Showing papers by "Turku Centre for Computer Science published in 2009"

Extracting Complex Biological Events with Rich Graph-Based Feature Sets

Abstract: We describe a system for extracting complex events among genes and proteins from biomedical literature, developed in context of the BioNLP'09 Shared Task on Event Extraction. For each event, its text trigger, class, and arguments are extracted. In contrast to the prevailing approaches in the domain, events can be arguments of other events, resulting in a nested structure that better captures the underlying biological statements. We divide the task into independent steps which we approach as machine learning problems. We define a wide array of features and in particular make extensive use of dependency parse graphs. A rule-based post-processing step is used to refine the output in accordance with the restrictions of the extraction task. In the shared task evaluation, the system achieved an F-score of 51.95% on the primary task, the best performance among the participants.

On the Densest MIMO Lattices From Cyclic Division Algebras

Abstract: It is shown why the discriminant of a maximal order within a cyclic division algebra must be minimized in order to get the densest possible matrix lattices with a prescribed nonvanishing minimum determinant. Using results from class field theory, a lower bound to the minimum discriminant of a maximal order with a given center and index (= the number of Tx/Rx antennas) is derived. Also numerous examples of division algebras achieving the bound are given. For example, a matrix lattice with quadrature amplitude modulation (QAM) coefficients that has 2.5 times as many codewords as the celebrated Golden code of the same minimum determinant is constructed. Also, a general algorithm due to Ivanyos and Ronyai for finding maximal orders within a cyclic division algebra is described and enhancements to this algorithm are discussed. Also some general methods for finding cyclic division algebras of a prescribed index achieving the lower bound are proposed.

An efficient algorithm for learning to rank from preference graphs

Abstract: In this paper, we introduce a framework for regularized least-squares (RLS) type of ranking cost functions and we propose three such cost functions. Further, we propose a kernel-based preference learning algorithm, which we call RankRLS, for minimizing these functions. It is shown that RankRLS has many computational advantages compared to the ranking algorithms that are based on minimizing other types of costs, such as the hinge cost. In particular, we present efficient algorithms for training, parameter selection, multiple output learning, cross-validation, and large-scale learning. Circumstances under which these computational benefits make RankRLS preferable to RankSVM are considered. We evaluate RankRLS on four different types of ranking tasks using RankSVM and the standard RLS regression as the baselines. RankRLS outperforms the standard RLS regression and its performance is very similar to that of RankSVM, while RankRLS has several computational benefits over RankSVM.

A Fuzzy Approach to Social Network Analysis

Abstract: Adjacency relations for social network analysis have usually been tackled in their bidimensional form, in the sense that relations are computed over pairs of objects. Nevertheless, this paper considers the bidimensional case as restrictive and it proposes an approach where the dimension of the analysis is not limited to binary relations. With the aid of fuzzy logic and OWA operators, it is showed that the interpretation of m-ary adjacency relations is the same of binary relations and therefore they can consistently be employed in social network analysis and some novel results be derived. Besides justifying the use of m-ary relations, the paper proposes a way to characterize them and, eventually, it will provide the reader with an example section.

Algebraic Systems and Pushdown Automata

Abstract: We concentrate in this chapter on the core aspects of algebraic series, pushdown automata, and their relation to formal languages. We choose to follow here a presentation of their theory based on the concept of properness. We introduce in Sect. 2 some auxiliary notions and results needed throughout the chapter, in particular the notions of discrete convergence in semirings and C-cycle free infinite matrices. In Sect. 3 we introduce the algebraic power series in terms of algebraic systems of equations. We focus on interconnections with context-free grammars and on normal forms. We then conclude the section with a presentation of the theorems of Shamir and Chomsky–Schutzenberger. We discuss in Sect. 4 the algebraic and the regulated rational transductions, as well as some representation results related to them. Section 5 is dedicated to pushdown automata and focuses on the interconnections with classical (non-weighted) pushdown automata and on the interconnections with algebraic systems. We then conclude the chapter with a brief discussion of some of the other topics related to algebraic systems and pushdown automata.

Least Periods of Factors of Infinite Words

Abstract: Work of the first author supported by a Discovery Grant from NSERC. Work of the second author supported by the Finnish Academy under grant 8206039.

On the normalisation of a priority vector associated with a reciprocal relation

Abstract: In this paper, we show that the widely used normalisation constraint does not apply to the priority vectors associated with reciprocal relations, also called fuzzy preference relations, whenever additive transitivity is involved. We show that misleading applications of this type of normalisation may lead to unsatisfactory results and we give some examples from the literature. Then, we propose an alternative normalisation procedure which is compatible with additive transitivity and leads to better results.

Model-Based Testing Using Scenarios and Event-B Refinements

Abstract: In this paper, we present a model-based testing approach based on user provided testing scenarios. In this approach, when a software model is refined to add or modify features, the corresponding testing scenarios are automatically refined to incorporate these changes. The test cases, to be applied on the system under test, are generated from these scenarios. We use the Event-B formalism for software models, while user scenarios are represented as Communicating Sequential Process (CSP) expressions. The presented case study demonstrates how our approach can be used to test different features of a system such as incorporated fault-tolerance mechanisms.

Ultimately constant abelian complexity of infinite words

Abstract: It is known that there are recurrent words with constant abelian complexity three, but not with constant complexity four. We prove that there are recurrent words with ultimately constant complexity c for every c.

Matrix representations, linear transformations, and kernels for disambiguation in natural language

Abstract: In the application of machine learning methods with natural language inputs, the words and their positions in the input text are some of the most important features. In this article, we introduce a framework based on a word-position matrix representation of text, linear feature transformations of the word-position matrices, and kernel functions constructed from the transformations. We consider two categories of transformations, one based on word similarities and the second on their positions, which can be applied simultaneously in the framework in an elegant way. We show how word and positional similarities obtained by applying previously proposed techniques, such as latent semantic analysis, can be incorporated as transformations in the framework. We also introduce novel ways to determine word and positional similarities. We further present efficient algorithms for computing kernel functions incorporating the transformations on the word-position matrices, and, more importantly, introduce a highly efficient method for prediction. The framework is particularly suitable to natural language disambiguation tasks where the aim is to select for a single word a particular property from a set of candidates based on the context of the word. We demonstrate the applicability of the framework to this type of tasks using context-sensitive spelling error correction on the Reuters News corpus as a model problem.

On the computation of class numbers of real abelian fields

Abstract: In this paper we give a procedure to search for prime divisors of class numbers of real abelian fields and present a table of odd primes < 10000 not dividing the degree that divide the class numbers of fields of conductor < 2000. Cohen-Lenstra heuristics allow us to conjecture that no larger prime divisors should exist. Previous computations have been largely limited to prime power conductors.

Some Undecidable Dynamical Properties for One-Dimensional Reversible Cellular Automata

Abstract: Using the fact that the tiling problem of Wang tiles is undecidable even if the given tile set is deterministic by two opposite corners, it is shown that the question whether there exists a trajectory which belongs to the given open and closed set is undecidable for one-dimensional reversible cellular automata This result holds even if the cellular automaton is mixing Furthermore, it is shown that left expansivity of a reversible cellular automaton is an undecidable property Also, the tile set construction gives yet another proof for the universality of one-dimensional reversible cellular automata

Multi network interface architectures for fault tolerant Network-on-Chip

Abstract: The topology level fault tolerance of Network-on-Chip (NoC) can be improved with multi network interface (multi-NI) architectures. Multi-NI NoC architectures are based on connecting at least two network interfaces on each core. The aim is to improve fault tolerance on the architectural level which means the delivery of packets even when there are faulty links or routers in the network. This paper presents architectures and algorithms for multi-NI NoCs. The analysis of the proposed architectures and algorithms shows that some of them improve the fault tolerance of NoC with a reasonable overhead by decreasing the average hop counts and keeping the cores connectable even in the case of faults. With a multi-NI architecture the number of successfully delivered packets has been even doubled.

Word equations with one unknown

Abstract: We consider properties of the solution set of a word equation with one unknown. We prove that the solution set of a word equation possessing infinite number of solutions is of the form (pq)*p where pq is primitive. Next, we prove that a word equation with at most four occurrences of the unknown possesses either infinitely many solutions or at most two solutions. We show that there are equations with at most four occurrences of the unknown possessing exactly two solutions. Finally, we prove that a word equation with at most 2k occurrences of the unknown possesses either infinitely many solutions or at most 8 log k + O(1) solutions. Hence, if we consider a class ek of equations with at most 2k occurrences of the unknown, then each equation in this class possesses either infinitely many solutions or O(log k) number of solutions. Our considerations allow to construct the first alphabet independent linear time algorithm for computing the solution set of an equation in a nontrivial class of equations.

Fair consistency evaluation for reciprocal relations and in group decision making

Abstract: In decision-making processes, it often occurs that the decision maker is asked to pairwise compare alternatives. His/her judgements over a set of pairs of alternatives can be collected into a matrix and some relevant properties, for instance, consistency, can be estimated. Consistency is a desirable property which implies that all the pairwise comparisons respect a principle of transitivity. So far, many indices have been proposed to estimate consistency. Nevertheless, in this paper we argue that most of these indices do not fairly evaluate this property. Then, we introduce a new consistency evaluation method and we propose to use it in group decision making problems in order to fairly weigh the decision maker's preferences according to their consistency. In our analysis, we consider two families of pairwise comparison matrices: additively reciprocal pairwise comparison matrices and multiplicatively reciprocal pairwise comparison matrices.

On the Complexity of Hmelevskii's Theorem and Satisfiability of Three Unknown Equations

Abstract: We analyze Hmelevskii's theorem, which states that the general solutions of constant-free equations on three unknowns are expressible by a finite collection of formulas of word and numerical parameters. We prove that the size of the finite representation is bounded by an exponential function on the size of the equation. We also prove that the shortest nontrivial solution of the equation, if it exists, is exponential, and that its existence can be solved in nondeterministic polynomial time.

Fault tolerant distributed routing algorithms for mesh Networks-on-Chip

Abstract: The reliability of networks-on-chip can be increased by developing routing algorithms that provide fault tolerance. Distributed routing algorithms are a class of algorithms where the routing decisions are made without knowledge of the global state of the network. The paper presents a distributed routing algorithm, which targets to maximal fault tolerance. Total of 11 different algorithms are analyzed in their fault tolerance and average hop count using a dedicated in-house C++ simulator. Five of the algorithms are variations of the proposed algorithm and the rest are references or algorithms presented in literature. The developed fully adaptive algorithm is shown to provide superior fault tolerance. The best deadlock-free algorithm is the variation based on the north-last turn model. In the class of minimal algorithms the dynamic xy algorithm provides the highest fault tolerance.

Determining directional distances between points and shorelines using sweep line technique

Abstract: Distances from points to closest shorelines in a given direction are used, for example, in some models for estimating wave exposure. Such distances, also called fetch lengths, can be determined using standard geographic information systems. However, performance may be a problem if these distances are required for a great number of study points. Two new algorithms for determining fetch lengths for study points in the same directions are presented in this paper. It is assumed that the two-dimensional map is stored in vector format, i.e. shorelines of islands and mainland are stored as polygons. The first algorithm works on a set of undirected line segments derived from the shoreline polygons. The other works on a raster representation of the map. The algorithm saves memory by postponing the rasterisation until necessary. Both of the new algorithms have superior efficiency to a previously reported algorithm when the number of study points is large.