In this paper we identify two types of problems with excessive feature sharing and the lack of discriminative learning in hierarchical compositional models: (a) similar category misclassifications and (b) phantom detections in background objects. We propose to overcome those issues by fully utilizing a discriminative features already present in the generative models of hierarchical compositions. We introduce descriptor called Histogram of Compositions to capture the information important for improving discriminative power and use it with a classifier to learn distinctive features important for successful discrimination. The generative model of hierarchical compositions is combined with the discriminative descriptor by performing hypothesis verification of detections produced by the hierarchical compositional model. We evaluate proposed descriptor on five datasets and show to improve the misclassification rate between similar categories as well as the misclassification rate of phantom detections on backgrounds. Additionally, we compare our approach against a state-of-the-art convolutional neural network and show to outperform it under significant occlusions.
COBISS.SI-ID: 1536363971
The node set of a two-mode network consists of two disjoint subsets and all its links are linking these two subsets. The links can be weighted. We developed a new method for identifying important subnetworks in two-mode networks. The method combines and extends the ideas from generalized cores in one-mode networks and from $(p, q)$-cores for two-mode networks. In this paper we introduce the notion of generalized two-mode cores and discuss some of their properties. An efficient algorithm to determine generalized two-mode cores and an analysis of its complexity are also presented. For illustration some results obtained in analyses of real-life data are presented.
COBISS.SI-ID: 17369177
This paper investigates the dust lifting phenomenon at the coal and iron ore stockpile at the Port of Koper, Slovenia. Dust lifting presents a serious environment and health hazard, thus the main objective of our study was to propose efficient measures for wind erosion reduction. A numerical model of the stockpile was created using the computational fluid dynamics (CFD) software. Wind velocity fields above the piles were calculated for the current stockpile layout and for several modified cases with rearranged piles and added porous barriers. Results from numerical modelling were used in the USEPA model to determine dust emission factors. Comparison of selected cases shows a positive, although limited effect of porous fences and barriers on reduction of local velocities and consequently, dust erosion rate. On the other hand, pile rearrangement has little effect and may not be practical for implementation. Realistic stockpile geometry with adjacent structures has shown to cause wind velocity distributions, which are not consistent with the most studied cases of regular pile shapes and layouts. The angle of incoming wind is a key factor influencing effectiveness of both solid and porous windbarriers. The proposed placement of porous barriers between the piles has shown to be effective in reducing wind exposure and dust emission. To maintain the dust emissions at an acceptably low level, other measures such as spraying with water and crust-forming liquids may still be necessary at higher wind velocities.
COBISS.SI-ID: 14317339
Hierarchical network clustering is an approach to find tightly and internally connected clusters (groups or communities) of nodes in a network based on its structure. Instead of nodes, it is possible to cluster links of the network. The sets of nodes belonging to clusters of links can overlap. While overlapping clusters of nodes are not always expected, they are natural in many applications. Using appropriate dissimilarity measures, we can complement the clustering strategy to consider, for example, the semanticmeaning of links or nodes based on their properties. We propose a new hierarchical link clustering algorithm which in comparison to existing algorithms considers node and/or link properties (descriptions, attributes) of the input network alongside its structure using monotonic dissimilarity measures. The algorithm determines communities that form connected subnetworks (relational constraint) containing locally similar nodes with respect to their description. It is only implicitly based on the corresponding line graph of the input network, thus reducing its space and time complexities. We investigate both complexities analytically and statistically. Using provided dissimilarity measures, our algorithm can, in addition to the general overlapping community structure of input networks, uncover also related subregions inside these communities in a form of hierarchy. We demonstrate this ability on real-world and artificial network examples.
COBISS.SI-ID: 17567833
Heavy duty vehicles (HDVs), such as those used for winter services (e.g. snow plowing), are responsible for around 25 % of CO2 emissions caused by road transportation. In this paper an optimization approach developed for the fleet management of winter services is presented, which was also evaluated in city of Maribor, Slovenia. The algorithm used is based on a mathematical graph theory, specifically on a solution to the Chinese postman problem, using two types of optimization: 1) optimization of the whole plan of the winter service and 2) optimization of the completed part of the planned service. The results of the optimization suggested a reduction of route length by 28.3 % when executing the whole plan. When considering a redistribution of the salting material at three other locations, our estimation has showed that additional savings up to 9.8 % could be achieved regarding to the selection of those locations. Less routes travelled consequently leads to a reduction of resource usage (diesel) and lower CO2 emissions. When executing 15 plans in a single winter service campaign, 210 L less diesel was used, resulting in 1.362 t less CO2 emissions per campaign, representing a reduction of 30 %.