67639 research outputs found
Sort by
Fair Text-to-Image Diffusion via Fair Mapping
No full textIn this paper, we address the limitations of existing text-to-image diffusion models in generating demographically fair results when given human-related descriptions. These models often struggle to disentangle the target language context from sociocultural biases, resulting in biased image generation. To overcome this challenge, we propose Fair Mapping, a flexible, model-agnostic, and lightweight approach that modifies a pre-trained text-to-image diffusion model by controlling the prompt to achieve fair image generation. One key advantage of our approach is its high efficiency. It only requires updating an additional linear network with few parameters at a low computational cost. By developing a linear network that maps conditioning embeddings into a debiased space, we enable the generation of relatively balanced demographic results based on the specified text condition. With comprehensive experiments on face image generation, we show that our method significantly improves image generation fairness with almost the same image quality compared to conventional diffusion models when prompted with descriptions related to humans. By effectively addressing the issue of implicit language bias, our method produces more fair and diverse image outputs.Di Wang and Lijie Hu are supported in part by the funding BAS/1/1689-01-01, URF/1/4663-01-01, REI/1/5232-01-01, REI/1/5332-01-01, and URF/1/5508-01-01 from KAUST, and funding from KAUST - Center of Excellence for Generative AI, under award number 5940. Hua Zhang and Jia Li are supported in part by the National Key Research and Development Program of China under Grant 2021YFB3100800; in part by the National Natural Science Foundation of China under Grant 62372448
A nodal ghost method based on variational formulation and regular square grid for elliptic problems on arbitrary domains in two space dimensions
No full textThis paper focuses on the numerical solution of elliptic partial differential equations (PDEs) with Dirichlet and mixed boundary conditions, specifically addressing the challenges arising from irregular domains. Both finite element method (FEM) and finite difference method (FDM), face difficulties in dealing with arbitrary domains. The paper introduces a novel nodal symmetric ghost method based on a variational formulation, which combines the advantages of FEM and FDM. The method employs bilinear finite elements on a structured mesh and provides a detailed implementation description. A rigorous a priori convergence rate analysis is also presented. The convergence rates are validated with many numerical experiments, in both one and two space dimensions.The work of C.A. and G.R. has been supported by the Spoke 10 Future AI Research (FAIR) of the Italian Research Center funded by the Ministry of University and Research as part of the National Recovery and Resilience Plan (PNRR). C.A. and G.R. have also been supported also by Italian Ministerial grant PRIN 2022 “Efficient numerical schemes and optimal control methods for time-dependent partial differential equations”, No. 2022N9BM3N - Finanziato dall'Unione europea - Next Generation EU – CUP: E53D23005830006. The work of C.A. and G.R. has also been supported by the Italian Ministerial grant PRIN 2022 PNRR “FIN4GEO: Forward and Inverse Numerical Modeling of hydrothermal systems in volcanic regions with application to geothermal energy exploitation”, No. P2022BNB97 - Finanziato dall'Unione europea - Next Generation EU – CUP: E53D23017960001. The authors are members of the Gruppo Nazionale Calcolo Scientifico-Istituto Nazionale di Alta Matematica (GNCS-INdAM)
Introduction of Non-Native Fish Species in Red Sea Aquaculture: Implications for Marine Ecosystem Integrity
No full textAquaculture is a rapidly expanding industry that holds significant potential to meet growing seafood demands and it is expected to alleviate pressures on wild stocks. The use of non-native fishes has been practiced worldwide as a strategy to enhance production and to promote financial sustainability in aquaculture. However, the introduction and cultivation of non-native species (hereafter, NNS) in aquaculture can pose severe risks to marine ecosystems, particularly in biodiverse regions like the Red Sea. This review explores insights about commercially produced species, the rationale of introducing NNS, and the potential associated risks, focusing on escapees, genetic pollution, and competition with native species, disease transmission, and habitat modification. The review also highlights the ecological consequences of such risks and proposes strategies to mitigate their impacts, emphasizing the need for comprehensive monitoring, regulatory frameworks, and sustainable aquaculture practices to safeguard marine ecosystem integrity in the region.Ministry of Environment Water and Agriculture, Riyadh, Saudi Arabia (Applied Research Support for Enhancing Fisheries Production, Initiative No. 368).The authors are thankful for the research support from the Ministry of Environment Water and Agriculture, Riyadh, Saudi Arabia (Applied Research Support for Enhancing Fisheries Production, Initiative No. 368), awarded to KAUST Beacon Development, King Abdullah University of Science and Technology, Thuwal, Jeddah, Saudi Arabia. During the preparation of this review, the authors used Chat GPT (March 2025 Version), OpenAI, San Francisco, CA, USA, free version, for the purposes of some guidelines for titles. The authors have reviewed and edited the output and take full responsibility for the content of this publication
Anorthosite Dissolution as a function of pH at 60 and 120°C: Implications for Subsurface Carbon Mineralization
No full textCarbon mineralization in reactive rocks is a promising approach for mitigating carbon dioxide (CO₂) emissions by converting CO₂ into stable carbonate minerals. Anorthosites, abundant igneous rocks composed primarily of calcium-rich plagioclase, hold significant potential for CO₂ capture and storage due to their rapid dissolution rates in acidic environments thereby promoting the formation of stable carbonate minerals. In this study, the dissolution behavior of anorthosites collected from Yanbu, Saudi Arabia, was evaluated at field-relevant conditions. Element release rates were experimentally measured in mixed-flow reactors in aqueous fluids at pH ranging from 2 to 12 and temperatures of 60 and 120°C. The results show that silicon release rates are consistent with those reported for intermediate plagioclase in the literature. A pronounced preferential initial calcium release was observed at all investigated conditions. Mass balance calculations suggest this preferential release is driven by calcium ion exchange with sodium ions and/or ammonium ions at the plagioclase surface. The preferential release of calcium continued throughout all experiments performed at pH greater than 3, where some experiments lasted up to 550 hours. The preferential release of calcium, combined with the observed rapid precipitation of aluminum-oxyhydroxide phases at near to neutral conditions, facilitates the formation of calcium carbonate minerals. Given the global abundance of anorthosites, these findings underscore their potential as host rocks for subsurface mineral carbon disposal, providing a robust and scalable solution for long-term CO₂ capture and storage.We thank King Abdullah University of Science and Technology (KAUST) for providing the resources and facilities necessary to conduct the experiments, as well as for the support from the Research Funding Office under Award No. 4357
No Mesh, No Problem: Estimating Coral Volume and Surface from Sparse Multi-View Images
No full textEffective reef monitoring requires the quantification of coral growth via accurate volumetric and surface area estimates, which is a challenging task due to the complex morphology of corals. We propose a novel, lightweight, and scalable learning framework that addresses this challenge by predicting the 3D volume and surface area of coral-like objects from 2D multi-view RGB images. Our approach utilizes a pre-trained module (VGGT) to extract dense point maps from each view; these maps are merged into a unified point cloud and enriched with per-view confidence scores. The resulting cloud is fed to two parallel DGCNN decoder heads, which jointly output the volume and the surface area of the coral, as well as their corresponding confidence estimate. To enhance prediction stability and provide uncertainty estimates, we introduce a composite loss function based on Gaussian negative log-likelihood in both real and log domains. Our method achieves competitive accuracy and generalizes well to unseen morphologies. This framework paves the way for efficient and scalable coral geometry estimation directly from a sparse set of images, with potential applications in coral growth analysis and reef monitoring
Lactate receptor HCAR1 affects axonal development and contributes to lactate's protection of axons and myelin in experimental neonatal hypoglycemia.
No full textLactate plays an important role in brain energy metabolism. It contributes to normal brain development and to neuroprotection in diabetic hypoglycemia, but its role in neonatal hypoglycemia is unclear. Moreover, lactate can work as a signaling substance via the lactate receptor HCAR1 (Hydroxycarboxylic acid receptor 1). Recent studies indicate that HCAR1 is protective in mouse models of neonatal hypoxic ischemia and has a role in metabolic regulation in glial cells during hypoglycemia. Here we have studied potential impacts of HCAR1 on axonal and myelin development in the cerebral cortex and corpus callosum of young (p21) wild type (WT) mice and HCAR1 KO mice and in cortical organotypic brain slice cultures. The HCAR1 KO mice showed lower axonal area relative to WT in both cortex and corpus callosum. However, the myelin area was unaffected by HCAR1 KO. Using particle- and colocalization analysis we show that HCAR1 KO predominantly reduces axonal size in unmyelinated axons. Using an organotypic brain slice model of neonatal hypoglycemia, we find that lactate protects both axonal and myelin development in hypoglycemia, partially via HCAR1. Lastly, live imaging with a pH-sensitive dye on acute cortical brain slices indicates that cellular lactate uptake is influenced by HCAR1. In conclusion, our findings support a role of HCAR1 in axonal development and in lactate's protective effects in hypoglycemia.Significance statement Lactate is a critical metabolite for brain energy metabolism, with established roles in neuroprotection and development. Our study provides new insights into the role of the lactate receptor HCAR1 in axonal and myelin development in the neonatal brain. We demonstrate that HCAR1 influences axonal size, particularly in unmyelinated axons, and mediates lactate's protective effects during neonatal hypoglycemia. Using in vivo and ex vivo approaches, including organotypic brain slice cultures and live imaging, we show that HCAR1 influences cellular lactate uptake and protects axonal and myelin integrity under hypoglycemic conditions. These findings highlight the dual role of lactate as an energy substrate and signaling molecule via HCAR1, with implications for understanding brain development and resilience to metabolic stress.Electron microscopy was carried out at the EM facility at the Institute of Oral Biology, UiO.
Special thanks to Yiqing Cai at the Institute of Oral Biology for assistance with tissue processing
and the preparation of ultrathin sections. Furthermore, we extend our gratitude to Krister
Andersson for help with PCR (genotyping), Jon Storm-Mathisen for valuable scientific
discussions and, Farrukh Abbas Chaudhry for the use of facilities and instruments at the Division of Anatomy, UiO. We also want to thank the group of Prof. Kåre-Olav Stensløkken for
help with glucose measurements. Lastly, we thank Hilde Galtung at the institute of Oral Biology
for lending out her cell lab for culturing organotypic brain slices
Open and FAIR data for nanofiltration in organic media: A unified approach
No full textOrganic solvent nanofiltration (OSN), also called solvent-resistant nanofiltration (SRNF), has emerged as a promising technology for the removal of impurities, recovery of solutes, and the regeneration of solvents in various industries, such as the pharmaceutical and the (petro)chemical industries. Despite the widespread use of OSN/SRNF, the presence of scattered, non-standardized data, and the absence of openly accessible data pose critical challenges to the development of new membrane materials and processes, their comparison to the state-of-the-art materials, and their fundamental understanding. To overcome these hurdles, data from peer-reviewed research articles and commercial datasheets were curated via a standardized procedure to obtain an extensive dataset on the membrane materials, synthesis parameters, operational conditions, physicochemical properties, and performance of OSN/SRNF membranes. Thanks to a truly impressive joint effort of the OSN/SRNF community, the dataset contains, as per April 2024, 5,006 unique membrane filtrations from 294 publications for 42 solvents under several process parameters. This findable, accessible, interoperable, reproducible, and open (FAIR/O) dataset is available on both the OSN Database and the newly inaugurated Open Membrane Database for SRNF (OMD4SRNF). These databases provide multiple visualization and data exploration tools. Here, the standardized procedure applied to curate the data and the functionality of the databases are outlined, as well as the online user interface to deposit new data by external users on the OMD4SRNF. This community-led project has been supported by all the co-authors of this work. Most importantly, they additionally agreed to systematically deposit their future peer-reviewed data on OSN/SRNF into the databases. We thereby pave the road for FAIR/O data in the field of OSN/SRNF to increase transparency, enable more accurate data analysis, and foster collaboration and innovation.The website and back-office development of the OMD, and by extension the OMD4SRNF, is performed with the help of Thomas Stassin and Pierre Biezemans from codefathers.be. The research reported in this publication was supported by funding from King Abdullah University of Science and Technology (KAUST) and KU Leuven. This work was supported by the Research Foundation Flanders (FWO) through R.V. junior postdoctoral fellowship (1216222N), S.C. and M.L. doctoral fellowships (1SF9825N and 1S93122N) and project grant G0C6623N. This work was supported by the Flemish government through Industrieel Onderzoeksfonds (IOF) grants (VTI-23-00181, C3/21/066 and IOFM/17/008) and by VLAIO through the RENOVATE 2 program (HBC.2022.0536). Y.R. and R.P.L. were supported by the National Science Foundation (DMREF-1921873). J.C. acknowledges the Spanish Agencia Estatal de Investigación (PID2022-138582OB-I00 through MCIN/AEI/10.13039/501100011033/ and “ERDF A way of making Europe”) as well as the Aragón Government (T68_23R). This work was partially supported by Kobe University Strategic International Collaborative Research Grant (Type B Fostering Joint Research). J.R.M. and S.B. were supported by the National Alliance for Water Innovation and the U.S. Department of Energy Office of Energy Efficiency and Renewable Energy (EERE), Industrial Efficiency and Decarbonization Office, under Funding Opportunity Announcement Number DE-FOA-0001905. K.Z. and Q.F. acknowledge grants from the National Key R&D Program of China (2023YFC3709005), the Ministry of Science and Technology, PRC, the Bureau of International Cooperation, Chinese Academy of Sciences (132C35KYSB20160018). K.Z. thanks Ocean University of China for providing Zhufeng Distinguished Professor Fellowship. M.G. gratefully acknowledges financial support from the US National Science Foundation (NSF-CBET) under the CAREER Program (#2043648). U.K. and H.S. were supported by the German Research Foundation within the Research Unit FOR5538 IMPD4Cat, grant KR2491/14-1. A.J.V. and S.S. acknowledge the Max-Planck-Society for funding. A.V. and A.Y. were supported by State Program of TIPS RAS
Advancing Geophysical Data Processing with Attention Enhanced Self-Supervised Denoisers
No full textDeep learning has shown strong potential in seismic data processing, but supervised approaches are often limited by the lack of clean, labeled data. This thesis explores self-supervised learning as a robust alternative, applying it to two geophysical challenges: seismic deblending and ambient noise correlation denoising. For seismic deblending, we enhance a blind-trace self-supervised denoiser with three attention mechanisms—Channel, Spatial, and Self-Attention—within a plug-and-play (PnP) inversion framework. Applied to 2D and 3D synthetic datasets, attention-aided models, especially Self-Attention, improve signal-to- noise ratio and enhance structural continuity in both denoising and inversion stages. In the passive setting, we apply the same models to daily ambient noise correlations from the Zafran OBS network in the Red Sea. Results show that denoising improves waveform coherence and can extend dispersion curve picks to longer periods, which is essential for ambient noise tomography. Performance varies by case, highlighting the importance of input data quality. This work demonstrates the versatility of attention-enhanced, self-supervised denoising in both active and passive seismic contexts, offering a scalable solution when ground truth is unavailable
Multi-frequency wavefield solutions for variable velocity models using meta-learning enhanced low-rank physics-informed neural network
No full textPhysics-informed neural networks (PINNs) face significant challenges in modeling multi-frequency wavefields in complex velocity models due to their slow convergence, difficulty in representing high-frequency details, and lack of generalization to varying frequencies and velocity scenarios. To address these issues, we propose Meta-LRPINN, a novel framework that combines low-rank parameterization using singular value decomposition (SVD) with meta-learning and frequency embedding. Specifically, we decompose the weights of PINN’s hidden layers using SVD and introduce an innovative frequency embedding hypernetwork (FEH) that links input frequencies with the singular values, enabling efficient and frequency-adaptive wavefield representation. Meta-learning is employed to provide robust initialization, improving optimization stability and reducing training time. Additionally, we implement adaptive rank reduction and FEH pruning during the meta-testing phase to further enhance efficiency. Numerical experiments, which are presented on multi-frequency scattered wavefields for different velocity models, demonstrate that Meta-LRPINN achieves much faster convergence speed and much higher accuracy compared to baseline methods such as Meta-PINN and vanilla PINN. Also, the proposed framework shows strong generalization to out-of-distribution frequencies while maintaining computational efficiency. These results highlight the potential of our Meta-LRPINN for scalable and adaptable seismic wavefield modeling.We thank Editor Ludovic M´etivier and the reviewers for their constructive comments and suggestions, which greatly improved the quality of the manuscript. This publication is based on work supported by the King Abdullah University of Science and Technology (KAUST). The authors thank the DeepWave sponsors for their support. This work utilized the resources of the Supercomputing Laboratory at King Abdullah University of Science and Technology (KAUST) in Thuwal, Saudi Arabia
Application of Deep Learning for Single Cell Multi-Omics: A State-of-the-Art Review
No full textSince its inception in 2009 to being highlighted as the method of the year in 2013, single cell sequencing technology has shown tremendous potential to study various omics profiles or data at an unprecedented resolution. The advances in single cell technology have led to the development of multi-omics techniques which can profile more than one modality from a single cell simultaneously. Thus, providing a significant measure of information which can be utilized to study the cell state and functions eventually the disease and health. The multi-omics profiling has led to a significant increase in production of single cell data. The single cell data is complex due to the heterogeneous nature, thus offers various challenges to deal with such largely complex data. Several computational methods have been proposed to get insights from the single cell multi-omics data. A comprehensive review describing the methods would be great step towards the growth of the field of single cell analysis. Here we provide an in-depth survey of the deep learning computational methods for single cell applications. We provide a brief history of sequencing technologies with a timeline depicting the evolution of various profiling techniques developed over the time. We identify various deep learning techniques that have been employed for single cell applications. This paper presents in-depth survey of deep learning based methods for various downstream applications such as imputation, batch effect (BE) removal, single cell integration and more. We identify various challenges and issues associated with each application which are critical to be addressed. This review will serve as a source of knowledge for new researchers aspiring to begin their research journey in building computational methods to overcome various challenges faced by the field