检索结果-内蒙古大学图书馆

Pitfalls of Epistemic Uncertainty Quantification through Loss Minimisation 36

学校读者我要写书评

暂无评论

Pitfalls of Epistemic Uncertainty Quantification through Los...

36th Conference on Neural Information Processing Systems, NeurIPS 2022

作者： Bengs, Viktor Hüllermeier, Eyke Waegeman, Willem Germany Munich Center for Machine Learning Germany Department of Data Analysis and Mathematical Modeling Ghent University Belgium

ISBN: (纸本)9781713871088

Uncertainty quantification has received increasing attention in machine learning in the recent past. In particular, a distinction between aleatoric and epistemic uncertainty has been found useful in this regard. The latter refers to the learner's (lack of) knowledge and appears to be especially difficult to measure and quantify. In this paper, we analyse a recent proposal based on the idea of a second-order learner, which yields predictions in the form of distributions over probability distributions. While standard (first-order) learners can be trained to predict accurate probabilities, namely by minimising suitable loss functions on sample data, we show that loss minimisation does not work for second-order predictors: The loss functions proposed for inducing such predictors do not incentivise the learner to represent its epistemic uncertainty in a faithful way. © 2022 Neural information processing systems foundation. All rights reserved.

关键词： Probability distributions

Machine Learning Models and Technology for Classification of Forest on Satellite data 20

学校读者我要写书评

暂无评论

Machine Learning Models and Technology for Classification of...

20th International Conference on Smart Technologies, EUROCON 2023

作者： Salii, Yevhenii Kuzin, Volodymyr Hohol, Anton Kussul, Nataliia Yailymova, Hanna National Technical University of Ukraine 'Igor Sikorsky Kyiv Polytechnic Institute' Department of Mathematical Modeling and Data Analysis Kyiv Ukraine National University of 'Kyiv-Mohyla Academy' Department of Informatics Kyiv Ukraine National Technical University of Ukraine 'Igor Sikorsky Kyiv Polytechnic Institute' Department of Space Information Technologies and Systems Space Research Institute NASU-SSAU Department of Mathematical Modeling and Data Analysis Kyiv Ukraine

ISBN: (纸本)9781665463973

The paper deals with the problem of semantic segmentation of satellite imagery to deliver forest type map with high resolution. To solve the problem, we propose 4 machine learning models. Two of them are based on Random Forest (RF) and other two - on Convolutional Neural Network (CNN) - U-Net. As an input we use 2 images of Sentinel-2 (one for summer and one for winter, 4 spectral bands from each). As an output (labels) we use the Copernicus Forest Type dataset for 2018 year. Our models showed promising results on validation data. Of all models the one based on U-Net ended up being the most efficient in forest classification with overall accuracy 91.7%. At the same time the best RF model scored with 86.5%. After comparing the results, in order to check our model transferability we created and compared forest map of northern part of Kyiv region of 2018 and 2022. The experiment confirmed the robustness of the model and it's scalability. The developed models have been implemented in the cloud platform specialized on satellite data - CREODIAS. The developed map can provide valuable data for foresters, biologists, or other researchers to make decisions about forest management and conservation, as well as to ensure that Europe's forests are managed in an ecologically sustainable way. © 2023 IEEE.

关键词： Machine learning

Is Epistemic Uncertainty Faithfully Represented by Evidential Deep Learning Methods?

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Jürgens, Mira Meinert, Nis Bengs, Viktor Hüllermeier, Eyke Waegeman, Willem Department of Data Analysis and Mathematical Modeling Ghent University Belgium Neustrelitz Germany Germany

关键词： Contrastive Learning

ON SECOND-ORDER SCORING RULES FOR EPISTEMIC UNCERTAINTY QUANTIFICATION

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Bengs, Viktor Hüllermeier, Eyke Waegeman, Willem Munich Center for Machine Learning Germany Department of Data Analysis and Mathematical Modeling Ghent University Belgium

It is well known that accurate probabilistic predictors can be trained through empirical risk minimisation with proper scoring rules as loss functions. While such learners capture so-called aleatoric uncertainty of predictions, various machine learning methods have recently been developed with the goal to let the learner also represent its epistemic uncertainty, i.e., the uncertainty caused by a lack of knowledge and data. An emerging branch of the literature proposes the use of a second-order learner that provides predictions in terms of distributions on probability distributions. However, recent work has revealed serious theoretical shortcomings for second-order predictors based on loss minimisation. In this paper, we generalise these findings and prove a more fundamental result: There seems to be no loss function that provides an incentive for a second-order learner to faithfully represent its epistemic uncertainty in the same manner as proper scoring rules do for standard (first-order) learners. As a main mathematical tool to prove this result, we introduce the generalised notion of second-order scoring *** Codes 68T37 (Primary) 68T30 (Secondary) Copyright © 2023, The Authors. All rights reserved.

关键词： Probability distributions

A CALIBRATION TEST FOR EVALUATING SET-BASED EPISTEMIC UNCERTAINTY REPRESENTATIONS

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Jürgens, Mira Mortier, Thomas Hüllermeier, Eyke Bengs, Viktor Waegeman, Willem Department of Data Analysis and Mathematical Modeling Ghent University Belgium Department of Environment Ghent University Belgium Department of Informatics Munich Center for Machine Learning LMU Munich Germany

The accurate representation of epistemic uncertainty is a challenging yet essential task in machine learning. A widely used representation corresponds to convex sets of probabilistic predictors, also known as credal sets. One popular way of constructing these credal sets is via ensembling or specialized supervised learning methods, where the epistemic uncertainty can be quantified through measures such as the set size or the disagreement among members. In principle, these sets should contain the true data-generating distribution. As a necessary condition for this validity, we adopt the strongest notion of calibration as a proxy. Concretely, we propose a novel statistical test to determine whether there is a convex combination of the set’s predictions that is calibrated in distribution. In contrast to previous methods, our framework allows the convex combination to be instance dependent, recognizing that different ensemble members may be better calibrated in different regions of the input space. Moreover, we learn this combination via proper scoring rules, which inherently optimize for calibration. Building on differentiable, kernel-based estimators of calibration errors, we introduce a nonparametric testing procedure and demonstrate the benefits of capturing instance-level variability on of synthetic and real-world experiments. © 2025, CC BY.

关键词： Self-supervised learning

Is epistemic uncertainty faithfully represented by evidential deep learning methods? 24

学校读者我要写书评

暂无评论

Is epistemic uncertainty faithfully represented by evidentia...

Proceedings of the 41st International Conference on Machine Learning

作者： Mira Jürgens Nis Meinert Viktor Bengs Eyke Hüllermeier Willem Waegeman Department of Data Analysis and Mathematical Modeling Ghent University Belgium Institue of Communications and Navigation German Aerospace Center (DLR) Neustrelitz Germany Department of Informatics University of Munich (LMU) Germany

关键词：

PITFALLS OF EPISTEMIC UNCERTAINTY QUANTIFICATION THROUGH LOSS MINIMISATION

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Bengs, Viktor Hüllermeier, Eyke Waegeman, Willem Germany Munich Center for Machine Learning Germany Department of Data Analysis and Mathematical Modeling Ghent University Belgium

Uncertainty quantification has received increasing attention in machine learning in the recent past. In particular, a distinction between aleatoric and epistemic uncertainty has been found useful in this regard. The latter refers to the learner's (lack of) knowledge and appears to be especially difficult to measure and quantify. In this paper, we analyse a recent proposal based on the idea of a second-order learner, which yields predictions in the form of distributions over probability distributions. While standard (first-order) learners can be trained to predict accurate probabilities, namely by minimising suitable loss functions on sample data, we show that loss minimisation does not work for second-order predictors: The loss functions proposed for inducing such predictors do not incentivise the learner to represent its epistemic uncertainty in a faithful *** Codes 68T37 (Primary) 68T30 (Secondary) Copyright © 2022, The Authors. All rights reserved.

关键词： Probability distributions

Machine Learning Models and Technology for Classification of Forest on Satellite data

学校读者我要写书评

暂无评论

Machine Learning Models and Technology for Classification of...

Eurocon

作者： Yevhenii Salii Volodymyr Kuzin Anton Hohol Nataliia Kussul Hanna Yailymova Department of Mathematical Modeling and Data Analysis National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute” Kyiv Ukraine Department of Informatics National University of “Kyiv-Mohyla Academy” Kyiv Ukraine Department of Space Information Technologies and Systems Space Research Institute NASU-SSAU Department of Mathematical Modeling and Data Analysis National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute” Kyiv Ukraine

关键词：

Assessing Ukraine’s Solar Power Potential: A Comprehensive analysis Using Satellite data and Fuzzy Logic

学校读者我要写书评

暂无评论

Assessing Ukraine’s Solar Power Potential: A Comprehensive ...

IEEE International Symposium on Geoscience and Remote Sensing (IGARSS)

作者： Sofiia Drozd Nataliia Kussul Department of Mathematical Modeling and Data Analysis National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute” Kyiv Ukraine Department of Space Information Systems and Technologies Space Research Institute NASU-SSAU Kyiv Ukraine

ISBN: (数字)9798350360325

ISBN: (纸本)9798350360332

This study evaluates the land suitability for the placement of solar power stations in Ukraine, utilizing satellite data on climate factors (Global Horizontal Irradiance, temperature, precipitation, wind speed), topography (elevation, slope), and land use. Fuzzy logic, pairwise comparisons, and weighted linear combination methods were utilized to develop a high-resolution (100 m) land suitability map for the installation of solar power plants. The results show that more than half (54.5%) of Ukraine’s territory has a high suitability score (exceeding 0.65) for solar power stations, particularly in the southern and eastern regions, such as Odessa, Kherson, Mykolaiv, Zaporizhia, Donetsk, and Crimea. Only 10.68% of the land has a suitability score less than 0.6, and 18.18% is deemed absolutely unsuitable (with a score of 0, due to land cover), primarily located in the western and northern parts of the country. This indicates that Ukraine has significant potential for green energy production. The study provides an effective and useful tool for decision-making on the optimal location of solar power facilities in Ukraine.

关键词： Green energy Fuzzy logic Temperature distribution Satellites Wind speed Production Surfaces