1. Instance selection for dynamic algorithm configuration with reinforcement learning : improving generalizationCarolin Benjamins, Gjorgjina Cenikj, Ana Nikolikj, Aditya Mohan, Tome Eftimov, Marius Lindauer, 2024, published scientific conference contribution Keywords: dynamic algorithm configuration, reinforcement learning, instance selection, generalization
Published in DiRROS: 16.09.2024; Views: 653; Downloads: 151
 Full text (767,63 KB)
This document has many files! More... |
2. Generalization ability of feature-based performance prediction models : a statistical analysis across benchmarksAna Nikolikj, Ana Kostovska, Gjorgjina Cenikj, Carola Doerr, Tome Eftimov, 2024, published scientific conference contribution Abstract: This study examines the generalization ability of algorithm performance prediction models across various bench-mark suites. Comparing the statistical similarity between the problem collections with the accuracy of performance prediction models that are based on exploratory landscape analysis features, we observe that there is a positive correlation between these two measures. Specifically, when the high-dimensional feature value distributions between training and testing suites lack statistical significance, the model tends to generalize well, in the sense that the testing errors are in the same range as the training errors. Two experiments validate these findings: one involving the standard benchmark suites, the BBOB and CEC collections, and another using five collections of affine combinations of BBOB problem instances.
Keywords: meta-learning, single-objective optimization, module importance
Published in DiRROS: 16.09.2024; Views: 46; Downloads: 26
Full text (1,29 MB)
This document has many files! More... |
3. |
4. |
5. DELWAVE 1.0 : deep learning surrogate model of surface wave climate in the Adriatic BasinPeter Mlakar, Antonio Ricchi, Sandro Carniel, Davide Bonaldo, Matjaž Ličer, 2024, original scientific article Abstract: We propose a new point-prediction model, the DEep Learning WAVe Emulating model (DELWAVE), which successfully emulates the behaviour of a numerical surface ocean wave model (Simulating WAves Nearshore, SWAN) at a sparse set of locations, thus enabling numerically cheap large-ensemble prediction over synoptic to climate timescales. DELWAVE was trained on COSMO-CLM (Climate Limited-area Model) and SWAN input data during the period of 1971–1998, tested during 1998–2000, and cross-evaluated over the far-future climate time window of 2071–2100. It is constructed from a convolutional atmospheric encoder block, followed by a temporal collapse block and, finally, a regression block. DELWAVE reproduces SWAN model significant wave heights with a mean absolute error (MAE) of between 5 and 10 cm, mean wave directions with a MAE of 10–25°, and a mean wave period with a MAE of 0.2 s. DELWAVE is able to accurately emulate multi-modal mean wave direction distributions related to dominant wind regimes in the basin. We use wave power analysis from linearised wave theory to explain prediction errors in the long-period limit during southeasterly conditions. We present a storm analysis of DELWAVE, employing threshold-based metrics of precision and recall to show that DELWAVE reaches a very high score (both metrics over 95 %) of storm detection. SWAN and DELWAVE time series are compared to each other in the end-of-century scenario (2071–2100) and compared to the control conditions in the 1971–2000 period. Good agreement between DELWAVE and SWAN is found when considering climatological statistics, with a small (≤ 5 %), though systematic, underestimate of 99th-percentile values. Compared to control climatology over all wind directions, the mismatch between DELWAVE and SWAN is generally small compared to the difference between scenario and control conditions, suggesting that the noise introduced by surrogate modelling is substantially weaker than the climate change signal.
Keywords: surrogate modelling, deep learning, DEep Learning WAVe Emulating model, DELWAVE, Simulating WAves Nearshore, SWAN
Published in DiRROS: 05.08.2024; Views: 190; Downloads: 135
Full text (9,04 MB)
This document has many files! More... |
6. Detection and localization of hyperfunctioning parathyroid glands on [18F]fluorocholine PET/CT using deep learning – model performance and comparison to human expertsLeon Jarabek, Jan Jamšek, Anka Cuderman, Sebastijan Rep, Marko Hočevar, Tomaž Kocjan, Mojca Jensterle Sever, Žiga Špiclin, Žiga Maček Ležaić, Filip Cvetko, Luka Ležaič, 2022, original scientific article Abstract: In the setting of primary hyperparathyroidism (PHPT), [18F]fluorocholine PET/CT (FCH-PET) has excellent diagnostic performance, with experienced practitioners achieving 97.7% accuracy in localising hyperfunctioning parathyroid tissue (HPTT). Due to the relative triviality of the task for human readers, we explored the performance of deep learning (DL) methods for HPTT detection and localisation on FCH-PET images in the setting of PHPT. Patients and methods. We used a dataset of 93 subjects with PHPT imaged using FCH-PET, of which 74 subjects had visible HPTT while 19 controls had no visible HPTT on FCH-PET. A conventional Resnet10 as well as a novel mPETResnet10 DL model were trained and tested to detect (present, not present) and localise (upper left, lower left, upper right or lower right) HPTT. Our mPETResnet10 architecture also contained a region-of-interest masking algorithm that we evaluated qualitatively in order to try to explain the model’s decision process. Results. The models detected the presence of HPTT with an accuracy of 83% and determined the quadrant of HPTT with an accuracy of 74%. The DL methods performed statistically worse (p < 0.001) in both tasks compared to human readers, who localise HPTT with the accuracy of 97.7%. The produced region-of-interest mask, while not showing a consistent added value in the qualitative evaluation of model’s decision process, had correctly identified the foreground PET signal. Conclusions. Our experiment is the first reported use of DL analysis of FCH-PET in PHPT. We have shown that it is possible to utilize DL methods with FCH-PET to detect and localize HPTT. Given our small dataset of 93 subjects, results are nevertheless promising for further research
Keywords: primary hyperparathyroidism, deep learning, nuclear medicine
Published in DiRROS: 25.07.2024; Views: 318; Downloads: 105
Full text (810,33 KB)
This document has many files! More... |
7. A ǂFramework for applying data-driven AI/ML models in reliabilityRok Hribar, Margarita Antoniou, Gregor Papa, 2024, independent scientific component part or a chapter in a monograph Abstract: In this chapter, we present a framework for applying artificial intelligence (AI)/machine learning (ML) in reliability, in the context of the iRel40 project. Data-driven models are becoming an increasingly fruitful tool for detecting patterns in complex data and identifying the circumstances in which they occur. Using only data, gathered along the value chain, data-driven methods are now being used to detect indications of potential early failures, signs of wear out or degradation, and other unwanted events within the development, fabrication, or service phases of the electronic components and systems. We present general considerations that were found to be important during the iRel40 project, when designing pipelines that combine data processing with the AI/ML models for predicting or detecting reliability issues. This chapter serves as an introduction to the definitions and concepts used within the specific use cases that rely on the AI/ML methodology within the iRel40 project.
Keywords: machine learning, artificial intelligence, data-driven models
Published in DiRROS: 23.07.2024; Views: 137; Downloads: 57
 Link to file |
8. Reliability improvements for in-wheel motorGašper Petelin, Rok Hribar, Stane Ciglarič, Jernej Herman, Anton Biasizzo, Peter Korošec, Gregor Papa, 2024, independent scientific component part or a chapter in a monograph Abstract: Setting up a reliable electric propulsion system in the automotive sector requires an intelligent condition monitoring device capable of reliably assessing the state and the health of the electric motor. To allow for a massive integration of such monitoring devices, they must be inexpensive and small. These requirements limit their accuracy. However, we show in this chapter that these limitations can be significantly reduced by appropriate processing of the sensor data. We have used machine learning models (random forest and XGBoost) to transform very noisy motor winding insulation resistance measurements made by a low-cost device into a much more reliable value that can compete with measurements made by a high-priced state-of-the-art measurement system. The proposed method is an important building block for a future smart condition monitoring system and enables a cost-effective and accurate assessment of the condition of electric motor health in connection with the condition of their winding insulation.
Keywords: machine learning models, low-cost device, electric motor
Published in DiRROS: 23.07.2024; Views: 173; Downloads: 77
Link to file |
9. Comparison of in-situ chlorophyll-a time series and sentinel-3 ocean and land color instrument data in Slovenian national waters (Gulf of Trieste, Adriatic Sea)El Khalil Cherif, Patricija Mozetič, Janja Francé, Vesna Flander-Putrle, Jana Faganeli Pucer, Martin Vodopivec, 2021, original scientific article Abstract: While satellite remote sensing of ocean color is a viable tool for estimating large-scale patterns of chlorophyll-a (Chl-a) and global ocean primary production, its application in coastal waters is limited by the complex optical properties. An exploratory study was conducted in the Gulf of Trieste (Adriatic Sea) to assess the usefulness of Sentinel-3 satellite data in the Slovenian national waters. OLCI (Ocean and Land Colour Instrument) Chl-a level 2 products (OC4Me and NN) were compared to monthly Chl-a in-situ measurements at fixed sites from 2017 to 2019. In addition, eight other methods for estimating Chl-a concentration based on reflectance in different spectral bands were tested (OC3M, OC4E, MedOC4, ADOC4, AD4, 3B-OLCI, 2B-OLCI and G2B). For some of these methods, calibration was performed on in-situ data to achieve a better agreement. Finally, L1-regularized regression and random forest were trained on the available dataset to test the capabilities of the machine learning approach. The results show rather poor performance of the two originally available products. The same is true for the other eight methods and the fits to the measured values also show only marginal improvement. The best results are obtained with the blue-green methods (OC3, OC4 and AD4), especially the AD4SI (a designated fit of AD4) with R = 0.56 and RMSE = 0.4 mg/m³, while the near infrared (NIR) methods show underwhelming performance. The machine learning approach can only explain 30% of the variability and the RMSE is of the same order as for the blue-green methods. We conclude that due to the low Chl-a concentration and the moderate turbidity of the seawater, the reflectance provided by the Sentinel-3 OLCI spectrometer carries little information about Chl-a in the Slovenian national waters within the Gulf of Trieste and is therefore of limited use for our purposes. This requires that we continue to improve satellite products for use in those marine waters that have not yet proven suitable. In this way, satellite data could be effectively integrated into a comprehensive network that would allow a reliable assessment of ecological status, taking into account environmental regulations.
Keywords: hydrobiology, coastal waters, Gulf of Trieste, chlorophyll-a, Sentinel-3, OLCI, machine learning
Published in DiRROS: 19.07.2024; Views: 206; Downloads: 625
Full text (5,77 MB)
This document has many files! More... |
10. Learning deep representations of enzyme thermal adaptationGang Li, Filip Buric, Jan Zrimec, Sandra Viknander, Jens Nielsen, Aleksej Zelezniak, Martin K. M. Engqvist, 2022, original scientific article Abstract: Temperature is a fundamental environmental factor that shapes the evolution of organisms. Learning thermal determinants of protein sequences in evolution thus has profound significance for basic biology, drug discovery, and protein engineering. Here, we use a data set of over 3 million BRENDA enzymes labeled with optimal growth temperatures (OGTs) of their source organisms to train a deep neural network model (DeepET). The protein-temperature representations learned by DeepET provide a temperature-related statistical summary of protein sequences and capture structural properties that affect thermal stability. For prediction of enzyme optimal catalytic temperatures and protein melting temperatures via a transfer learning approach, our DeepET model outperforms classical regression models trained on rationally designed features and other deep-learning-based representations. DeepET thus holds promise for understanding enzyme thermal adaptation and guiding the engineering of thermostable enzymes.
Keywords: bioinformatics, deep neural networks, enzyme catalytic temperatures, optimal growth temperatures, protein thermostability, transfer learning
Published in DiRROS: 17.07.2024; Views: 245; Downloads: 169
Full text (2,61 MB)
This document has many files! More... |
