1. A hierarchical knowledge graph embedding framework for link predictionShuang Liu, Chengwang Hou, Jiana Meng, Peng Chen, Simon Kolmanič, 2024, original scientific article Keywords: knowledge graph embedding, knowledge graph completion, negative sampling, link prediction Published in DKUM: 04.02.2025; Views: 0; Downloads: 2
Full text (1,76 MB) |
2. Long-term temperature prediction with hybrid autoencoder algorithmsJorge Pérez-Aracil, Dušan Fister, C. M. Marina, César Peláez-Rodriguez, L. Cornejo-Bueno, P. A. Gutiérrez, Matteo Giuliani, A. Castelleti, Sancho Salcedo-Sanz, 2024, original scientific article Abstract: This paper proposes two hybrid approaches based on Autoencoders (AEs) for long-term temperature prediction. The first algorithm comprises an AE trained to learn temperature patterns, which is then linked to a second AE, used to detect possible anomalies and provide a final temperature prediction. The second proposed approach involves training an AE and then using the resulting latent space as input of a neural network, which will provide the final prediction output. Both approaches are tested in long-term air temperature prediction in European cities: seven European locations where major heat waves occurred have been considered. The longterm temperature prediction for the entire year of the heatwave events has been analysed. Results show that the proposed approaches can obtain accurate long-term (up to 4 weeks) temperature prediction, improving Persistence and Climatology in the benchmark models compared. In heatwave periods, where the persistence of the temperature is extremely high, our approach beat the persistence operator in three locations and works similarly in the rest of the cases, showing the potential of this AE-based method for long-term temperature prediction. Keywords: autoencoder, temperature prediction, hybrid models, heatwave Published in DKUM: 29.01.2025; Views: 0; Downloads: 1
Full text (1,82 MB) |
3. Deep learning predictive models for terminal call rate prediction during the warranty periodAljaž Ferencek, Davorin Kofjač, Andrej Škraba, Blaž Sašek, Mirjana Kljajić Borštnar, 2020, original scientific article Abstract: Background: This paper addresses the problem of products’ terminal call rate (TCR) prediction during the warranty period. TCR refers to the information on the amount of funds to be reserved for product repairs during the warranty period. So far, various methods have been used to address this problem, from discrete event simulation and time series, to machine learning predictive models.
Objectives: In this paper, we address the above named problem by applying deep learning models to predict terminal call rate.
Methods/Approach: We have developed a series of deep learning models on a data set obtained from a manufacturer of home appliances, and we have analysed their quality and performance.
Results: Results showed that a deep neural network with 6 layers and a convolutional neural network gave the best results.
Conclusions: This paper suggests that deep learning is an approach worth exploring further, however, with the disadvantage being that it requires large volumes of quality data. Keywords: manufacturing, product lifecycle, management product failure, machine learning, prediction Published in DKUM: 21.01.2025; Views: 0; Downloads: 2
Full text (834,28 KB) This document has many files! More... |
4. Efficient encoding and decoding of voxelized models for machine learning-based applicationsDamjan Strnad, Štefan Kohek, Borut Žalik, Libor Váša, Andrej Nerat, 2025, original scientific article Abstract: Point clouds have become a popular training data for many practical applications of machine learning in the fields of environmental modeling and precision agriculture. In order to reduce high space requirements and the effect of noise in the data, point clouds are often transformed to a structured representation such as a voxel grid. Storing, transmitting and consuming voxelized geometry, however, remains a challenging problem for machine learning pipelines running on devices with limited amount of on-chip memory with low access latency. A viable solution is to store the data in a compact encoded format, and perform on-the-fly decoding when it is needed for processing. Such on-demand expansion must be fast in order to avoid introducing substantial additional delay to the pipeline. This can be achieved by parallel decoding, which is particularly suitable for massively parallel architecture of GPUs on which the majority of machine learning is currently executed. In this paper, we present such method for efficient and parallelizable encoding/decoding of voxelized geometry. The method employs multi-level context-aware prediction of voxel occupancy based on the extracted binary feature prediction table, and encodes the residual grid with a pointerless sparse voxel octree (PSVO). We particularly focused on encoding the datasets of voxelized trees, obtained from both synthetic tree models and LiDAR point clouds of real trees. The method achieved 15.6% and 12.8% reduction of storage size with respect to plain PSVO on synthetic and real dataset, respectively. We also tested the method on a general set of diverse voxelized objects, where an average 11% improvement of storage space was achieved. Keywords: voxel grid, feature prediction, tree models, prediction-based encoding, key voxels, residuals, sparse voxel octree Published in DKUM: 09.01.2025; Views: 0; Downloads: 2
Full text (20,93 MB) |
5. A case study on entropy-aware block-based linear transforms for lossless image compressionBorut Žalik, David Podgorelec, Ivana Kolingerová, Damjan Strnad, Štefan Kohek, 2024, original scientific article Abstract: Data compression algorithms tend to reduce information entropy, which is crucial, especially in the case of images, as they are data intensive. In this regard, lossless image data compression is especially challenging. Many popular lossless compression methods incorporate predictions and various types of pixel transformations, in order to reduce the information entropy of an image. In this paper, a block optimisation programming framework is introduced to support various experiments on raster images, divided into blocks of pixels. Eleven methods were implemented within , including prediction methods, string transformation methods, and inverse distance weighting, as a representative of interpolation methods. Thirty-two different greyscale raster images with varying resolutions and contents were used in the experiments. It was shown that reduces information entropy better than the popular JPEG LS and CALIC predictors. The additional information associated with each block in is then evaluated. It was confirmed that, despite this additional cost, the estimated size in bytes is smaller in comparison to the sizes achieved by the JPEG LS and CALIC predictors. Keywords: computer science, information entropy, prediction, inverse distance transform, string transformations Published in DKUM: 07.01.2025; Views: 0; Downloads: 4
Full text (5,13 MB) |
6. Authoritative subspecies diagnosis tool for European honey bees based on ancestry informative SNPsJamal Momeni, Melanie Parejo, Rasmus O. Nielsen, Jorge Langa, Iratxe Montes, Laetitia Papoutsis, Leila Farajzadeh, Christian Brendixen, Eliza Cǎuia, Aleš Gregorc, 2021, original scientific article Abstract: Background: With numerous endemic subspecies representing four of its five evolutionary lineages, Europe holds a large fraction of Apis mellifera genetic diversity. This diversity and the natural distribution range have been altered by anthropogenic factors. The conservation of this natural heritage relies on the availability of accurate tools for subspecies diagnosis. Based on pool-sequence data from 2145 worker bees representing 22 populations sampled across Europe, we employed two highly discriminative approaches (PCA and FST) to select the most informative SNPs for ancestry inference.
Results: Using a supervised machine learning (ML) approach and a set of 3896 genotyped individuals, we could show that the 4094 selected single nucleotide polymorphisms (SNPs) provide an accurate prediction of ancestry inference in European honey bees. The best ML model was Linear Support Vector Classifier (Linear SVC) which correctly assigned most individuals to one of the 14 subspecies or different genetic origins with a mean accuracy of 96.2% ± 0.8 SD. A total of 3.8% of test individuals were misclassified, most probably due to limited differentiation between the subspecies caused by close geographical proximity, or human interference of genetic integrity of reference subspecies, or a combination thereof.
Conclusions: The diagnostic tool presented here will contribute to a sustainable conservation and support breeding activities in order to preserve the genetic heritage of European honey bees. Keywords: Apis mellifera, European suspecies, conservation, machine learning, prediction, biodiversity Published in DKUM: 01.10.2024; Views: 0; Downloads: 2
Full text (1,06 MB) This document has many files! More... |
7. Bike sharing and cable car demand forecasting using machine learning and deep learning multivariate time series approachesCésar Peláez-Rodriguez, Jorge Pérez-Aracil, Dušan Fister, Ricardo Torres- López, Sancho Salcedo-Sanz, 2024, original scientific article Keywords: cities green mobility, bike sharing demand prediction, cable car demand prediction, machine learning, deep learning Published in DKUM: 22.08.2024; Views: 76; Downloads: 5
Full text (4,33 MB) |
8. Deep learning criminal networksHaroldo V. Ribeiro, Diego D. Lopes, Arthur A. B. Pessa, Alvaro F. Martins, Bruno R. da Cunha, Sebastián Gonçalves, Ervin K. Lenzi, Quentin S. Hanley, Matjaž Perc, 2023, original scientific article Abstract: Recent advances in deep learning methods have enabled researchers to develop and apply algorithms for the analysis and modeling of complex networks. These advances have sparked a surge of interest at the interface between network science and machine learning. Despite this, the use of machine learning methods to investigate criminal networks remains surprisingly scarce. Here, we explore the potential of graph convolutional networks to learn patterns among networked criminals and to predict various properties of criminal networks. Using empirical data from political corruption, criminal police intelligence, and criminal financial networks, we develop a series of deep learning models based on the GraphSAGE framework that are able to recover missing criminal partnerships, distinguish among types of associations, predict the amount of money exchanged among criminal agents, and even anticipate partnerships and recidivism of criminals during the growth dynamics of corruption networks, all with impressive accuracy. Our deep learning models significantly outperform previous shallow learning approaches and produce high-quality embeddings for node and edge properties. Moreover, these models inherit all the advantages of the GraphSAGE framework, including the generalization to unseen nodes and scaling up to large graph structures. Keywords: organized crime, complexity, crime prediction, GraphSAGE Published in DKUM: 20.06.2024; Views: 236; Downloads: 13
Full text (2,36 MB) This document has many files! More... |
9. FLoCIC: A Few Lines of Code for Raster Image CompressionBorut Žalik, Damjan Strnad, Štefan Kohek, Ivana Kolingerová, Andrej Nerat, Niko Lukač, Bogdan Lipuš, Mitja Žalik, David Podgorelec, 2023, original scientific article Keywords: computer science, algorithm, prediction, interpolative coding, PNG, JPEG LS, JPEG 2000 lossless Published in DKUM: 22.05.2024; Views: 165; Downloads: 23
Full text (10,78 MB) This document has many files! More... |
10. Reduction of surface defects by optimization of casting speed using genetic programming : an industrial case studyMiha Kovačič, Uroš Župerl, Leo Gusel, Miran Brezočnik, 2023, original scientific article Abstract: Štore Steel Ltd. produces more than 200 different types of steel with a continuous caster installed in 2016. Several defects, mostly related to thermomechanical behaviour in the mould, originate from the continuous casting process. The same casting speed of 1.6 m/min was used for all steel grades. In May 2023, a project was launched to adjust the casting speed according to the casting temperature. This adjustment included the steel grades with the highest number of surface defects and different carbon content: 16MnCrS5, C22, 30MnVS5, and 46MnVS5. For every 10 °C deviation from the prescribed casting temperature, the speed was changed by 0.02 m/min. During the 2-month period, the ratio of rolled bars with detected surface defects (inspected by an automatic control line) decreased for the mentioned steel grades. The decreases were from 11.27 % to 7.93 %, from 12.73 % to 4.11 %, from 16.28 % to 13.40 %, and from 25.52 % to 16.99 % for 16MnCrS5, C22, 30MnVS5, and 46MnVS5, respectively. Based on the collected chemical composition and casting parameters from these two months, models were obtained using linear regression and genetic programming. These models predict the ratio of rolled bars with detected surface defects and the length of detected surface defects. According to the modelling results, the ratio of rolled bars with detected surface defects and the length of detected surface defects could be minimally reduced by 14 % and 189 %, respectively, using casting speed adjustments. A similar result was achieved from July to November 2023 by adjusting the casting speed for the other 27 types of steel. The same was predicted with the already obtained models. Genetic programming outperformed linear regression. Keywords: continuous casting of steel, surface defects, automatic control, machine learning, modelling, optimisation, prediction, linear regression, genetic programming Published in DKUM: 25.03.2024; Views: 284; Downloads: 19
Full text (1,19 MB) This document has many files! More... |