Cc Carmen Bratosin

Distributed genetic process mining

Process mining aims at discovering process models from data logs in order to offer insight into the real use of information systems. Most of the existing process mining algorithms fail to discover complex constructs or have problems dealing with noise and infrequent behavior. The genetic process mining algorithm overcomes these issues by using genetic operators to search for the fittest solution in the space of all possible process models. The main disadvantage of genetic process mining is the required computation time. In this paper we present a coarse-grained distributed variant of the genetic miner that reduces the computation time. The degree of the improvement obtained highly depends on the parameter values and event logs characteristics. We perform an empirical evaluation to determine guidelines for setting the parameters of the distributed algorithm.

A Reference Model for Grid Architectures and Its Analysis

Computing and data intensive applications in physics, medicine, biology, graphics, and business intelligence require large and distributed infrastructures to address todays and tomorrows challenges. For example, process mining applications are faced with terrabytes of event data and computationally expensive algorithms. Increasingly, computer grids are used to deal with such challenges. However, despite the availability of many software packages for grid applications, a good conceptual model of the grid is missing. Grid computing is often approached in an ad-hoc and engineering-like manner. This paper provides formal description of the grid in terms of a colored Petri net (CPN). The CPN can be seen as a reference model for grids and clarifies the basic concepts at a conceptual level. Moreover, the CPN allows for various kinds of analysis ranging from verification to performance analysis. In this paper, we show that our reference model allows for the analysis of various distribution strategies using simulation.

Discovering process models with genetic algorithms using sampling

Process mining, a new business intelligence area, aims at discovering process models from event logs. Complex constructs, noise and infrequent behavior are issues that make process mining a complex problem. A genetic mining algorithm, which applies genetic operators to search in the space of all possible process models, deals with the aforementioned challenges with success. Its drawback is high computation time due to the high time costs of the fitness evaluation. Fitness evaluation time linearly depends on the number of process instances in the log. By using a sampling-based approach, i.e. evaluating fitness on a sample from the log instead of the whole log, we drastically reduce the computation time. When the desired fitness is achieved on the sample, we check the fitness on the whole log; if it is not achieved yet, we increase the sample size and continue the computation iteratively. Our experiments show that sampling works well even for relatively small logs, and the total computation time is reduced by 6 up to 15 times.

Distributed genetic process mining using sampling

Process mining aims at discovering process models from event logs. Complex constructs, noise and infrequent behavior are issues that make process mining a complex problem. A genetic mining algorithm, which applies genetic operators to search in the space of all possible process models, can successfully deal with the aforementioned challenges. In this paper, we reduce the computation time by using a distributed setting. The population is distributed between the islands of a computer network (e.g. a grid). To further accelerate the method we use sample-based fitness evaluations, i.e. we evaluate the individuals on a sample of the event log instead of the entire event log, gradually increasing the sample size if necessary. Our experiments show that both sampling and distributing the event log significantly improve the performance. The actual speed-up is highly dependent of the combination of the population size and sample size.

Evaluating a Data Removal Strategy for Grid Environments Using Colored Petri Nets

We use Colored Petri Nets (CPNs) for the modeling and performance analysis of grid architectures. We define a strategy for the optimization of grid storage usage, based on the addition of data removal tasks to grid workflows. We evaluate the strategy by simulating our CPN model of the grid. Experiments show that the strategy significantly reduces the amount of storage space needed to execute a grid application.