Ingo Feinerer

Computing product configurations via UML and integer linear programming

The Unified Modelling Language (UML) can be used to specify complex systems: component types are modelled as classes, interdependencies as associations with multiplicities and labels. This paper describes how to handle constraints on associations and multiplicities declaratively by translating them to inequalities over integers without adding complexity. This method provides well-defined semantics and allows for efficient algorithms for reasoning tasks like detecting inconsistencies. We identify some challenges arising from the use of class diagrams for product configuration, and propose solutions for some of them. The paper concludes with the discussion of an example derived from a real-world configuration problem in the railway domain.

On the undecidability of the equivalence of second-order tuple generating dependencies

Second-order tuple generating dependencies (SO tgds) were introduced by Fagin et al. to capture the composition of simple schema mappings. Testing the equivalence of SO tgds would be important for applications like model management and mapping optimization. However, we prove the undecidability of the logical equivalence of SO tgds. Moreover, under weak additional assumptions, we also show the undecidability of a relaxed notion of equivalence between two SO tgds, namely the so-called conjunctive query equivalence. HighlightsWe prove undecidability of logical equivalence of SO tuple generating dependencies.This result holds even for conjunctive query equivalent mappings.CQ-equivalence is shown undecidable for mappings based on SO tgds and source KDs.

Consistency and Minimality of UML Class Specifications with Multiplicities and Uniqueness Constraints

The Unified Modeling Language (UML) has become a universal tool for the formal object-oriented specification of hard- and software. In particular, UML class diagrams and so-called multiplicities, which restrict the number of links between objects, are essential when using UML for applications like the specification of admissible configurations of components. In this paper we give a formal definition of the semantics of UML class diagrams and multiplicities. We extend results obtained in the context of entity relationship diagrams to cover UML specific extensions like the (non-)uniqueness attribute of binary associations. We show that the consistency of such specifications can be checked in polynomial time, and give an algorithm for computing minimal configurations (models). The core of our approach is a translation of UML class diagrams to Diophantine inequations.

Text Clustering with String Kernels in R

We present a package which provides a general framework, including tools and algorithms, for text mining in R using the S4 class system. Using this package and the kernlab R package we explore the use of kernel methods for clustering (e.g., kernel k-means and spectral clustering) on a set of text documents, using string kernels. We compare these methods to a more traditional clustering technique like k-means on a bag of word representation of the text and evaluate the viability of kernel-based methods as a text clustering technique.

Reducing multiplicities in class diagrams

In class diagrams, so-called multiplicities are integer ranges attached to association ends. They constrain the number of instances of the associated class that an instance may be linked to, or in an alternative reading, the number of links to instances of the associated class. In complex diagrams with several chains of associations between two classes (arising e.g. in configuration management) it may happen that the lower or upper bound of a range can never be attained because of restrictions imposed by a parallel chain. In this paper we investigate how multiplicities behave when chaining associations together, and we characterise situations where intervals can be tightened due to information from other chains. Detecting and eliminating such redundancies provides valuable feedback to the user, as redundancies may hint at some underlying misconception.

A comparison of tools for teaching formal software verification

We compare four tools regarding their suitability for teaching formal software verification, namely the Frege Program Prover, the Key system, Perfect Developer, and the Prototype Verification System (PVS). We evaluate them on a suite of small programs, which are typical of courses dealing with Hoare-style verification, weakest preconditions, or dynamic logic. Finally we report our experiences with using Perfect Developer in class.

Numeric semantics of class diagrams with multiplicity and uniqueness constraints

We translate class diagrams with multiplicity constraints and uniqueness attributes to inequalities over non-negative integers. Based on this numeric semantics we check the satisfiability and consistency of class diagrams and compute minimal models. We show that this approach is efficient and provides succinct user feedback in the case of errors. In an experimental section we demonstrate that general off-the-shelf solvers for integer linear programming perform as well on real-world and synthetic benchmarks as specialised algorithms do, facilitating the extension of the formal model by further numeric constraints like cost functions. Our results are embedded in a research programme on reasoning about class diagrams and are motivated by applications in configuration management. Compared to other (for instance logic-based) approaches our aim is to hide the complexity of formal methods behind familiar user interfaces like class diagrams and to concentrate on problems that can be solved efficiently in order to be able to provide immediate feedback to users.

Efficient large-scale configuration via integer linear programming

Abstract Configuration of large-scale applications in an engineering context requires a modeling environment that allows the design engineer to draft the configuration problem in a natural way and efficient methods that can process the modeled setting and scale with the number of components. Existing configuration methods in artificial intelligence typically perform quite well in certain subareas but are hard to use for general-purpose modeling without mathematical or logics background (the so-called knowledge acquisition bottleneck) and/or have scalability issues. As a remedy to this important issue both in theory and in practical applications, we use a standard modeling environment like the Unified Modeling Language that has been proposed by the configuration community as a suitable object-oriented formalism for configuration problems. We provide a translation of key concepts of class diagrams to inequalities and identify relevant configuration aspects and show how they are treated as an integer linear program. Solving an integer linear program can be done efficiently, and integer linear programming scales well to large configurations consisting of several thousands components and interactions. We conduct an empirical study in the context of package management for operating systems and for the Linux kernel configuration. We evaluate our methodology by a benchmark and obtain convincing results in support for using integer linear programming for configuration applications of realistic size and complexity.

Searching for the structure of early American psychology: Networking Psychological Review, 1909-1923.

This study continues a previous investigation of the intellectual structure of early American psychology by presenting and analyzing 3 networks that collectively include every substantive article published in Psychological Review during the 15-year period from 1909 to 1923. The networks were laid out such that articles (represented by the networks nodes) that possessed strongly correlated vocabularies were positioned closer to each other spatially than articles with weakly correlated vocabularies. We identified distinct research communities within the networks by locating and interpreting the clusters of lexically similar articles. We found that the Psychological Review was in some turmoil during this period compared with its first 15 years attributable, first, to Baldwins unexpected departure in 1910; second, to the pressures placed on the discipline by United States entry into World War I; and, third, to the emergence of specialty psychology journals catering to research communities that had once published in the Review. The journal emerged from these challenges, however, with a better-defined mission: to serve as the chief repository of theoretical psychology in the United States.

Class Diagrams with Equated Association Chains

We investigate properties of class diagrams with multiplicity constraints - as they appear e.g. in model-based engineering or database design - augmented by equational constraints on association chains. Constraints are typically used to generate additional code that throws an exception when a constraint is violated during run-time. Our aim is different: We develop methods to check already at modelling time whether all constraints can be satisfied, to provide suitable user feedback, and to compute optimal instances of the model. In this paper we extend our approach by a family of constraints that has proven relevant in practice, namely equations between chains of associations. Such equational constraints are necessary if we want to specify that the objects reachable via one chain of associations should in fact be the same as reachable via another one.