To Örebro University

Single-image super-resolution aims to recover high-frequency detail from a single low-resolution image, but practical applications often require balancing distortion against perceptual quality. Existing methods typically produce a single fixed reconstruction and offer limited test-time control over this trade-off. This paper presents DR-SCAN, a dual-branch deep residual network for single-image super-resolution in which, during test-time inference, weights can be assigned to either of the branches to dynamically steer their respective contributions to the reconstructed output. An interactive interface enables users to re-weight the shallow and deep branches at inference or run a one-click LPIPS search, to navigate the distortion–perception trade-off without retraining the model. Ablation experiments confirm that both the second branch and the channel–spatial attention that is used within the residual blocks are essential for the network for better reconstruction, while the interactive tuning routine demonstrates the practical value of post-hoc branch fusion.

Traditional tabular augmentation methods, such as SMOTE and Gaussian sampling, treat features as generic vectors, disregarding the domain-specific constraints often present in scientific tabular data. This work introduces a domain-aware augmentation approach that leverages Large Language Models (LLMs) to encode scientific knowledge through policy generation. The effectiveness of this approach is demonstrated using a case study on geochemical compositions, where data must satisfy closure constraints and exhibit intrinsic correlations that geometric interpolation methods fail to preserve. Evaluated on an imbalanced geochemical rock classification dataset, the LLM-based augmentation achieves 95.74% accuracy and a 0.9544 macro-F1 score, outperforming SMOTE, Gaussian sampling, and no-augmentation baselines while requiring fewer synthetic samples.

High-resolution imaging is essential in remote sensing and geoscience for precise environmental and geological analysis. DR-SCAN (Dual-Branch Residual Spatial and Channel Attention Networks), a neural network architecture for image super-resolution across these domains, is introduced. Evaluated on the UCMerced Land Use and DeepRock-SR datasets, DR-SCAN demonstrates adaptability to diverse remote sensing landscapes and effectiveness in resolving pore-scale geological features. Feature map visualizations highlight the model’s ability to prioritize crit-ical spatial features, enhancing interpretability for domain-specific applications.

In this paper, we present a novel approach to relational object affordance learning by leveraging the knowledge distillation paradigm, where large language models (LLMs) serve as generative teacher models. Distinct from traditional affordance learning approaches, which heavily depend on manual annotations, our approach leverages LLMs to automatically generate binary affordance labels and functional rationale explanations, grounded in object semantics and physical plausibility. This reduces the need for labor-intensive labeling while harnessing the rich semantic knowledge embedded in LLMs. To transfer this knowledge, we train a discriminative student model on the generated outputs, ensuring both predictive accuracy and semantic alignment with the teacher model. The student benefits from dual supervision; affordance labels guide classification, while rationales enhance functional understanding. Experimental results demonstrate that our generative-to-discriminative distillation method improves computational efficiency while maintaining a generalizable understanding of affordances across diverse object-object-action scenarios. 

Goal imagination in robotics is an emerging concept and involves the capability to automatically generate realistic goals, which, in turn, requires the assessment of the feasibility of transitioning from the current conditions of an initial scene to thedesired goal state. Existing research has explored the utilization of diverse image-generative models to create images depicting potential goal states based on the current state and instructions. In this paper, we illustrate the limitations of current state-of-the-art image generative models in accurately assessing the feasibility of specific actions in particular situations. Consequently, we present how integrating large language models, which possess profound knowledge of real-world objects and affordances, can enhance the performance of image-generative models in discerning plausible from implausible actions and simulating the outcomes of actions in a given context. This will be a step towards achieving the pragmatic goal of imagination in robotics.

The green area factor model is a crucial tool for conserving and creating urban greenery and ecosystem services within neighborhood land. This model serves as a valuable index, streamlining the planning, assessment, and comparison of local-scale green infrastructures. However, conventional on-site measurements of the green area factor are resource intensive. In response, this study pioneers a computational approach that integrates ecological and social dimensions to estimate the green area factor. Employing satellite remote sensing and advanced deep learning techniques, the methodology utilizes satellite orthophotos of urban areas subjected to semantic segmentation, identifying and categorizing diverse green elements. Ground truths are established through on-site measurements of green area factors and satellite orthophotos from benchmarking sites in <spacing diaeresis>Orebro, Sweden. Results reveal an 82.0% average F1-score for semantic segmentations, signifying a favourable correlation between computationally estimated and measured green area factors. The proposed methodology is potential for adapting to various urban settings. In essence, this research introduces a promising, cost-effective solution for assessing urban greenness, particularly beneficial for urban administrators and planners aiming for insightful and comprehensive green strategies in city planning.

We explore the use of various machine learning (ML) models for classifying lithologies utilizing data from X-ray fluorescence (XRF) and X-ray computed tomography (XCT). Typically, lithologies are identified over several meters, which restricts the use of ML models due to limited training data. To address this issue, we augment the original interval dataset, where lithologies are marked over extensive sections, into finer segments of 10cm, to produce a high resolution dataset with vastly increased sample size. Additionally, we examine the impact of adjacent lithologies on building a more generalized ML model. We also demonstrate that combining XRF and XCT data leads to an improved classification accuracy compared to using only XRF data, which is the common practice in current studies, or solely relying on XCT data.

We explore an attenuation and shape-based identification of euhedral pyrites in high-resolution X-ray Computed Tomography (XCT) data using deep neural networks. To deal with the scarcity of annotated data we generate a complementary training set of synthetic images. To investigate and address the domain gap between the synthetic and XCT data, several deep learning models, with and without domain adaption, are trained and compared. We find that a model trained on a small set of human annotations, while displaying over-fitting, can rival the human annotators. The unsupervised domain adaptation approaches are successful in bridging the domain gap, which significantly improves their performance. A domain-adapted model, trained on a dataset that fuses synthetic and real data, is the overall best-performing model. This highlights the possibility of using synthetic datasets for the application of deep learning in mineralogy.

Finding a balance between meeting test coverage and minimizing the testing resources is always a challenging task both in software (SW) and hardware (HW) testing. Therefore, employing machine learning (ML) techniques for test optimization purposes has received a great deal of attention. However, utilizing machine learning techniques frequently requires large volumes of valuable data to be trained. Although, the data gathering is hard and also expensive, manual data analysis takes most of the time in order to locate the source of failure once they have been produced in the so-called fault localization. Moreover, by applying ML techniques to historical production test data, relevant and irrelevant features can be found using strength association, such as correlation- and mutual information-based methods. In this paper, we use production data records of 100 units of a 5G radio product containing more than 7000 test results. The obtained results show that insightful information can be found after clustering the test results by their strength association, most linear and monotonic, which would otherwise be challenging to identify by traditional manual data analysis methods.

Zinc (Zn) is an essential element that influences many cellular functions. Depending on bioavailability, Zn can cause both deficiency and toxicity. Zn bioavailability is influenced by water hardness. Therefore, water quality analysis for health-risk assessment should consider both Zn concentration and water hardness. However, exposure media selection for traditional toxicology tests are set to defined hardness levels and do not represent the diverse water chemistry compositions observed in nature. Moreover, these tests commonly use whole organism endpoints, such as survival and reproduction, which require high numbers of test animals and are labor intensive. Gene expression stands out as a promising alternative to provide insight into molecular events that can be used for risk assessment. In this work, we apply machine learning techniques to classify the Zn concentrations and water hardness from Daphnia magna gene expression by using quantitative PCR. A method for gene ranking was explored using techniques from game theory, namely, Shapley values. The results show that standard machine learning classifiers can classify both Zn concentration and water hardness simultaneously, and that Shapley values are a versatile and useful alternative for gene ranking that can provide insight about the importance of individual genes.