Catálogo de publicaciones - libros

Workflow mining algorithms are used to improve and/or refine design of existing workflows. Workflows are composed of sequential, parallel, conflict and iterative structures. In this paper we present results of experimental complexity study of the alpha workflow mining algorithm. We studied time and space complexity as dependent on workflow’s internal structure and on the number of workflow tasks.

Contemporary information systems are facing challenging tasks involving advanced data analysis, pattern discovery, and knowledge utilization. Data mining can be successfully employed to sieve through huge amounts of raw data in search for interesting patterns. Knowledge discovered during data mining activity can be used to provide value-added services to users, customers, and organizations.

The adoption of the Web as one of the main media for business-to-customer (B2C) communication provides novel opportunities for using data mining to personalize and enhance customer interfaces. In this paper we introduce the notion of recommendation rules — a simple knowledge model that can be successfully used in the Web environment to improve the quality of B2C relationship by highly personalized communication. We present the formalism and we show how to efficiently generate recommendation rules from a large body of customer data.

In the paper continuation of our research described earlier is shortly discussed. The main goal of current investigation is to build combined learning model (probably quasi-optimal) merging some knowledge models, developed by means of different machine learning algorithms. In order to reach this goal, a set of generic operations were implemented and tested on melanocytic dataset.

This paper compares the behaviour of three metaheuristics for the function optimization problem on a set of classical functions handling a lot number of variables and known to be hard. The first algorithm to be described is Particle Swarm Optimization (PSO). The second one is based on the paradigm of Artificial Immune System (AIS). Both algorithms are then compared with a Genetic Algorithm (GA). New insights on how these algorithms behave on a set of difficult objective functions with a lot number of variables are provided.

We present a language for property specification for workflows and a tool for property checks. The language is based on the Propositional Linear Temporal Logic and the structure of workflow logs. These language and tool help companies to diagnose business processes, react to changes in business environment and collect formal definitions of business properties. We give examples of specifications of business properties that set relations between events of business processes

We propose a theoretical model to realize DNA made circuits based on algorithms, to perform arithmetic and logical operations. The physical components of the resulting Arithmetic-Logic Unit are a variety of elements such as biochemical laboratories, test tubes and human operators. The advantage of the model is the possibility to perform arithmetic operations with huge binary numbers.

The field of Reinforcement Learning, a sub-field of machine learning, represents an important direction for research in Artificial Intelligence, the way for improving an agent’s behavior, given a certain feed-back about its performance. In this paper we propose an original interface for programming reinforcement learning simulations in known environments. Using this interface, there are possible simulations both for reinforcement learning based on the states’ utilities and learning based on actions’ values (Q-learning).

In this paper we present architecture of recently built experimental anomaly detection system based on the paradigm of artificial immune system and working in a network environment. We show how network traffic data are mapped into antibodies or antigens of artificial immune system and how similarities between signatures of attackers and antibodies are measured. We present an example of the work of the system in the real network environment.

In the paper the new algorithm and results of its application to designation the importance of the TDS attribute in identifying the melanocytic skin lesions are described. The algorithm consists of decision table, TDS belief net, and the nearest neighbor method applied to the bases, which was obtained by reduction the number of attributes from thirteen to four. The obtained results show that this algorithm is very promising.

Decision trees and rules are completing methods of knowledge representation. Both have advantages in some applications. Algorithms that convert trees to rules are common. In the paper an algorithm that converts rules to decision tree and its implementation in inductive database VINLEN is presented.