6,913 research outputs found
Modeling ice block failure within drift ice and ice rubble
No full textFunding Information: The authors are grateful for the financial support from the Academy of Finland through Project No. 309830, Ice Block Breakage: Experiments and Simulations (ICEBES). The authors acknowledge CSC–IT Center for Science, Finland, for computational resources under Project No. 2000971, Mechanics and Fracture of Ice. Publisher Copyright: © 2022 authors. Published by the American Physical Society.A major challenge within material science is the proper modeling of force transmission through fragmenting materials under compression. A particularly demanding material is sea ice, which on small scales is an anisotropic material with quasibrittle characteristics under failure. Here we use the particle-based model HiDEM and laboratory-scale experiments on saline ice to develop a material model for fragmenting ice. The material behavior of the HiDEM model-ice, and the experiments are compatible on force transmission and fragmentation if: (i) the typical HiDEM glacier-scale particle size of meters is brought down to millimeters corresponding to the grain size of the laboratory ice, (ii) the often used HiDEM lattice structure is replaced by a planar random structure with an anisotropy in the direction normal to the randomized plane, and (iii) the instant tensile and bending failure criterion, used in HiDEM on glacier scale, is replaced by a cohesive softening failure potential for energy dissipation. The main outcomes of this exercise is that many of the, more or less, traditional ice modeling schemes are proven to be incomplete. In particular, local crushing of ice is not valid as a generic failure mode for fragmented ice under compression. Rather, shear failure, as described by Mohr-Coulomb theory is demonstrated to be the dominant failure mode.Peer reviewe
A Relational Model for One-Shot Classification of Images and Pen Strokes
No full textFunding Information: We thank CSC (IT Center for Science, Finland) for computational resources and the Academy of Finland for the support within the Flagship programme: Finnish Center for Artificial Intelligence (FCAI). Publisher Copyright: © 2022 The AuthorsWe show that a deep learning model with built-in relational inductive bias can bring benefits to sample-efficient learning, without relying on extensive data augmentation. Our study shows that excellent results can be achieved with a model in which the relational inductive bias is applied to images, while building an efficient one-shot classifier on top of raw strokes is more challenging. The proposed one-shot classification model performs relational matching of a pair of inputs in the form of local and pairwise attention. Our approach solves with almost perfect accuracy the one-shot image classification Omniglot challenge when combined with a Hungarian matching algorithm and attains competitive results on the same task on characters represented as rotation-augmented strokes.Peer reviewe
Report of work experience at CSC-IT Center for Science.pdf
No full textThis is a summary and report of the main tasks and activities completed, tools and technologies used, and challenges faced during a two-week internship at CSC-IT Center for Science Ltd., Finland.</p
Work report on two week experience at CSC-IT Center for Science
No full textEncapsulation of the two weeks spent at CSC-IT Center for Science from all aspects of working at a tech company. Reflection on new knowledge gained and challenges along with their solutions in that time period. </p
DLBFoam: An open-source dynamic load balancing model for fast reacting flow simulations in OpenFOAM
No full textFunding Information: The present study has been financially supported by the Academy of Finland (grant number 318024). The computational resources for this study were provided by CSC - Finnish IT Center for Science. The first author has been financially supported by the Merenkulun S??ti?. Funding Information: The present study has been financially supported by the Academy of Finland (grant number 318024 ). The computational resources for this study were provided by CSC - Finnish IT Center for Science. The first author has been financially supported by the Merenkulun Säätiö . Publisher Copyright: © 2021 The Author(s)Computational load imbalance is a well-known performance issue in multiprocessor reacting flow simulations utilizing directly integrated chemical kinetics. We introduce an open-source dynamic load balancing model named DLBFoam to address this issue within OpenFOAM, an open-source C++ library for Computational Fluid Dynamics (CFD). Due to the commonly applied operator splitting practice in reactive flow solvers, chemistry can be treated as an independent stiff ordinary differential equation (ODE) system within each computational cell. As a result of the highly non-linear characteristics of chemical kinetics, a large variation in the convergence rates of the ODE integrator may occur, leading to a high load imbalance across multiprocessor configurations. However, the independent nature of chemistry ODE systems leads to a problem that can be parallelized easily (called an embarrassingly parallel problem in the literature) during the flow solution. The presented model takes advantage of this feature and balances the chemistry load across available resources. Additionally, a reference mapping model is utilized to further speed-up the simulations. When DLBFoam it utilized with both these features enabled, a speed-up by a factor of 10 is reported for reactive flow benchmark cases. To the best of our knowledge, this model is the first open-source implementation of chemistry load balancing in the literature. (C) 2021 The Author(s). Published by Elsevier B.V.Peer reviewe
Recommendations for Preservation Data Policies
No full textPoster at Open Repositories 2014, Helsinki, Finland, June 9-13, 2014Posters, Demos and Developer "How-To's"In this paper, we summarise selected recommendations that should be taken into account when drawing up data policies concerning digital preservation. It is important to understand what current data policies address and if they miss out on important topics, such as specific requirements for data preservation. This gives an indication of the possible impact of such data policies on individual communities, for example for repository managers, and allows recommendations to be drawn up to guide forthcoming policies. The recommendations suggested in this paper are based on both desktop research on selected data policies and an online survey conducted by the APARSEN project during autumn 2013.Lehtonen, Juha (CSC - IT Center for Science, Finland)Helin, Heikki (CSC - IT Center for Science, Finland)Dallmeier-Tiessen, Suenje (CERN – European Organization for Nuclear Research, Switzerland)Guercio, Mariella (CINI – Consorzio Interuniversitario Nazionale per l’Informatica, Italy)Herterich, Patricia (CERN – European Organization for Nuclear Research, Switzerland)Kaur, Kirnn (British Library, United Kingdom)Lavasa, Artemis (CERN – European Organization for Nuclear Research, Switzerland)Salmivalli, Riina (CSC - IT Center for Science, Finland
Multitask Recalibrated Aggregation Network for Medical Code Prediction
No full text| openaire: EC/H2020/101016775/EU//INTERVENE Funding Information: Acknowledgments. This work was supported by the Academy of Finland (grant 336033) and EU H2020 (grant 101016775). We acknowledge the computational resources provided by the Aalto Science-IT project. The authors wish to acknowledge CSC - IT Center for Science, Finland, for computational resources. Publisher Copyright: © 2021, The Author(s).Medical coding translates professionally written medical reports into standardized codes, which is an essential part of medical information systems and health insurance reimbursement. Manual coding by trained human coders is time-consuming and error-prone. Thus, automated coding algorithms have been developed, building especially on the recent advances in machine learning and deep neural networks. To solve the challenges of encoding lengthy and noisy clinical documents and capturing code associations, we propose a multitask recalibrated aggregation network. In particular, multitask learning shares information across different coding schemes and captures the dependencies between different medical codes. Feature recalibration and aggregation in shared modules enhance representation learning for lengthy notes. Experiments with a real-world MIMIC-III dataset show significantly improved predictive performance.Peer reviewe
Numerical evidence on deflagration fronts in a methane/n-dodecane dual-fuel shear layer under engine relevant conditions
No full textFunding Information: This study has been funded by the Academy of Finland (grant numbers 318024, 332784, and 297248). We acknowledge CSC, Finnish IT Center for Science for providing the computational resources. Funding Information: This study has been funded by the Academy of Finland (grant numbers 318024 , 332784 , and 297248 ). We acknowledge CSC, Finnish IT Center for Science for providing the computational resources. Publisher Copyright: © 2023 The Author(s)To date, high resolution spray-assisted dual-fuel (DF) studies have focused on capturing the ignition process while the subsequent post-ignition events have been largely neglected due to modeling requirements and high computational cost. Here, we use a simplified approach for studying ignition front evolution after ignition. Three-dimensional scale-resolved simulations of igniting shear layers (0≤Re≤1500) are studied to better understand reaction fronts in engine-relevant conditions. We carry out quasi-DNS in a DF combustion setup consisting of premixed n-dodecane/methane/air/EGR at 700K as a fuel stream and premixed methane/air as the oxidizer at a pressure of 60 atmospheres and an ambient temperature of 900 K. The flow solution resolution is δ/10, where δ=laminar flame thickness. The present study primarily focuses on the hypothesized flame formation and its characterization. Under these conditions, the simulations indicate two-stage ignition further leading to reaction front initiation and dual-fuel flame establishment. For Re<1500, a reaction front resembling DF deflagration is demonstrated close to the auto-ignition timescales. At Re=1500, mixing effects promote more rapid dilution and the DF deflagration front formation is slightly delayed although still observed. For the first time, at rather short timescales of 0.2−0.4 IDT (ignition delay time) after the ignition, we provide numerical evidence on DF deflagration front emergence in shear-driven DF combustion processes via 3D numerical simulations for 0<Re≤1500.Peer reviewe
Say, S (as) Semantics - And Mean it! Path to Semantically Interoperable Digital Research Services
No full textThe more we invest in open science and research, the more we need to ensure that metadata enabling discovering and digital preservation of research material is of high-quality and semantically coherent. Still, interoperability of information systems and the lack of shared semantics, both between humans and machines, is an internationally recognised issue. In Finland we are in the process of implementing information systems and harmonising the legacy data models in the way that it makes use of the shared semantics, standards and other best practices according to the common architectural vision. This basic infrastructure for information management is built by combining terminological theory, linked data and adaptable data modelling practices. The idea of the Semantic Interoperability Model and new tools, IOW - Interoperability Workbench, supporting it are presented in the context of research and science in Finland, but the vision of the linked information components is generic.Peer reviewe
Learning to Predict Head Pose in Remotely-Rendered Virtual Reality
No full textFunding Information: This work has been supported by the Academy of Finland under grant numbers 332306, 332307, and 357533. We would like to thank the CSC – IT Center for Science and the Aalto Science-IT project for provisioning the computational resources used for the evaluation. Publisher Copyright: © 2023 Owner/Author(s).Accurate characterization of Head Mounted Display (HMD) pose in a virtual scene is essential for rendering immersive graphics in Extended Reality (XR). Remote rendering employs servers in the cloud or at the edge of the network to overcome the computational limitations of either standalone or tethered HMDs. Unfortunately, it increases the latency experienced by the user; for this reason, predicting HMD pose in advance is highly beneficial, as long as it achieves high accuracy. This work provides a thorough characterization of solutions that forecast HMD pose in remotely-rendered virtual reality (VR) by considering six degrees of freedom. Specifically, it provides an extensive evaluation of pose representations, forecasting methods, machine learning models, and the use of multiple modalities along with joint and separate training. In particular, a novel three-point representation of pose is introduced together with a data fusion scheme for long-Term short-Term memory (LSTM) neural networks. Our findings show that machine learning models benefit from using multiple modalities, even though simple statistical models perform surprisingly well. Moreover, joint training is comparable to separate training with carefully chosen pose representation and data fusion strategies.Peer reviewe
- …
