This paper presents a comparative analysis of selected deep learning methods applied to anomaly detection in data streams. The anomaly detection results obtained on the popular Yahoo! Webscope S5 dataset are used for the computational experiments. The two commonly used and recommended models in the literature, which are the basis for this analysis, are the following: the LSTM and its more complicated variant, the LSTM autoencoder. Additionally, the usefulness of an innovative LSTM-CNN approach is evaluated. The results indicate that the LSTM-CNN approach can successfully be applied for anomaly detection in data streams as its performance compares favorably with that of the two mentioned standard models. For the performance evaluation, the is used.
31 May 2018 in International Journal of Intelligent Systems
Automatic summary of databases is an important tool in strategic decision‐making. This paper applies the concept of linguistic summaries of databases to outlier detection. The definition of an outlier is closely related to the type of data analyzed and its context. Outlier detection is an important data‐mining technique, which finds applications in a wide range of domains. It can identify defects, remove impurities from the data, and, most of all, it is significant to decision‐making processes. The authors propose a novel definition of an outlier, based on linguistic quantifiers and linguistic summaries. Linguistic quantifiers are employed to express the cardinality of a set of outliers in a natural language. Thus, this paper demonstrates that linguistic summaries proposed by Yager, which provide the ability to model imprecise information, can serve as an effective tool for outlier detection.
Geometric deep learning (GDL) generalizes convolutional neural networks (CNNs) to non-Euclidean domains. In this work, a GDL technique, allowing the application of CNN on graphs, is examined. It defines convolutional filters with the use of the Gaussian mixture model (GMM). As those filters are defined in continuous space, they can be easily rotated without the need for some additional interpolation. This, in turn, allows constructing systems having rotation equivariance property. The characteristic of the proposed approach is illustrated with the problem of ear detection, which is of great importance in biometric systems enabling image based, discrete human identification. The analyzed graphs were constructed taking into account superpixels representing image content. This kind of representation has several advantages. On the one hand, it significantly reduces the amount of processed data, allowing building simpler and more effective models. On the other hand, it seems to be closer to the conscious process of human image understanding as it does not operate on millions of pixels. The contributions of the paper lie both in GDL application area extension (semantic segmentation of the images) and in the novel concept of trained filter transformations. We show that even significantly reduced information about image content and a relatively simple, in comparison with classic CNN, model (smaller number of parameters and significantly faster processing) allows obtaining detection results on the quality level similar to those reported in the literature on the UBEAR dataset. Moreover, we show experimentally that the proposed approach possesses in fact the rotation equivariance property allowing detecting rotated structures without the need for labor consuming training on all rotated and non-rotated images.
7 March 2012 in Journal of Ambient Intelligence and Humanized Computing
Semantic oriented image analysis has always been considered a challenging task, as it does not concentrate on segmentation process itself, but on interpretation of various image fragments. Contextuality of the process has recently gained significant research interest, with knowledge from image domain being repeatedly highlighted as crucial in achieving satisfactory method effectiveness. The present article elaborates on the recently described contextual hierarchical active partitions (CHAP) technique and its distributed reformulation. CHAP framework lets domain knowledge to be injected to the automated medical study analysis in a seamless and transparent manner by enabling a human expert to interactively participate in the process, e.g. by solving subtasks currently too difficult to solve by automated agents. Separation of agents makes it easy to design complex analysis algorithms from well tested and predictable components making it easy to inject human expertise at any point as needed.
1 March 2009 in International Journal of Applied Mathematics and Computer Science
The main paradigm of image understanding and a concept for its practical machine realisation are presented. The crucial elements of the presented approach are the formalisation of human knowledge about the class of images that are to be automatically interpreted, a linguistic description and the realization of cognitive resonance.