent learns the general pattern of the population’s behavior, and its performance is evaluated based on all of the interactions in which it participates. We propose a generic approach which may help the agent compete against unknown opponents in different environments where CE and SCE interactions exist, where the agent has a relatively large number of alternatives and where its achievements in the first several dozen interactions are important. The underlying mechanism we propose for CE interactions is a new meta-algorithm, deviated virtual learning (DVL), which extends existing methods to efficiently cope with environments comprising a large number of alternative decisions at each decision point. Another competitive approach is the Bayesian approach, which learns the opponents’ statistical distribution, given prior knowledge about the type of distribution. For the SCE, we propose the simultaneous deviated virtual reinforcement learning algorithm (SDVRL), the segmentation meta-algorithm as a method for extending different basic algorithms, and a heuristic called fixed success probabilities (FSP). Experiments comparing the performance of the proposed algorithms with algorithms taken from the literature, as well as other intuitive meta-algorithms, reveal superiority of the proposed algorithms in average payoff and stability as well as in accuracy in converging to the optimal action, both in CE and SCE problems.

AAMAS Conference 2008 Conference Paper

How Automated Agents Treat Humans and Other Automated Agents in Situations of Inequity: An Experimental Study

• Ron Katz
• Sarit Kraus

This paper explores the question of how agent designers perceive and treat their agent’s opponents. In particular, it examines the influence of the opponent’s identity (human vs. automated agent) in negotiations. We empirically demonstrate that when people interact spontaneously they treat human opponents differently than automated agents in the context of equity and fairness considerations. However, these difference vanish when people design and implement agents that will interact on their behalf. Nevertheless, the commitment of the agents to honor agreements with people is higher than their commitment to other agents. In the experiments, which comprised 147 computer science students, we used the Colored Trails game as the negotiation environment. We suggest possible explanations for the relationships among online players, agent designers, human opponents and automated opponents.

IJCAI Conference 2007 Conference Paper

• Efrat Manisterski
• Ron Katz
• Sarit Kraus

This paper presents a novel approach for providing automated trading agents to a population, focusing on bilateral negotiation with unenforceable agreements. A new type of agents, called semi-cooperative (SC) agents is proposed for this environment. When these agents negotiate with each other they reach a pareto-optimal solution that is mutually beneficial. Through extensive experiments we demonstrate the superiority of providing such agents for humans over supplying equilibrium agents or letting people design their own agents. These results are based on our observation that most people do not modify SC agents even though they are not in equilibrium. Our findings introduce a new factor ---human response to provided agents --- that should be taken into consideration when developing agents that are provided to a population.

AAAI Conference 2007 Conference Paper

Gender-Sensitive Automated Negotiators

• Ron Katz

This paper introduces an innovative approach for automated negotiating using the gender of human opponents. Our approach segments the information acquired from previous opponents, stores it in two databases, and models the typical behavior of males and of females. The two models are used in order to match an optimal strategy to each of the two subpopulations. In addition to the basic separation, we propose a learning algorithm which supplies an online indicator for the gender separability-level of the population, which tunes the level of separation the algorithm activates. The algorithm we present can be generally applied in different environments with no need for configuration of parameters. Experiments in 4 different one-shot domains, comparing the performance of the gender based separation approach with a basic approach which is not gender sensitive, revealed higher payoffs of the former in almost all the domains. Moreover, using the proposed learning algorithm further improved the results.

AAMAS Conference 2006 Conference Paper

Efficient agents for Cliff Edge environments with a large set of decision options

• Ron Katz
• Sarit Kraus

AAAI Conference 2006 Conference Paper

Modeling Human Decision Making in Cliff-Edge Environments

• Ron Katz

In this paper we propose a model for human learning and decision making in environments of repeated Cliff-Edge (CE) interactions. In CE environments, which include common daily interactions, such as sealed-bid auctions and the Ultimatum Game (UG), the probability of success decreases monotonically as the expected reward increases. Thus, CE environments are characterized by an underlying conflict between the strive to maximize profits and the fear of causing the entire deal to fall through. We focus on the behavior of people who repeatedly compete in one-shot CE interactions, with a different opponent in each interaction. Our model, which is based upon the Deviated Virtual Reinforcement Learning (DVRL) algorithm, integrates the Learning Direction Theory with the Reinforcement Learning algorithm. We also examined several other models, using an innovative methodology in which the decision dynamics of the models were compared with the empirical decision patterns of individuals during their interactions. An analysis of human behavior in auctions and in the UG reveals that our model fits the decision patterns of far more subjects than any other model.