Article
Computer Science, Artificial Intelligence
Max-Heinrich Laves, Malte Tolle, Alexander Schlaefer, Sandy Engelhardt
Summary: POTOBIM is an approach to inverse problems in medical imaging that optimizes both the prior distribution and posterior temperature using Bayesian optimization. This leads to improved reconstruction accuracy and uncertainty estimation.
MEDICAL IMAGE ANALYSIS
(2022)
Editorial Material
Environmental Sciences
Grey S. Nearing, Frederik Kratzert, Alden Keefe Sampson, Craig S. Pelissier, Daniel Klotz, Jonathan M. Frame, Cristina Prieto, Hoshin V. Gupta
Summary: This paper is derived from a keynote talk given at Google's 2020 Flood Forecasting Meets Machine Learning Workshop. Recent experiments applying deep learning to rainfall-runoff simulation show that there is more information in large-scale hydrological data sets than previously thought. The paper calls for the hydrology community to focus on developing a quantitative understanding of the value of hydrological process understanding in a modeling discipline increasingly dominated by machine learning.
WATER RESOURCES RESEARCH
(2021)
Article
Robotics
Rika Antonova, Jingyun Yang, Priya Sundaresan, Dieter Fox, Fabio Ramos, Jeannette Bohg
Summary: In this research, the problem of inferring simulation parameters for deformable objects is addressed. By defining the state space of an object and embedding noisy keypoints extracted from images into a distribution, the motion of the object in different images can be represented as a trajectory of distribution embeddings in the deformable state space. This approach allows for incorporating noisy state observations into Bayesian inference tools and estimating posterior distributions over simulation parameters for highly deformable objects.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Biochemical Research Methods
Theophile Sanchez, Erik Madison Bray, Pierre Jobic, Jeremy Guez, Anne-Catherine Letournel, Guillaume Charpiat, Jean Cury, Flora Jay
Summary: dnadna is a flexible Python-based software for deep learning inference in population genetics. It provides user-friendly workflows for implementing new architectures and tasks, and allows re-optimization of implemented networks. Users can apply pre-trained networks to predict evolutionary history without extensive knowledge in deep learning or coding.
Article
Engineering, Civil
Yingying Yao, Yufeng Zhao, Xin Li, Dapeng Feng, Chaopeng Shen, Chuankun Liu, Xingxing Kuang, Chunmiao Zheng
Summary: The study shows that using deep learning technology for hydrological predictions in the regions around the Tibetan Plateau can effectively improve the accuracy of flow predictions. For DL models, the influence of flow data with different temporal resolutions on predictions is minimal, and the model's performance depends on the number of flow observations and the hydrological characteristics of the catchment. Climate forcing data are the main determining factor for flow prediction performance, while other factors have less significant impacts.
JOURNAL OF HYDROLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Miguel Vasco, Hang Yin, Francisco S. Melo, Ana Paiva
Summary: This work presents a novel hierarchical generative model NEXUS for cross-modality inference and introduces a benchmark dataset MHD for evaluation. Results show that NEXUS outperforms current state-of-the-art multimodal generative models in cross-modality inference capabilities.
Article
Robotics
Javier Rodriguez-Puigvert, Ruben Martinez-Cantin, Javier Civera
Summary: Uncertainty quantification is crucial for robotic perception. In this letter, we evaluate two scalable approaches for uncertainty quantification in single-view supervised depth learning: MC dropout and deep ensembles. We find that adding dropout in all layers of the encoder leads to better results and has a lower memory footprint compared to deep ensembles.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Engineering, Civil
Zhendong Zhang, Haihua Tang, Hui Qin, Bin Luo, Chao Zhou, Huayan Zhou
Summary: The demand for more accurate simulation of physical river basin raises higher requirements for hydrological forecast stations, variables, accuracy, period, and uncertainty. Therefore, obtaining multi-step ahead probabilistic forecasting of multiple hydrological variables for multiple stations is the key issue in this study.
JOURNAL OF HYDROLOGY
(2023)
Article
Computer Science, Information Systems
Emna Baccour, Naram Mhaisen, Alaa Awad Abdellatif, Aiman Erbad, Amr Mohamed, Mounir Hamdi, Mohsen Guizani
Summary: Artificial intelligence has made significant breakthroughs in IoT applications and services, and the emergence of pervasive AI has expanded the role of ubiquitous IoT systems, from mainly data collection to executing distributed computations. This paper provides a comprehensive survey of the latest techniques and strategies for overcoming resource challenges in pervasive AI systems.
IEEE COMMUNICATIONS SURVEYS AND TUTORIALS
(2022)
Article
Engineering, Marine
Lei Lei, Tang Tengfei, Yang Gang, Guo Jing
Summary: This paper proposes a hierarchical neural network-based hierarchical perception model for the underwater glider to perceive the hydrological information in the ocean environment. The model employs a one-dimensional convolutional neural network and long short-term memory network to detect and predict the thermocline layer and deep layer information, and conducts finite element analysis to explore the buoyancy loss of the underwater glider in the ocean.
Article
Computer Science, Interdisciplinary Applications
Pengpeng Li, Jiping Liu, An Luo, Yong Wang, Jun Zhu, Shenghua Xu
Summary: This study developed a multisource POI matching method based on deep learning, achieving high precision in POI matching through the use of different Chinese word segmentation methods and neural network models, outperforming traditional methods.
COMPUTERS ENVIRONMENT AND URBAN SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Youru Li, Xiaobo Guo, Wenfang Lin, Mingjie Zhong, Qunwei Li, Zhongyi Liu, Wenliang Zhong, Zhenfeng Zhu
Summary: Despite the better performance of path-based and embedding-based models with knowledge graphs in recommendation systems compared to other deep learning methods, there is still limited improvement due to the lack of modeling users' dynamic interests. To address this, the authors propose a multi-granularity dynamic interest sequence learning method that utilizes knowledge-enhanced path mining and interest fluctuation signal discovery to obtain semantic-enhanced paths. The paths are then embedded using SEP2Vec and merged through an entropy-aware pooling layer to obtain user preference representation for learning dynamic user interest sequences.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Yunqi Wang, Furui Liu, Zhitang Chen, Yik-Chung Wu, Jianye Hao, Guangyong Chen, Pheng-Ann Heng
Summary: Domain generalization aims to learn knowledge invariant across different distributions while semantically meaningful for downstream tasks from multiple source domains, to improve the model's generalization ability on unseen target domains. The proposed method introduces invariance of the mechanisms into the learning process, which enforces stability of the causal prediction by the classifier across domains. Experiments on several benchmark datasets demonstrate the performance of the proposed method against SOTAs.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Chemistry, Multidisciplinary
Kejian Liu, Wei Wang, Rongju Wang, Xuran Cui, Liying Zhang, Xianzhi Yuan, Xianyong Li
Summary: This study proposes a lightweight, plug-and-play interest enhancement module that fuses interest vectors from two independent models to address the modeling of long- and short-term user interests. By analyzing the dataset, deviations in the recommendation performance of long- and short-term interest models are identified and compensated for through feature enhancement and loss correction during training. The proposed module combines and compares multiple independent long-term and short-term interest models on multiple domain datasets, achieving outstanding performance in challenging recommendation scenarios.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Information Systems
Mingxin Gan, Hang Zhang
Summary: Learning representations of user interests and item characteristics is crucial for recommendation tasks. Existing graph neural network-based methods have limitations in modeling complex user interests and incorporating attribute-related information. To address these limitations, we propose a variational inference-based graph autoencoder (VIGA) model that explores a multivariate distribution over latent representations for recommendation.
INFORMATION SCIENCES
(2023)
Article
Environmental Sciences
Camille Vautier, Tamara Kolbe, Tristan Babey, Jean Marcais, Benjamin W. Abbott, Anniet M. Laverman, Zahra Thomas, Luc Aquilina, Gilles Pinay, Jean-Raynald de Dreuzy
Summary: The study indicates that predicting groundwater nitrate concentration can be effectively approximated with a limited number of key parameters, while historical nitrogen input time series can be simplified without substantially altering predictions. Mean and standard deviation of residence time distribution can help make reasonable predictions at watershed to regional scales.
SCIENCE OF THE TOTAL ENVIRONMENT
(2021)
Article
Computer Science, Interdisciplinary Applications
Joubine Aghili, Jean-Raynald de Dreuzy, Roland Masson, Laurent Trenty
Summary: This paper presents an extension of Discrete Fracture Matrix (DFM) models to compositional two-phase Darcy flow accounting for phase transitions and Fickian diffusion. The hybrid-dimensional model is validated by numerical comparison with a reference equi-dimensional model, showing better accuracy and physical consistency.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Engineering, Civil
N. Guiheneuf, A. Dausse, J-R de Dreuzy, B. L. Parker
Summary: This study introduces a new framework for interpreting the long-term pumping test results in fractured sandstone and shale formations, along with scaling analysis driven by geological evidence. The consistent scaling of transmissivity T to storativity S provides essential information about flow-bearing structures, demonstrating the validity of different flow structures in the fractured rocks.
JOURNAL OF HYDROLOGY
(2021)
Article
Soil Science
Alexandre Coche, Tristan Babey, Alain Rapaport, Laure Vieuble Gonod, Patricia Garnier, Naoise Nunan, Jean-Raynald de Dreuzy
Summary: The bacterial traits in soils play a significant role in the decomposition of organic matter, but have received little attention. A bioreactive transport model was developed to investigate the interactive impacts of spatial dispersion and bacterial traits on mineralization. The study found that bacterial dispersion and traits have a substantial influence on the mineralization process of organic substances in soils.
SOIL BIOLOGY & BIOCHEMISTRY
(2022)
Article
Environmental Sciences
Alexandre Gauvain, Sarah Leray, Jean Marcais, Clement Roques, Camille Vautier, Frederic Gresselin, Luc Aquilina, Jean-Raynald de Dreuzy
Summary: The study found that geomorphological structures have a significant impact on the Transit Time Distributions in shallow aquifers, with the coefficient of variation linearly related to the average distance of groundwater volume to the river. With seepage, the TTD exhibits three separate modes and the coefficient of variation also depends on the extent of the seepage area.
WATER RESOURCES RESEARCH
(2021)
Article
Environmental Sciences
Luca Guillaumot, Jean Marcais, Camille Vautier, Aurelie Guillou, Virginie Vergnaud, Camille Bouchez, Remi Dupas, Patrick Durand, Jean-Raynald de Dreuzy, Luc Aquilina
Summary: The fate of agricultural nitrate in aquifers is influenced by rapid subsurface transfer, denitrification processes, and storage. Quantifying these processes remains challenging due to variable subsurface contributions and unknown aquifer characteristics. The study suggests that nitrate concentration in rivers is determined by aquifer flows stratification, with denitrification potentially controlled by the accessibility of reduced minerals and weathering interfaces.
SCIENCE OF THE TOTAL ENVIRONMENT
(2021)
Article
Environmental Sciences
Jean Marcais, Louis A. Derry, Luca Guillaumot, Luc Aquilina, Jean-Raynald de Dreuzy
Summary: The study found that the transit time distributions of streamwater in the watershed display different proportions of old waters, mainly due to groundwater contributions to the stream. Seasonal variations in transit time are influenced by the variable contributions of different flowpaths and the stratification of groundwater residence times. A parsimonious model is developed to capture the groundwater contribution and its effect on transient transit times. Calibration of hydraulic conductivity, porosities, and tracer data successfully reproduces the concentrations and dynamics of the stream. Groundwater flow contribution is controlled by the hydraulic conductivity, while its age is controlled by the porosities.
WATER RESOURCES RESEARCH
(2022)
Article
Engineering, Civil
Nicolas Cornette, Clement Roques, Alexandre Boisson, Quentin Courtois, Jean Marcais, Josette Launay, Guillaume Pajot, Florence Habets, Jean-Raynald de Dreuzy
Summary: Surface and subsurface flows interact at different scales, influencing water partitioning between base flow, seepage flow, and overland flow. However, quantifying this interaction remains challenging.
JOURNAL OF HYDROLOGY
(2022)
Article
Geosciences, Multidisciplinary
Sylvain Pasquet, Jean Marcais, Jorden L. Hayes, Peter B. Sak, Lin Ma, Jerome Gaillardet
Summary: Weathering and erosion processes are essential for CZ evolution and availability of natural resources. This study proposes a novel workflow using near-surface geophysics to characterize the architecture of the deep CZ at a catchment scale on a volcanic tropical island, revealing complex weathering patterns.
GEOPHYSICAL RESEARCH LETTERS
(2022)
Article
Geosciences, Multidisciplinary
Luca Guillaumot, Laurent Longuevergne, Jean Marcais, Nicolas Lavenant, Olivier Bour
Summary: Groundwater recharge estimation in fractured aquifers is challenging due to the variability of soil properties and the lack of data. This study introduces a new approach that considers groundwater lateral flow and improves recharge estimation by analyzing water table fluctuations in the frequency domain. The findings highlight the importance of rainfall distribution and unsaturated zone thickness in recharge generation.
HYDROLOGY AND EARTH SYSTEM SCIENCES
(2022)
Article
Geosciences, Multidisciplinary
Clement Roques, David E. Rupp, Jean-Raynald De Dreuzy, Laurent Longuevergne, Elizabeth R. Jachens, Gordon Grant, Luc Aquilina, John S. Selker
Summary: We found that the vertical compartmentalization of hillslopes has a significant impact on groundwater flow and recession discharge. Streamflow recession behavior can deviate from predictions made by groundwater theory when hydraulic properties are vertically compartmentalized.
HYDROLOGY AND EARTH SYSTEM SCIENCES
(2022)
