Shulamit Reches

A statistical decision-making model for choosing among multiple alternatives

Automated agents often have several alternatives to choose from in order to solve a problem. Usually the agent does not know in advance which alternative is the best one, so some exploration is required. However, in most cases there is a cost associated with exploring the domain, which must be minimized in order to be worthwhile. We concentrate on cases where the agent has some prior knowledge about each alternative, which is expressed in terms of units of information. A unit of information about an alternative is the result of choosing the alternative - for example, in the e-commerce domain one unit of information can be a customers impression or feedback about a supplier; in the heuristic domain one unit of information can be the observed result of running one simulation with a given heuristic function. In our environments the agent has a-priori only a small number of units of information about each alternative, and it would like to use this knowledge in deciding between its alternatives. Nevertheless, since the agent has only a limited number of units of information, deciding between the alternatives solely based on these units may be risky. In extreme cases, they can even mislead the agent to choose the worst alternative rather than the best one.

MLBP: MAS for large-scale biometric pattern recognition

Security systems can observe and hear almost anyone every-where. However, it is impossible to employ an adequate number of human experts to analyze the information explosion. In this paper we present an autonomous multi-agent framework which, as an input, obtains biometric information acquired at a set of locations. The framework aims in real-time to point out individuals who act according to a suspicious pattern across these locations. The system works in large-scale scenarios. We present a scenario to demonstrate the usefulness of the framework. The goal is to point out individuals who visited a sequence of airports using face recognition algorithms. Simulation results show a high overall accuracy of our system in real-time.

Decision making with dynamic uncertain events

When to make a decision is a key question in decision making problems characterized by uncertainty. In this paper we deal with decision making in environments where information arrives dynamically. We address the tradeoff between waiting and stopping strategies. On the one hand, waiting to obtain more information reduces uncertainty, but it comes with a cost. Stopping and making a decision based on an expected utility reduces the cost of waiting, but the decision is based on uncertain information. We propose an optimal algorithm and two approximation algorithms. We prove that one approximation is optimistic - waits at least as long as the optimal algorithm, while the other is pessimistic - stops not later than the optimal algorithm. We evaluate our algorithms theoretically and empirically and show that the quality of the decision in both approximations is near-optimal and much faster than the optimal algorithm. Also, we can conclude from the experiments that the cost function is a key factor to chose the most effective algorithm.

MLBPR: MAS for Large-Scale Biometric Pattern Recognition

Security systems can observe and hear almost anyone everywhere. However, it is impossible to employ an adequate number of human experts to analyze the information explosion. In this paper, we present a multi-agent framework which works in large-scale scenarios and responds in real time. The input for the framework is biometric information acquired at a set of locations. The framework aims to point out individuals who act according to a suspicious pattern across these locations. The framework works in large-scale scenarios. We present two scenarios to demonstrate the usefulness of the framework. The goal in the first scenario is to point out individuals who visited a sequence of airports, using face recognition algorithms. The goal in the second scenario is to point out individuals who called a set of phones, using speaker recognition algorithms. Theoretical performance analysis and simulation results show a high overall accuracy of our system in real-time.

Choosing a Candidate Using Efficient Allocation of Biased Information

This article deals with a decision-making problem concerning an agent who wants to choose a partner from multiple candidates for long-term collaboration. To choose the best partner, the agent can rely on prior information he knows about the candidates. However, to improve his decision, he can request additional information from information sources. Nonetheless, acquiring information from external information sources about candidates may be biased due to different personalities of the agent searching for a partner and the information source. In addition, information may be costly. Considering the bias and the cost of the information sources, the optimization problem addressed in this article is threefold: (1) determining the necessary amount of additional information, (2) selecting information sources from which to request the information, and (3) choosing the candidates on whom to request the additional information. We propose a heuristic to solve this optimization problem. The results of experiments on simulated and real-world domains demonstrate the efficiency of our algorithm.

Efficiently gathering information in costly domains

This paper proposes a novel technique for allocating information gathering actions in settings where agents need to choose among several alternatives, each of which provides a stochastic outcome to the agent. Samples of these outcomes are available to agents prior to making decisions and obtaining further samples is associated with a cost. The paper formalizes the task of choosing the optimal sequence of information gathering actions in such settings and establishes it to be NP-Hard. It suggests a novel estimation technique for the optimal number of samples to obtain for each of the alternatives. The approach takes into account the trade-offs associated with using prior samples to choose the best alternative and paying to obtain additional samples. This technique is evaluated empirically in several different settings using real data. Results show that our approach was able to significantly outperform alternative algorithms from the literature for allocating information gathering actions in similar types of settings. These results demonstrate the efficacy of our approach as an efficient, tractable technique for deciding how to acquire information when agents make decisions under uncertain conditions.