We develop a theory of various kinds of large-scale clustering of inertial particles in a rotating density stratified or inhomogeneous turbulent fluid flows. The large-scale particle clustering occurs in scales which are much larger than the integral scale of turbulence, and it is described in terms of the effective pumping velocity in a turbulent flux of particles. We show that for a fast rotating strongly anisotropic turbulence, the large-scale clustering occurs in the plane perpendicular to rotation axis in the direction of the fluid density stratification. We apply the theory of the large-scale particle clustering for explanation of the formation of planetesimals (progenitors of planets) in accretion protoplanetary discs. We determine the radial profiles of the radial and azimuthal components of the effective pumping velocity of particles which have two maxima corresponding to different regimes of the particle--fluid interactions: at the small radius it is the Stokes regime, while at the larger radius it is the Epstein regime. With the decrease the particle radius, the distance between the maxima increases. This implies that smaller-size particles are concentrated nearby the central body of the accretion disk, while larger-size particles are accumulated far from the central body. The dynamic time of the particle clustering is about $τ_{\rm dyn} \sim 10^5$--$10^6$ years, while the turbulent diffusion time is about $10^7$ years, that is much larger than the characteristic formation time of large-scale particle clusters ($\sim τ_{\rm dyn}$).
We propose Dyn-HaMR, to the best of our knowledge, the first approach to reconstruct 4D global hand motion from monocular videos recorded by dynamic cameras in the wild. Reconstructing accurate 3D hand meshes from monocular videos is a crucial task for understanding human behaviour, with significant applications in augmented and virtual reality (AR/VR). However, existing methods for monocular hand reconstruction typically rely on a weak perspective camera model, which simulates hand motion within a limited camera frustum. As a result, these approaches struggle to recover the full 3D global trajectory and often produce noisy or incorrect depth estimations, particularly when the video is captured by dynamic or moving cameras, which is common in egocentric scenarios. Our DynHaMR consists of a multi-stage, multi-objective optimization pipeline, that factors in (i) simultaneous localization and mapping (SLAM) to robustly estimate relative camera motion, (ii) an interacting-hand prior for generative infilling and to refine the interaction dynamics, ensuring plausible recovery under (self-)occlusions, and (iii) hierarchical initialization through a combination of state-of-the-art hand tracking methods. Through extensive evaluations on both in-the-wild and indoor datasets, we show that our approach significantly outperforms state-of-the-art methods in terms of 4D global mesh recovery. This establishes a new benchmark for hand motion reconstruction from monocular video with moving cameras. Our project page is at https://dyn-hamr.github.io/.
Our study focuses on Traditional Chinese Medical (TCM) named entity recognition (NER), which involves identifying and extracting specific entity names from TCM record. This task has significant implications for doctors and researchers, as it enables the automated identification of relevant TCM terms, ultimately enhancing research efficiency and accuracy. However, the current Bidirectional Encoder Representations from Transformers-Long Short Term Memory-Conditional Random Fields (BERT-LSTM-CRF) model for TCM NER is constrained by a traditional structure, limiting its capacity to fully harness the advantages provided by Bidirectional Encoder Representations from Transformers (BERT) and long short term memory (LSTM) models. Through comparative experiments, we also observed that the straightforward superimposition of models actually leads to a decrease in recognition results. To optimize the structure of the traditional BERT-BiLSTM-CRF model and obtain more effective text representations, we propose the Dyn-Att Net model, which introduces dynamic attention and a parallel structure. By integrating BERT and LSTM models with the dynamic attention mechanism, our model effectively captures semantic, contextual, and sequential relations within text sequences, resulting in high accuracy. To validate the effectiveness of our model, we compared it with nine other models in TCM dataset namely the publicly available PaddlePaddle dataset. Our Dyn-Att Net model, based on BERT, outperforms the other models, achieving an F1 score of 81.91%, accuracy of 92.06%, precision of 80.26%, and recall of 83.76%. Furthermore, its robust generalization capability is substantiated through validation on the APTNER, MSRA, and EduNER datasets. Overall, the Dyn-Att Net model not only enhances NER accuracy within the realm of traditional Chinese medicine, but also showcases considerable potential for cross-domain generalization. Moreover, the Dyn-Att Net model’s parallel architecture facilitates efficient computation, contributing to time-saving efforts in NER tasks.
Computing-in-memory (CIM), a promising computing paradigm, has demonstrated great energy-efficiency by integrating computing units into memory. However, previous research on CIM has rarely utilized sparsity in activation and weight concurrently. Moreover, new challenges arise when harnessing sparsity in both activation and weight (two-sided sparsity), such as imbalanced workload and low hardware substrate utilization.To fully unleash the acceleration potential brought by two-sided sparsity, we implemented an accelerator called Dyn-Bitpool which innovates on two fronts: 1) a balanced working scheme called "pool first and cross lane sharing" to maximize the available performance benefiting from bit-level sparsity in activation; 2) dynamic topology of CIM arrays to effectively handle low CIM macro utilization issue stemming from value-level sparsity in weight. All the contributions collaborate to speed up Dyn-Bitpool by 1.89x and 2.64x on average on Googlenet, Mobilenetv3, Resnet50, ResNeXt101 and VGG19, compared with two state-of-the-art accelerators featuring CIM.
Parameter-efficient transfer learning (PETL) is a promising task, aiming to adapt the large-scale pre-trained model to downstream tasks with a relatively modest cost. However, current PETL methods struggle in compressing computational complexity and bear a heavy inference burden due to the complete forward process. This paper presents an efficient visual recognition paradigm, called Dynamic Adapter (Dyn-Adapter), that boosts PETL efficiency by subtly disentangling features in multiple levels. Our approach is simple: first, we devise a dynamic architecture with balanced early heads for multi-level feature extraction, along with adaptive training strategy. Second, we introduce a bidirectional sparsity strategy driven by the pursuit of powerful generalization ability. These qualities enable us to fine-tune efficiently and effectively: we reduce FLOPs during inference by 50%, while maintaining or even yielding higher recognition accuracy. Extensive experiments on diverse datasets and pretrained backbones demonstrate the potential of Dyn-Adapter serving as a general efficiency booster for PETL in vision recognition tasks.
This study constitutes the second phase of a research endeavor aimed at evaluating the feasibility of employing Long Short-Term Memory (LSTM) neural networks as a replacement for Reynolds-Averaged Navier-Stokes (RANS) turbulence models. In the initial phase of this investigation (titled Modeling Turbulent Flows with LSTM Neural Networks, arXiv:2307.13784v1 [physics.flu-dyn] 25 Jul 2023), the application of an LSTM-based recurrent neural network (RNN) as an alternative to traditional RANS models was demonstrated. LSTM models were used to predict shear Reynolds stresses in both developed and developing turbulent channel flows, and these predictions were propagated through RANS simulations to obtain mean flow fields of turbulent flows. A comparative analysis was conducted, juxtaposing the LSTM results from computational fluid dynamics (CFD) simulations with outcomes from the $\kappa-\epsilon$ model and data from direct numerical simulations (DNS). These initial findings indicated promising performance of the LSTM approach. This second phase delves further into the challenges encountered and presents robust solutions. Additionally, new results are provided, demonstrating the efficacy of the LSTM model in predicting turbulent behavior in perturbed flows. While the overall study serves as a proof-of-concept for the application of LSTM networks in RANS turbulence modeling, this phase offers compelling evidence of its potential in handling more complex flow scenarios.
Recently, the editing of neural radiance fields (NeRFs) has gained considerable attention, but most prior works focus on static scenes while research on the appearance editing of dynamic scenes is relatively lacking. In this paper, we propose a novel framework to edit the local appearance of dynamic NeRFs by manipulating pixels in a single frame of training video. Specifically, to locally edit the appearance of dynamic NeRFs while preserving unedited regions, we introduce a local surface representation of the edited region, which can be inserted into and rendered along with the original NeRF and warped to arbitrary other frames through a learned invertible motion representation network. By employing our method, users without professional expertise can easily add desired content to the appearance of a dynamic scene. We extensively evaluate our approach on various scenes and show that our approach achieves spatially and temporally consistent editing results. Notably, our approach is versatile and applicable to different variants of dynamic NeRF representations.
Dynamic graph neural networks (DGNNs) have been widely used in modeling and representation learning of graph structure data. Current dynamic representation learning focuses on either discrete learning which results in temporal information loss, or continuous learning which involves heavy computation. In this study, we proposed a novel DGNN, sparse dynamic (Sparse‐Dyn). It adaptively encodes temporal information into a sequence of patches with an equal amount of temporal‐topological structure. Therefore, while avoiding using snapshots which cause information loss, it also achieves a finer time granularity, which is close to what continuous networks could provide. In addition, we also designed a lightweight module, Sparse Temporal Transformer, to compute node representations through structural neighborhoods and temporal dynamics. Since the fully connected attention conjunction is simplified, the computation cost is far lower than the current state‐of‐the‐art. Link prediction experiments are conducted on both continuous and discrete graph data sets. By comparing several state‐of‐the‐art graph embedding baselines, the experimental results demonstrate that Sparse‐Dyn has a faster inference speed while having competitive performance.
Kenza Tlales, Kheir-eddine Otmani, G. Ntoukas
et al.
We present a machine learning-based mesh refinement technique for steady and unsteady flows. The clustering technique proposed by Otmani et al. arXiv:2207.02929 [physics.flu-dyn] is used to mark the viscous and turbulent regions for the flow past a cylinder at Re=40 (steady laminar flow) and Re=3900 (unsteady turbulent flow). Within this clustered region, we increase the polynomial order to show that it is possible to obtain similar levels of accuracy to a uniformly refined mesh. The method is effective as the clustering successfully identifies the two flow regions, a viscous/turbulent dominated region (including the boundary layer and wake) and an inviscid/irrotational region (a potential flow region). The data used within this framework are generated using a high-order discontinuous Galerkin solver, allowing to locally refine the polynomial order (p-refinement) in each element of the clustered region. For the steady laminar test case we are able to reduce the computational cost up to 32% and for the unsteady turbulent case up to 33%.
Kenza Tlales, Kheir-Eddine Otmani, Gerasimos Ntoukas
et al.
We present a machine learning-based mesh refinement technique for steady and unsteady flows. The clustering technique proposed by Otmani et al. arXiv:2207.02929 [physics.flu-dyn] is used to mark the viscous and turbulent regions for the flow past a cylinder at Re=40 (steady laminar flow) and Re=3900 (unsteady turbulent flow). Within this clustered region, we increase the polynomial order to show that it is possible to obtain similar levels of accuracy to a uniformly refined mesh. The method is effective as the clustering successfully identifies the two flow regions, a viscous/turbulent dominated region (including the boundary layer and wake) and an inviscid/irrotational region (a potential flow region). The data used within this framework are generated using a high-order discontinuous Galerkin solver, allowing to locally refine the polynomial order (p-refinement) in each element of the clustered region. For the steady laminar test case we are able to reduce the computational cost up to 32% and for the unsteady turbulent case up to 33%.
Many websites rely on third parties for services (e.g., DNS, CDN, etc.). However, it also exposes them to shared risks from attacks (e.g., Mirai DDoS attack [24]) or cascading failures (e.g., GlobalSign revocation error [21]). Motivated by such incidents, we analyze the prevalence and impact of third-party dependencies, focusing on three critical infrastructure services: DNS, CDN, and certificate revocation checking by CA. We analyze both direct (e.g., Twitter uses Dyn) and indirect (e.g., Netflix uses Symantec as CA which uses Verisign for DNS) dependencies. We also take two snapshots in 2016 and 2020 to understand how the dependencies evolved. Our key findings are: (1) 89% of the Alexa top-100K websites critically depend on third-party DNS, CDN, or CA providers i.e., if these providers go down, these websites could suffer service disruption; (2) the use of third-party services is concentrated, and the top-3 providers of CDN, DNS, or CA services can affect 50%-70% of the top-100K websites; (3) indirect dependencies amplify the impact of popular CDN and DNS providers by up to 25X; and (4) some third-party dependencies and concentration increased marginally between 2016 to 2020. Based on our findings, we derive key implications for different stakeholders in the web ecosystem.
Dynamic link prediction (DLP) makes graph prediction based on historical information. Since most DLP methods are highly dependent on the training data to achieve satisfying prediction performance, the quality of the training data is crucial. Backdoor attacks induce the DLP methods to make wrong prediction by the malicious training data, i.e., generating a subgraph sequence as the trigger and embedding it to the training data. However, the vulnerability of DLP toward backdoor attacks has not been studied yet. To address the issue, we propose a novel backdoor attack framework on DLP, denoted as Dyn-Backdoor. Specifically, Dyn-Backdoor generates diverse initial-triggers by a generative adversarial network (GAN). Then partial links of the initial-triggers are selected to form a trigger set, according to the gradient information of the attack discriminator in the GAN, so as to reduce the size of triggers and improve the concealment of the attack. Experimental results show that Dyn-Backdoor launches effective backdoor attacks on several state-of-the-art DLP models with a success rate more than 90%. Additionally, we conduct a possible defense against Dyn-Backdoor to testify its resistance in defensive settings, highlighting the needs of defenses for backdoor attacks on DLP.
M. Calvo, Andrés Fernández-Martín, Aida Gutiérrez-García
et al.
Prior research using static facial stimuli (photographs) has identified diagnostic face regions (i.e., functional for recognition) of emotional expressions. In the current study, we aimed to determine attentional orienting, engagement, and time course of fixation on diagnostic regions. To this end, we assessed the eye movements of observers inspecting dynamic expressions that changed from a neutral to an emotional face. A new stimulus set (KDEF-dyn) was developed, which comprises 240 video-clips of 40 human models portraying six basic emotions (happy, sad, angry, fearful, disgusted, and surprised). For validation purposes, 72 observers categorized the expressions while gaze behavior was measured (probability of first fixation, entry time, gaze duration, and number of fixations). Specific visual scanpath profiles characterized each emotional expression: The eye region was looked at earlier and longer for angry and sad faces; the mouth region, for happy faces; and the nose/cheek region, for disgusted faces; the eye and the mouth regions attracted attention in a more balanced manner for surprise and fear. These profiles reflected enhanced selective attention to expression-specific diagnostic face regions. The KDEF-dyn stimuli and the validation data will be available to the scientific community as a useful tool for research on emotional facial expression processing.
Convection and magnetic field generation in the Earth and planetary interiors are driven by both thermal and compositional gradients. In this work numerical simulations of finite-amplitude double-diffusive convection and dynamo action in rapidly rotating spherical shells full of incompressible two-component electrically-conducting fluid are reported. Four distinct regimes of rotating double-diffusive convection identified in a recent linear analysis (Silva et al., 2019, Geophys. Astrophys. Fluid Dyn., doi:10.1080/03091929.2019.1640875) are found to persist significantly beyond the onset of instability while their regime transitions remain abrupt. In the semi-convecting and the fingering regimes characteristic flow velocities are small compared to those in the thermally- and compositionally-dominated overturning regimes, while zonal flows remain weak in all regimes apart from the thermally-dominated one. Compositionally-dominated overturning convection exhibits significantly narrower azimuthal structures compared to all other regimes while differential rotation becomes the dominant flow component in the thermally-dominated case as driving is increased. Dynamo action occurs in all regimes apart from the regime of fingering convection. While dynamos persist in the semi-convective regime they are very much impaired by small flow intensities and very weak differential rotation in this regime which makes poloidal to toroidal field conversion problematic. The dynamos in the thermally-dominated regime include oscillating dipolar, quadrupolar and multipolar cases similar to the ones known from earlier parameter studies. Dynamos in the compositionally-dominated regime exhibit subdued temporal variation and remain predominantly dipolar due to weak zonal flow in this regime. These results significantly enhance our understanding of the primary drivers of planetary core flows and magnetic fields.
This paper introduces a dynamic parametric wind farm model that is capable of simulating floating wind turbine platform motion coupled with wake transport under time-varying wind conditions. The simulator is named FOWFSim-Dyn as it is a dynamic extension of the previously developed steady-state Floating Offshore Wind Farm Simulator (FOWFSim). One-dimensional momentum conservation is used to model dynamic propagation of wake centerline locations and average velocities, while momentum recovery is approximated with the assumption of a constant temporal wake expansion rate. Platform dynamics are captured by treating a floating offshore wind farm as a distribution of particles that are subject to aerodynamic, hydrodynamic, and mooring line forces. The finite difference method is used to discretize the momentum conservation equations to yield a nonlinear state-space model. Simulated data are validated against steady-state experimental wind tunnel results obtained from the literature. Predictions of wake centerlines differed from experimental results by at most 8.19% of the rotor diameter. Simulated wake velocity profiles in the far-wake region differed from experimental measurements by less than 3.87% of the free stream wind speed. FOWFSim-Dyn thus possesses a satisfactory level of fidelity for engineering applications. Finally, dynamic simulations are conducted to ensure that time-varying predictions match physical expectations and intuition.
Three eigenvalue bounds are derived for the instability of ideal compressible stratified magnetohydrodynamic shear flows in which the base velocity, density, and magnetic field vary in two directions. The first bound can be obtained by combining the Howard semi-circle theorem with the energy principle of the Lagrangian displacement. Remarkably, no special conditions are needed to use this bound, and for some cases, we can establish the stability of the flow. The second and third bounds come out from a generalisation of the Miles-Howard theory and have some similarity to the semi-ellipse theorem by Kochar & Jain (J. Fluid Mech., vol. 91, 1979, 489) and the bound found by Cally (Astrophys. Fluid Dyn., vol. 31,1983, 43), respectively. An important byproduct of this investigation is that the Miles-Howard stability condition holds only when there is no applied magnetic field and, in addition, the directions of the shear and the stratification are aligned everywhere.
J. Estivalezes, Wojciech Aniszewski, Franck Auguste
et al.
A series of benchmarks based on the physical situation of "phase inversion" between two incompressible liquids is presented. These benchmarks aim at progressing toward the direct numerical simulation of two-phase flows. Several CFD codes developed in French laboratories and using either Volume of Fluid or Level Set interface tracking methods are utilized to provide physical solutions of the benchmarks, convergence studies and code comparisons. Two typical configurations are retained, with integral scale Reynolds numbers of 1.37 10 4 and 4.33 10 5 , respectively. The physics of the problem are probed through macroscopic quantities such as potential and kinetic energies, interfa-cial area, enstrophy or volume ratio of the light fluid in the top part 1 arXiv:1906.02655v1 [physics.flu-dyn]