Search

We studied the transport and deposition behavior of point particles in Rayleigh-B\'enard convection cells subjected to Couette-type wall shear. Direct numerical simulations (DNS) are performed for Rayleigh number (Ra) in the range \(10^7 \leq Ra \leq 10^9\) with a fixed Prandtl number \(Pr = 0.71\), while the wall-shear Reynolds number (\(Re_w\)) is in the range \(0 \leq Re_w \leq 12000\). With the increase of \(Re_w\), the large-scale rolls expanded horizontally, evolving into zonal flow in two-dimensional simulations or streamwise-oriented rolls in three-dimensional simulations. We observed that for particles with a small Stokes number St, they either circulated within the large-scale rolls when buoyancy dominated or drifted near the walls when shear dominated. For medium St particles, pronounced spatial inhomogeneity and preferential concentration were observed regardless of the prevailing flow state. For large St particles, the turbulent flow structure had a minor influence on particles' motion; although clustering still occurred, wall shear had a negligible influence compared to that for medium St particles. We then presented the settling curves to quantify the particle deposition ratio on the walls. Our DNS results aligned well with previous theoretical predictions, which state that small St particles settle with an exponential deposition ratio and large St particles settle with a linear deposition ratio. For medium St particles, where complex particles-turbulence interaction emerges, we developed a new model describing the settling process with an initial linear stage followed by a non-linear stage. Unknown parameters in our model can be determined either by fitting the settling curves or using empirical relations. Compared with DNS results, our model also accurately predicts the average residence time across a wide range of St for various \(Re_w\)., Comment: 33 pages, 22 figures

With the increasing impact of algorithmic decision-making on human lives, the interpretability of models has become a critical issue in machine learning. Counterfactual explanation is an important method in the field of interpretable machine learning, which can not only help users understand why machine learning models make specific decisions, but also help users understand how to change these decisions. Naturally, it is an important task to study the robustness of counterfactual explanation generation algorithms to model changes. Previous literature has proposed the concept of Naturally-Occurring Model Change, which has given us a deeper understanding of robustness to model change. In this paper, we first further generalize the concept of Naturally-Occurring Model Change, proposing a more general concept of model parameter changes, Generally-Occurring Model Change, which has a wider range of applicability. We also prove the corresponding probabilistic guarantees. In addition, we consider a more specific problem, data set perturbation, and give relevant theoretical results by combining optimization theory.

In the realm of Artificial Intelligence (AI), the importance of Explainable Artificial Intelligence (XAI) is increasingly recognized, particularly as AI models become more integral to our lives. One notable single-instance XAI approach is counterfactual explanation, which aids users in comprehending a model's decisions and offers guidance on altering these decisions. Specifically in the context of image classification models, effective image counterfactual explanations can significantly enhance user understanding. This paper introduces a novel method for computing feature importance within the feature space of a black-box model. By employing information fusion techniques, our method maximizes the use of data to address feature counterfactual explanations in the feature space. Subsequently, we utilize an image generation model to transform these feature counterfactual explanations into image counterfactual explanations. Our experiments demonstrate that the counterfactual explanations generated by our method closely resemble the original images in both pixel and feature spaces. Additionally, our method outperforms established baselines, achieving impressive experimental results.

Counterfactual explanation generation is a powerful method for Explainable Artificial Intelligence. It can help users understand why machine learning models make specific decisions, and how to change those decisions. Evaluating the robustness of counterfactual explanation algorithms is therefore crucial. Previous literature has widely studied the robustness based on the perturbation of input instances. However, the robustness defined from the perspective of perturbed instances is sometimes biased, because this definition ignores the impact of learning algorithms on robustness. In this paper, we propose a more reasonable definition, Weak Robust Compatibility, based on the perspective of explanation strength. In practice, we propose WRC-Test to help us generate more robust counterfactuals. Meanwhile, we designed experiments to verify the effectiveness of WRC-Test. Theoretically, we introduce the concepts of PAC learning theory and define the concept of PAC WRC-Approximability. Based on reasonable assumptions, we establish oracle inequalities about weak robustness, which gives a sufficient condition for PAC WRC-Approximability.

Decay-time-dependent $CP$-violation effects in transitions of neutral $B$ mesons to $CP$-eigenstates can be visualised by oscillations in the asymmetry, as a function of decay time, between decay yields from mesons tagged as initially having $\overline{B}$ or $B$ flavour. Such images, for example for $B^0 \to J/\psi K^0_{\rm S}$ decays where the magnitude of the oscillation is proportional to $\sin(2\beta)$ with $\beta$ being an angle of the Cabibbo--Kobayashi--Maskawa Unitarity Triangle, provide a straightforward illustration of the underlying physics. Until now there has been no comparable method to provide visualisation for the case of decays to multibody final states that are not $CP$-eigenstates, where interference between $CP$-even and -odd amplitudes provides additional physics sensitivity. A method is proposed to weight the data so that the terms of interest can be projected out and used to obtain asymmetries that visualise the relevant effects. Application of the weighting to $B^0_s$ decays, where effects due to non-zero width difference are not negligible, provides a novel method to observe $CP$ violation in interference between mixing and decay without tagging the production flavour., Comment: 17 pages, 5 figures

We utilize the Open Accelerator (OpenACC) approach for graphics processing unit (GPU) accelerated particle-resolved thermal lattice Boltzmann (LB) simulation. We adopt the momentum-exchange method to calculate fluid-particle interactions to preserve the simplicity of the LB method. To address load imbalance issues, we extend the indirect addressing method to collect fluid-particle link information at each timestep and store indices of fluid-particle link in a fixed index array. We simulate the sedimentation of 4,800 hot particles in cold fluids with a domain size of $4000^{2}$, and the simulation achieves 1750 million lattice updates per second (MLUPS) on a single GPU. Furthermore, we implement a hybrid OpenACC and message passing interface (MPI) approach for multi-GPU accelerated simulation. This approach incorporates four optimization strategies, including building domain lists, utilizing request-answer communication, overlapping communications with computations, and executing computation tasks concurrently. By reducing data communication between GPUs, hiding communication latency through overlapping computation, and increasing the utilization of GPU resources, we achieve improved performance, reaching 10846 MLUPS using 8 GPUs. Our results demonstrate that the OpenACC-based GPU acceleration is promising for particle-resolved thermal lattice Boltzmann simulation., Comment: 46 pages, 18 figures

We report pore-scale statistical properties of temperature and thermal energy dissipation rate in a two-dimensional porous Rayleigh-B\'enard (RB) cell. High-resolution direct numerical simulations were carried out for the fixed Rayleigh number ($Ra$) of $10^{9}$ and the Prandtl numbers ($Pr$) of 5.3 and 0.7. We consider sparse porous media where the solid porous matrix is impermeable to both fluid and heat flux. The porosity ($\phi$) range $0.86 \leq \phi \le 0.98$, the corresponding Darcy number ($Da$) range $10^{-4}

In Lifelong Learning (LL), agents continually learn as they encounter new conditions and tasks. Most current LL is limited to a single agent that learns tasks sequentially. Dedicated LL machinery is then deployed to mitigate the forgetting of old tasks as new tasks are learned. This is inherently slow. We propose a new Shared Knowledge Lifelong Learning (SKILL) challenge, which deploys a decentralized population of LL agents that each sequentially learn different tasks, with all agents operating independently and in parallel. After learning their respective tasks, agents share and consolidate their knowledge over a decentralized communication network, so that, in the end, all agents can master all tasks. We present one solution to SKILL which uses Lightweight Lifelong Learning (LLL) agents, where the goal is to facilitate efficient sharing by minimizing the fraction of the agent that is specialized for any given task. Each LLL agent thus consists of a common task-agnostic immutable part, where most parameters are, and individual task-specific modules that contain fewer parameters but are adapted to each task. Agents share their task-specific modules, plus summary information ("task anchors") representing their tasks in the common task-agnostic latent space of all agents. Receiving agents register each received task-specific module using the corresponding anchor. Thus, every agent improves its ability to solve new tasks each time new task-specific modules and anchors are received. On a new, very challenging SKILL-102 dataset with 102 image classification tasks (5,033 classes in total, 2,041,225 training, 243,464 validation, and 243,464 test images), we achieve much higher (and SOTA) accuracy over 8 LL baselines, while also achieving near perfect parallelization. Code and data can be found at https://github.com/gyhandy/Shared-Knowledge-Lifelong-Learning, Comment: Transactions on Machine Learning Research (TMLR) paper

In this paper, we present our solutions for the 5th Workshop and Competition on Affective Behavior Analysis in-the-wild (ABAW), which includes four sub-challenges of Valence-Arousal (VA) Estimation, Expression (Expr) Classification, Action Unit (AU) Detection and Emotional Reaction Intensity (ERI) Estimation. The 5th ABAW competition focuses on facial affect recognition utilizing different modalities and datasets. In our work, we extract powerful audio and visual features using a large number of sota models. These features are fused by Transformer Encoder and TEMMA. Besides, to avoid the possible impact of large dimensional differences between various features, we design an Affine Module to align different features to the same dimension. Extensive experiments demonstrate that the superiority of the proposed method. For the VA Estimation sub-challenge, our method obtains the mean Concordance Correlation Coefficient (CCC) of 0.6066. For the Expression Classification sub-challenge, the average F1 Score is 0.4055. For the AU Detection sub-challenge, the average F1 Score is 0.5296. For the Emotional Reaction Intensity Estimation sub-challenge, the average pearson's correlations coefficient on the validation set is 0.3968. All of the results of four sub-challenges outperform the baseline with a large margin.

We adopt the reinforcement learning algorithm to train the self-propelling agent migrating long-distance in a thermal turbulent environment. We choose the Rayleigh-B\'enard turbulent convection cell with an aspect ratio ($\Gamma$, which is defined as the ratio between cell length and cell height) of 2 as the training environment. Our results showed that, compared to a naive agent that moves straight from the origin to the destination, the smart agent can learn to utilize the carrier flow currents to save propelling energy. We then apply the optimal policy obtained from the $\Gamma=2$ cell and test the smart agent migrating in convection cells with $\Gamma$ up to 32. In a larger $\Gamma$ cell, the dominant flow modes of horizontally stacked rolls are less stable, and the energy contained in higher-order flow modes increases. We found that the optimized policy can be successfully extended to convection cells with a larger $\Gamma$. In addition, the ratio of propelling energy consumed by the smart agent to that of the naive agent decreases with the increase of $\Gamma$, indicating more propelling energy can be saved by the smart agent in a larger $\Gamma$ cell. We also evaluate the optimized policy when the agents are being released from the randomly chosen origin, which aims to test the robustness of the learning framework, and possible solutions to improve the success rate are suggested. This work has implications for long-distance migration problems, such as unmanned aerial vehicles patrolling in a turbulent convective environment, where planning energy-efficient trajectories can be beneficial to increase their endurance., Comment: 32 pages, 14 figures

We studied the flow organization and heat transfer properties in two-dimensional and three-dimensional Rayleigh-B\'enard cells that are imposed with different types of wall shear. The external wall shear is added with the motivation of manipulating flow mode to control heat transfer efficiency. We imposed three types of wall shear that may facilitate the single-roll, the horizontally stacked double-roll, and the vertically stacked double-roll flow modes, respectively. Direct numerical simulations are performed for fixed Rayleigh number $Ra = 10^{8}$ and fixed Prandtl number $Pr = 5.3$, while the wall-shear Reynolds number ($Re_{w}$) is in the range $60 \le Re_{w} \le 6000$. Generally, we found enhanced heat transfer efficiency and global flow strength with the increase of $Re_{w}$. However, even with the same magnitude of global flow strength, the heat transfer efficiency varies significantly when the cells are under different types of wall shear. An interesting finding is that by increasing the wall-shear strength, the thermal turbulence is relaminarized, and more surprisingly, the heat transfer efficiency in the laminar state is higher than that in the turbulent state. We found that the enhanced heat transfer efficiency at the laminar regime is due to the formation of more stable and stronger convection channels. We propose that the origin of thermal turbulence laminarization is the reduced amount of thermal plumes. Because plumes are mainly responsible for turbulent kinetic energy production, when the detached plumes are swept away by the wall shear, the reduced number of plumes leads to weaker turbulent kinetic energy production. We also quantify the efficiency of facilitating heat transport via external shearing, and find that for larger $Re_{w}$, the enhanced heat transfer efficiency comes at a price of a larger expenditure of mechanical energy., Comment: 27 pages, 16 figures

We assess the performance of the hybrid Open Accelerator (OpenACC) and Message Passing Interface (MPI) approach for multi-graphics processing units (GPUs) accelerated thermal lattice Boltzmann (LB) simulation. The OpenACC accelerates computation on a single GPU, and the MPI synchronizes the information between multiple GPUs. With a single GPU, the two-dimension (2D) simulation achieved 1.93 billion lattice updates per second (GLUPS) with a grid number of $8193^{2}$, and the three-dimension (3D) simulation achieved 1.04 GLUPS with a grid number of $385^{3}$, which is more than 76% of the theoretical maximum performance. On multi-GPUs, we adopt block partitioning, overlapping communications with computations, and concurrent computation to optimize parallel efficiency. We show that in the strong scaling test, using 16 GPUs, the 2D simulation achieved 30.42 GLUPS and the 3D simulation achieved 14.52 GLUPS. In the weak scaling test, the parallel efficiency remains above 99% up to 16 GPUs. Our results demonstrated that, with improved data and task management, the hybrid OpenACC and MPI technique is promising for thermal LB simulation on multi-GPUs., Comment: 31 pages, 12 figures

Learning the causal structure behind data is invaluable for improving generalization and obtaining high-quality explanations. We propose a novel framework, Invariant Structure Learning (ISL), that is designed to improve causal structure discovery by utilizing generalization as an indication. ISL splits the data into different environments, and learns a structure that is invariant to the target across different environments by imposing a consistency constraint. An aggregation mechanism then selects the optimal classifier based on a graph structure that reflects the causal mechanisms in the data more accurately compared to the structures learnt from individual environments. Furthermore, we extend ISL to a self-supervised learning setting where accurate causal structure discovery does not rely on any labels. This self-supervised ISL utilizes invariant causality proposals by iteratively setting different nodes as targets. On synthetic and real-world datasets, we demonstrate that ISL accurately discovers the causal structure, outperforms alternative methods, and yields superior generalization for datasets with significant distribution shifts., Comment: 16 pages (including Appendix), 4 figures

We present a numerical study of training a self-propelling agent to migrate in the unsteady flow environment. We control the agent to utilize the background flow structure by adopting the reinforcement learning algorithm to minimize energy consumption. We considered the agent migrating in two types of flows: one is simple periodical double-gyre flow as a proof-of-concept example, while the other is complex turbulent Rayleigh-B\'enard convection as a paradigm for migrating in the convective atmosphere or the ocean. The results show that the smart agent in both flows can learn to migrate from one position to another while utilizing background flow currents as much as possible to minimize the energy consumption, which is evident by comparing the smart agent with a naive agent that moves straight from the origin to the destination. In addition, we found that compared to the double-gyre flow, the flow field in the turbulent Rayleigh-B\'enard convection exhibits more substantial fluctuations, and the training agent is more likely to explore different migration strategies; thus, the training process is more difficult to converge. Nevertheless, we can still identify an energy-efficient trajectory that corresponds to the strategy with the highest reward received by the agent. These results have important implications for many migration problems such as unmanned aerial vehicles flying in a turbulent convective environment, where planning energy-efficient trajectories are often involved., Comment: 23 pages, 12 figures

We present a numerical study of vorticity production and transport in the two-dimensional Rayleigh-B\'enard (RB) convection. Direct numerical simulations are carried out in the Rayleigh number ($Ra$) range $10^{5}\le Ra \le 10^{6}$, the Prandtl number ($Pr$) of 0.71, and the aspect ratio ($\Gamma$) of the convection cell range $0.75\le \Gamma \le 6$. We found that the flow structure and temperature distribution vary with $\Gamma$ greatly due to multiple vortices interaction. Further investigation on the vorticity production and transport reveals that, in the RB convection, in addition to the vorticity production due to wall shear stress, buoyancy produces significant vorticity in the bulk region. The produced vorticity is transported via advection and diffusion. An interesting finding is that the main vortices and the corner vortices can be visualized via the contour of buoyancy-produced vorticity. Although a vigorous definition of the vortex is still lacking in the community, our efficient vortex visualization approach in the RB convection may shed light on further research toward vortex identification. We also found that the spatial distribution of vorticity flux along the wall is positively correlated with that of the Nusselt number ($Nu$), suggesting the amount of vorticity that enters the flow is directly related to the amount of thermal energy that enters the flow., Comment: 23 pages, 13 figures