Marta Vomlelová

The SACSO methodology for troubleshooting complex systems

The paper describes the task of performing efficient decision-theoretic troubleshooting of electromechanical devices. In general, this task is NP-complete, but under fairly strict assumptions, a greedy approach will yield an optimal sequence of actions, as discussed in the paper. This set of assumptions is weaker than the set proposed by Heckerman et al. (1995). However, the printing system domain, which motivated the research and which is described in detail in the paper, does not meet the requirements for the greedy approach, and a heuristic method is used. The method takes value of identification of the fault into account and it also performs a partial two-step look-ahead analysis. We compare the results of the heuristic method with optimal sequences of actions, and find only minor differences between the two.

Troubleshooting: NP-hardness and solution methods

Abstract Troubleshooting is one of the areas where Bayesian networks are successfully applied [9]. In this paper we show that the generally defined troubleshooting task is NP-hard. We propose a heuristic function that exploits the conditional independence of all actions and questions given the fault of the device. It can be used as a lower bound of the expected cost of repair in heuristic algorithms searching an optimal troubleshooting strategy. In the paper we describe two such algorithms: the depth first search algorithm with pruning and the AO* algorithm.

Complexity of decision‐theoretic troubleshooting

The goal of troubleshooting is to find an optimal solution strategy consisting of actions and observations for repairing a device. We assume a probabilistic model of dependence between possible faults, actions, and observations; the goal is to minimize the expected cost of repair (ECR). We show that the task of finding an optimal solution strategy is NP hard for various troubleshooting models; therefore, approximate algorithms are necessary.

A comparison of two approaches for solving unconstrained influence diagrams

Influence diagrams and decision trees represent the two most common frameworks for specifying and solving decision problems. As modeling languages, both of these frameworks require that the decision analyst specifies all possible sequences of observations and decisions (in influence diagrams, this requirement corresponds to the constraint that the decisions should be temporarily linearly ordered). Recently, the unconstrained influence diagram was proposed to address this drawback. In this framework, we may have a partial ordering of the decisions, and a solution to the decision problem therefore consists not only of a decision policy for the various decisions, but also of a conditional specification of what to do next. Relative to the complexity of solving an influence diagram, finding a solution to an unconstrained influence diagram may be computationally very demanding w.r.t. both time and space. Hence, there is a need for efficient algorithms that can deal with (and take advantage of) the idiosyncrasies of the language. In this paper we propose two such solution algorithms. One resembles the variable elimination technique from influence diagrams, whereas the other is based on conditioning and supports any-space inference. Finally, we present an empirical comparison of the proposed methods.

An extension of lazy evaluation for influence diagrams avoiding redundant variables in the potentials

Standard methods for solving influence diagrams consist in stepwise elimination of variables, and along with elimination of a variable a set of new potentials over new domains is calculated. It is well known that these methods tend to produce unnecessarily large domains resulting in excessive consumption of time and memory. The lazy evaluation method represents only a partial solution to the problem. In this paper we extend any potential with two graphs over its domain representing the dependencies of variables. When a node A is eliminated, all necessary structural information for establishing the minimal sets of domains for potentials is contained in these graphs. We push lazy evaluation a step further to avoid performing unnecessary multiplications and subsequent division with equivalent potentials.

Color-Sketch Simulator: A Guide for Color-Based Visual Known-Item Search

In order to evaluate the effectiveness of a color-sketch retrieval system for a given multimedia database, tedious evaluations involving real users are required as users are in the center of query sketch formulation. However, without any prior knowledge about the bottlenecks of the underlying sketch-based retrieval model, the evaluations may focus on wrong settings and thus miss the desired effect. Furthermore, users have usually no clues or recommendations to draw color-sketches effectively. In this paper, we aim at a preliminary analysis to identify potential bottlenecks of a flexible color-sketch retrieval model. We present a formal framework based on position-color feature signatures, enabling comprehensive simulations of users drawing a color sketch.

Monotonization of User Preferences

We consider the problem of user-item recommendation as a multiuser instance ranking learning. A user-item preference is monotonizable if the learning can restrict to monotone models. A preference model is monotone if it is a monotone composition of rankings on domains of explanatory attributes (possibly describing user behavior, item content but also data aggregations). Target preference ordering of users on items is given by a preference indicator (e.g. purchase, rating).

Understanding Transparent and Complicated Users as Instances of Preference Learning for Recommender Systems

In this paper we are concerned with user understanding in content based recommendation. We assume having explicit ratings with time-stamps from each user. We integrate three different movie data sets, trying to avoid features specific for single data and try to be more generic. We use several metrics which were not used so far in the recommender systems domain. Besides classical rating approximation with RMSE and ratio of order agreement we study new metrics for predicting Next-k and at least 1-hit at Next-k. Using these Next-k and 1-hit we try to model display of our recommendation --- we can display k objects and hope to achieve at least one hit. We trace performance of our methods and metrics also as a distribution along each single user. We define transparent and complicated users with respect to number of methods which achieved at least one hit. We provide results of experiments with several combinations of methods, data sets and metrics along these three axes.

Basis Functions as Pivots in Space of Users Preferences

Our starting motivation is a user visiting an e-shop. E-shops usually offer conjunction of sharp filter conditions and one attribute ordering of results. We use a top-k query system where results are ordered by a multi-criterial monotone combination of soft filter conditions.