Operator Learning for Reconstructing Flow Fields from Sparse Measurements: an Energy Transformer Approach

Journal of Computational Physics, Год журнала: 2025, Номер unknown, С. 114148 - 114148

Опубликована: Июнь 1, 2025

Язык: Английский

---

Synthetic ground motions in heterogeneous geologies: the HEMEW-3D dataset for scientific machine learning

Опубликована: Янв. 9, 2024

Abstract. The ever-improving performances of physics-based simulations and the rapid developments deep learning are offering new perspectives to study earthquake-induced ground motion. Due large amount data required train neural networks, applications have so far been limited recorded or two-dimensional simulations. To bridge gap between high-fidelity numerical simulations, this work introduces a database earthquake HEMEW-3D comprises 30,000 elastic wave propagation in three-dimensional (3D) geological domains. Each domain is parametrized by different model built from random arrangement layers augmented fields that represent heterogeneities. For each simulation, motion synthetized at surface grid virtual sensors. high frequency waveforms (fmax = 5 Hz) allows extensive analyses Existing foreseen range statistic variability machine methods on models, learning-based predictions depending 3D heterogeneous geologies.

Язык: Английский

---

Synthetic ground motions in heterogeneous geologies from various sources: the HEMEW^S-3D database

Earth system science data, Год журнала: 2024, Номер 16(9), С. 3949 - 3972

Опубликована: Сен. 3, 2024

Abstract. The ever-improving performances of physics-based simulations and the rapid developments deep learning are offering new perspectives to study earthquake-induced ground motion. Due large amount data required train neural networks, applications have so far been limited recorded or two-dimensional (2D) simulations. To bridge gap between high-fidelity numerical simulations, this work introduces a database earthquake HEterogeneous Materials Elastic Waves with Source variability in 3D (HEMEWS-3D) comprises 30 000 elastic wave propagation geological domains. Each domain is parametrized by different model built from random arrangement layers augmented fields that represent heterogeneities. waves originate randomly located pointwise source moment tensor. For each simulation, motion synthesized at surface grid virtual sensors. high frequency waveforms (fmax⁡=5 Hz) allows for extensive analyses Existing foreseen range statistical machine methods on models deep-learning-based predictions depend heterogeneous geologies properties. Data available https://doi.org/10.57745/LAI6YU (Lehmann, 2023).

Язык: Английский

---

Rethinking skip connections in Spiking Neural Networks with Time-To-First-Spike coding

Frontiers in Neuroscience, Год журнала: 2024, Номер 18

Опубликована: Фев. 14, 2024

Time-To-First-Spike (TTFS) coding in Spiking Neural Networks (SNNs) offers significant advantages terms of energy efficiency, closely mimicking the behavior biological neurons. In this work, we delve into role skip connections, a widely used concept Artificial (ANNs), within domain SNNs with TTFS coding. Our focus is on two distinct types connection architectures: (1) addition-based and (2) concatenation-based connections. We find that connections introduce an additional delay spike timing. On other hand, circumvent but produce time gaps between after-convolution paths, thereby restricting effective mixing information from these paths. To mitigate issues, propose novel approach involving learnable for architecture. This successfully bridges gap convolutional branches, facilitating improved mixing. conduct experiments public datasets including MNIST Fashion-MNIST, illustrating advantage architectures. Additionally, demonstrate applicability beyond image recognition tasks extend it to scientific machine-learning tasks, broadening potential uses SNNs.

Язык: Английский

---

Face Expression Recognition with Vision Transformer and Local Mutual Information Maximization

IEEE Access, Год журнала: 2024, Номер 12, С. 169263 - 169276

Опубликована: Янв. 1, 2024

Язык: Английский

---