JMJanusz Marecki

Janusz Marecki

*This is an algorithmically generated author record based on publicly available information.

Janusz Marecki has a total of 13 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Solution Methods for Constrained Markov Decision Process with Continuous Probability Modulation

    by Marek Petrik, Dharmashankar Subramanian, Janusz Marecki

    We propose solution methods for previously-unsolved constrained MDPs in which actions can continuously modify the transition probabilities within some acceptable sets. While many methods have been proposed to solve regular MDPs with large state sets, there are few practical approaches for solving constrained MDPs with large action sets. In particular, we show that the continuous action sets can be replaced by their extreme points when the rewards are linear in the modulation. We also develop a tractable optimization formulation for concave reward functions and, surprisingly, also extend it to non- concave reward functions by using their concave envelopes. We evaluate the effectiveness of the approach on the problem of managing delinquencies in a portfolio of loans.

  • Approximation Methods for Infinite Bayesian Stackelberg Games: Modeling Distributional Payoff Uncertainty

    by Christopher Kiekintveld, Janusz Marecki, Milind Tambe

    Introduction Stackelberg games are increasingly important for informing real-world decision making, including a growing body of work that applies these techniques in security domains such as critical infrastructure protection (Bier, 2007; Sandler and D. G. A. M., 2003), computer networks (Alpcan and Basar, 2003; Nguyen and Basar 2009), and robot patrolling strategies (Agmon et al., 2009; Basilico, Gatti, and Amigoni, 2009; Gatti, 2008). Two software systems that use this type of game modeling are in use by the the Los Angeles International Airport (LAX) (Pita et al., 2008) and the Federal Air Marshals Service (FAMS) (Tsai et al., 2009) to assist with resource allocation decision. A key issue that has arisen in these applications is whether the models can accurately represent the uncertainty that domain experts have about the inputs used to construct the game models, including the preferences and capabilities of terrorist adversaries. To apply game-theoretic reasoning, the first step in the analysis is to construct a precise game model. The typical approach (e.g., in the LAX and FAMS applications) is to construct a model using a combination of the available data and expert opinions. Unfortunately, the data is often limited or imprecise, especially in regard to information about the terrorist adversaries. For example, it can be difficult to predict precisely how attackers will weigh casualties, economic consequences, media exposure, and other factors when selecting targets.


  • Coordinating Randomized Policies for Increasing the Security of Agent Systems

    by Praveen Paruchuri, Jonathan P. Pearce, Janusz Marecki, Milind Tambe, Fernando Ordóñez, Sarit Kraus

    Introduction Security, commonly defined as the ability to deal with intentional threats from other agents, is a major challenge for agents deployed in adversarial environments (Paruchuri et al., 2006). In this paper, we focus on adversarial domains in which the agents have limited information about the adversaries. Such adversarial scenarios arise in a wide variety of situations that are becoming increasingly important, such as patrol agents providing security for a group of houses or regions (Carroll et al., 2005; Paruchuri et al., 2007), UAVs monitoring a humanitarian mission (Beard and Mclain, 2003; Paruchuri et al., 2006), agents assisting in routine security checks at airports (Poole and Passantino, 2003), agents providing privacy in sensor network routing (Ozturk, Zhang, and Trappe, 2004), and agents maintaining anonymity in peer-to-peer networks (Borisov and Waddle, 2005). This paper brings together some of our recent work on how to plan for agents acting in uncertain environments in the presence of adversaries (Paruchuri et al., 2006, 2007, 2008). This research has introduced two very different approaches to increasing security in agent systems and has lead to the ARMOR (Assistant for Randomized Monitoring over Routes) system, which has been deployed for security scheduling at the LAX airport since August 2007 (Murr, 2007; Paruchuri et al., 2008; Pita et al., 2008). Here we will present the main results and algorithms proposed in these two approaches and highlight the relationship between them.

  • Introducing communication in Dis-POMDPs with locality of interaction

    by Makoto Tasaki, Yuichi Yabu, Yuki Iwanari, Makoto Yokoo, Janusz Marecki, Pradeep Varakantham, Milind Tambe

    The Networked Distributed POMDPs (ND-POMDPs) can model multiagent systems in uncertain domains and have begun to scale-up the number of agents. However, prior work in ND-POMDPs has failed to address communication. Without communication, the size of a