Institution profile

Umeå University

Academic institutioneurope · se
Official website
Research library215linked papers
Opportunities0open roles
Selected work

Representative Papers

Probabilistic NDVI Forecasting from Sparse Satellite Time Series and Weather Covariates

Feb 04, 2026arXiv.org

This study addresses the challenges of sparse and irregular satellite NDVI observations caused by cloud cover and the difficulty of short-term forecasting of crop vegetation dynamics under heterogeneous climatic conditions. The authors propose a probabilistic forecasting framework that employs a deep learning architecture to separately encode historical NDVI and meteorological observations along with future exogenous covariates, fusing multimodal information for multi-step quantile prediction. A novel temporally distance-weighted quantile loss function is introduced, complemented by feature engineering that incorporates both cumulative and extreme weather metrics, effectively capturing the delayed vegetation response to meteorological drivers and temporal uncertainty. Experiments on European satellite data demonstrate that the proposed method outperforms existing statistical, deep learning, and time series baselines in both point and probabilistic forecasting metrics, with ablation studies confirming historical NDVI as the dominant predictor and meteorological covariates providing significant performance gains.

1 citationsRead paper

State of the Art of LLM-Enabled Interaction with Visualization

Jan 21, 2026

This study addresses key challenges in the integration of large language models (LLMs) with visualizations—namely, difficulties in multimodal fusion, limited spatial reasoning capabilities, and the absence of standardized evaluation frameworks. Guided by PRISMA guidelines, the authors conduct a systematic literature review of 48 relevant studies and propose a novel six-dimensional taxonomy encompassing application domains, visualization tasks, interaction modalities, and more. This work presents the first comprehensive classification system specifically tailored to LLM–visualization interaction, elucidating prevalent integration patterns of LLMs in data querying, generation, explanation, and navigation. It further synthesizes dominant design paradigms and identifies critical research gaps, particularly concerning accessibility and contextual understanding, thereby establishing a theoretical foundation and future directions for evaluating and developing intelligent, conversational visualization systems.

1 citationsRead paper

Tensor Algebra Processing Primitives (TAPP): Towards a Standard for Tensor Operations

Jan 12, 2026

The absence of a unified tensor operation interface leads to tight coupling between applications and hardware, hindering performance portability and dependency management. This work proposes TAPP—a C-based Tensor Algebra Processing Primitives interface—that decouples applications from underlying implementations through formal modeling of tensor contraction operations. TAPP represents the first community-driven standard for general-purpose tensor operations, jointly endorsed by academia and industry. It provides a reference implementation and has been successfully integrated into widely used libraries and quantum chemistry software such as TBLIS, cuTENSOR, and DIRAC, demonstrating its feasibility, cross-platform portability, and effectiveness in simplifying software ecosystems.

1 citationsRead paper

Domain Adaptation of Carotid Ultrasound Images using Generative Adversarial Network

Jan 04, 2026arXiv.org

This work addresses the challenge of domain distribution discrepancies in carotid ultrasound images arising from different imaging devices. To mitigate this issue, the authors propose a novel generative adversarial network (GAN) architecture that formulates domain adaptation as an image-to-image translation task. The method simultaneously achieves texture transfer and reverberation noise suppression while preserving anatomical structures. Evaluated on two three-domain carotid ultrasound datasets, the approach substantially outperforms existing techniques such as CycleGAN, significantly enhancing cross-domain consistency. Quantitative results demonstrate high histogram correlation coefficients of 0.960 and 0.920, along with reduced Bhattacharyya distances of 0.040 and 0.085, thereby eliminating performance degradation on new devices and avoiding the need for costly retraining.

1 citationsRead paper

Harvesting energy consumption on European HPC systems: Sharing Experience from the CEEC project

Nov 04, 2025

Addressing energy-efficiency bottlenecks in European exascale HPC systems, this study conducts empirical power measurements and optimizations for representative CFD applications—waLBerla, FLEXI/GALÆXI, Neko, and NekRS—across heterogeneous platforms including LUMI, MareNostrum5, MeluXina, and JUWELS Booster. Methodologically, we propose an accelerator-native adaptation framework coupled with a mixed-precision (FP16/FP32) co-optimization strategy, and establish a unified cross-platform energy-efficiency evaluation metric. Experimental results demonstrate that GPU acceleration combined with judicious mixed-precision arithmetic achieves 30–50% energy reduction while preserving numerical accuracy, thereby significantly improving the energy efficiency (FLOPS/W). Our contributions include a reproducible, portable energy-optimization paradigm for HPC, validated across diverse architectures and application kernels. This work advances sustainable exascale computing by bridging algorithmic, hardware, and system-level energy-aware design principles.

1 citationsRead paper
Recent publications

Latest Papers