Computational Resources Research Articles

DRCmpVis: Visual Comparison of Physical Targets in Mobile Diminished and Mixed Reality.

Numerous physical objects in our daily lives are grouped or ranked according to a stereotyped presentation style. For example, in a library, books are typically grouped and ranked based on classification numbers. However, for better comparison, we often need to re-group or re-rank the books using additional attributes such as ratings, publishers, comments, publication years, keywords, prices, etc., or a combination of these factors. In this article, we propose a novel mobile DR/MR-based application framework named DRCmpVis to achieve in-context multi-attribute comparisons of physical objects with text labels or textual information. The physical objects are scanned in the real world using mobile cameras. All scanned objects are then segmented and labeled by a convolutional neural network and replaced (diminished) by their virtual avatars in a DR environment. We formulate three visual comparison strategies, including filtering, re-grouping, and re-ranking, which can be intuitively, flexibly, and seamlessly performed on their avatars. This approach avoids breaking the original layouts of the physical objects. The computation resources in virtual space can be fully utilized to support efficient object searching and multi-attribute visual comparisons. We demonstrate the usability, expressiveness, and efficiency of DRCmpVis through a user study, NASA TLX assessment, quantitative evaluation, and case studies involving different scenarios.

Evolution-Based Shape and Behavior Co-Design of Virtual Agents.

We introduce a novel co-design method for autonomous moving agents' shape attributes and locomotion by combining deep reinforcement learning and evolution with user control. Our main inspiration comes from evolution, which has led to wide variability and adaptation in Nature and has significantly improved design and behavior simultaneously. Our method takes an input agent with optional user-defined constraints, such as leg parts that should not evolve or are only within the allowed ranges of changes. It uses physics-based simulation to determine its locomotion and finds a behavior policy for the input design that is used as a baseline for comparison. The agent is randomly modified within the allowed ranges, creating a new generation of several hundred agents. The generation is trained by transferring the previous policy, which significantly speeds up the training. The best-performing agents are selected, and a new generation is formed using their crossover and mutations. The next generations are then trained until satisfactory results are reached. We show a wide variety of evolved agents, and our results show that even with only 10% of allowed changes, the overall performance of the evolved agents improves by 50%. If more significant changes to the initial design are allowed, our experiments' performance will improve even more to 150%. Our method significantly improved motion tasks without changing body structures, and it does not require considerable computation resources as it works on a single GPU and provides results by training thousands of agents within 30 minutes.

From code sharing to sharing of implementations: Advancing reproducible AI development for medical imaging through federated testing

BackgroundThe reproducibility crisis in AI research remains a significant concern. While code sharing has been acknowledged as a step toward addressing this issue, our focus extends beyond this paradigm. In this work, we explore “federated testing” as an avenue for advancing reproducible AI research and development especially in medical imaging. Unlike federated learning, where a model is developed and refined on data from different centers, federated testing involves models developed by one team being deployed and evaluated by others, addressing reproducibility across various implementations. MethodsOur study follows an exploratory design aimed at systematically evaluating the sources of discrepancies in shared model execution for medical imaging and outputs on the same input data, independent of generalizability analysis. We distributed the same model code to multiple independent centers, monitoring execution in different runtime environments while considering various real-world scenarios for pre- and post-processing steps. We analyzed deployment infrastructure by comparing the impact of different computational resources (GPU vs. CPU) on model performance. To assess federated testing in AI models for medical imaging, we performed a comparative evaluation across different centers, each with distinct pre- and post-processing steps and deployment environments, specifically targeting AI-driven positron emission tomography (PET) imaging segmentation. More specifically, we studied federated testing for an AI-based model for surrogate total metabolic tumor volume (sTMTV) segmentation in PET imaging: the AI algorithm, trained on maximum intensity projection (MIP) data, segments lymphoma regions and estimates sTMTV. ResultsOur study reveals that relying solely on open-source code sharing does not guarantee reproducible results due to variations in code execution, runtime environments, and incomplete input specifications. Deploying the segmentation model on local and virtual GPUs compared to using Docker containers showed no effect on reproducibility. However, significant sources of variability were found in data preparation and pre-/post- processing techniques for PET imaging. These findings underscore the limitations of code sharing alone in achieving consistent and accurate results in federated testing. ConclusionAchieving consistently precise results in federated testing requires more than just sharing models through open-source code. Comprehensive pipeline sharing, including pre- and post-processing steps, is essential. Cloud-based platforms that automate these processes can streamline AI model testing across diverse locations. Standardizing protocols and sharing complete pipelines can significantly enhance the robustness and reproducibility of AI models.

Markov Progressive Framework, a Universal Paradigm for Modeling Long Videos.

The computational complexity of video models increases linearly with the square number of frames. Thus, constrained bycomputational resources, training video models to learn long-term temporal semantics end-to-end is quite a challenge. Currently, the main-stream method is to split a raw video into clips, leading to incomplete fragmentary temporal information flow and failure of modeling long-term semantics. In this paper, we design the Markov Progressive framework (MaPro), a theoretical framework consisting of the progressive modeling method and a paradigm model tailored for it. Thecore idea of MaPro is to find a paradigm model consisting of proposed Markov operators which can be trained in multiple sequential steps and ensure that the multi-step progressive modeling is equivalent to the conventional end-to-endmodeling. By training the paradigm model under the progressive method, we are able to model long videos end-to-endwith limited resources and ensure the effective transmission of long-term temporal information. We provide implementations of this theoretical system on the mainstream CNN- and Transformer-based models, where they are modified to conform to the Markov paradigm. As a general and robust training method, we experimentally demonstrate that it yields significant performance improvements on different backbones and datasets. As an illustrative example, the proposed method improves the SlowOnly network by 4.1 mAP on Charades and 2.5 top-1 accuracy on Kinetics. And for TimeSformer, MaPro improves its performance on Kinetics by 2.0 top-1 accuracy. Importantly, all these improvements areachieved with a little parameter and computation overhead.

Joint Computation Offloading and Resource Allocation in Multi-Edge Smart Communities With Personalized Federated Deep Reinforcement Learning

Joint Computation Offloading and Resource Allocation in Multi-Edge Smart Communities With Personalized Federated Deep Reinforcement Learning

A new approach based on genetic algorithm for computation off-loading optimization in multi-access edge computing networks

The proliferation of smart devices and the increasing demand for resource- intensive applications present significant challenges in terms of computational efficiency, leading to surge in data traffic. While cloud computing offers partial solutions, its centralized architecture raises concerns about latency. Multi-access edge computing (MEC) emerges as promising alternative by deploying servers at the network edge to bring computations closer to user devices. However, op- timizing computation offloading in the dynamic MEC environment remains a complex challenge. This paper introduces novel genetic algorithm-based ap- proach for efficient computation offloading in MEC, considering processing and transmission delays, user preferences, and system constraints. The pro- posed approach integrates computation offloading and resource allocation al- gorithm based on evolutionary principles, combined with a greedy strategy to maximize overall system performance. By utilizing genetic algorithms, the pro- posed method enables dynamic adaptation to changing conditions, eliminating the need for intricate mathematical models and providing an appealing solution to the complexities inherent in MEC. The urgency of this research arises from the critical need to enhance mobile application performance. Simulation re- sults demonstrate the robustness and efficacy of our approach in achieving near- optimal solutions while efficiently balancing computation offloading, minimiz- ing latency, and maximizing resource utilization. Our approach offers flexibility and adaptability, contributing to advancement of MEC networks and addressing the requirements of latency-sensitive applications.

Cascade Vertical Federated Learning Towards Straggler Mitigation and Label Privacy Over Distributed Labels

Vertical federated learning (VFL) enables collaborative machine learning on vertically partitioned data with privacy-preservation. Most VFL methods face three daunting challenges in real-world applications. First, most existing VFL methods assume that at least one party holds the complete set of labels of all data samples. However, this assumption often violates the nature of many practical scenarios, where the parties only have partial labels. Second, the heterogeneity and dynamic of computational and communication resources in participated parties may cause the straggler problem and slow down training convergence. Third, the confidential label information could be exposed through malicious parties during VFL. To address these challenges, we propose a novel VFL algorithm named Cascade Vertical Federated Learning (CVFL), in which partitioned labels can be fully utilized to train neural networks with privacy-preservation. To mitigate the straggler problem, we design a novel optimization objective to increase straggler's contribution to the trained models. To mitigate the label privacy risks, we design a novel defense approach to protect the label privacy of CVFL. We conduct comprehensive experiments and the results demonstrate the effectiveness and efficiency of CVFL. Further, the proposed defense approach can achieve a better tradeoff between label privacy and model utility than two widely-used defense approaches.

A Survey on Deep Neural Network Pruning: Taxonomy, Comparison, Analysis, and Recommendations.

Modern deep neural networks, particularly recent large language models, come with massive model sizes that require significant computational and storage resources. To enable the deployment of modern models on resource-constrained environments and to accelerate inference time, researchers have increasingly explored pruning techniques as a popular research direction in neural network compression. More than three thousand pruning papers have been published from 2020 to 2024. However, there is a dearth of up-to-date comprehensive review papers on pruning. To address this issue, in this survey, we provide a comprehensive review of existing research works on deep neural network pruning in a taxonomy of 1) universal/specific speedup, 2) when to prune, 3) how to prune, and 4) fusion of pruning and other compression techniques. We then provide a thorough comparative analysis of eight pairs of contrast settings for pruning (e.g., unstructured/structured, one-shot/iterative, data-free/data-driven, initialized/pre-trained weights, etc.) and explore several emerging topics, including pruning for large language models, vision transformers, diffusion models, and large multimodal models, post-training pruning, and different levels of supervision for pruning to shed light on the commonalities and differences of existing methods and lay the foundation for further method development. Finally, we provide some valuable recommendations on selecting pruning methods and prospect several promising research directions for neural network pruning. To facilitate future research on deep neural network pruning, we summarize broad pruning applications (e.g., adversarial robustness, natural language understanding, etc.) and build a curated collection of datasets, networks, and evaluations on different applications. We maintain a repository on https://github.com/hrcheng1066/awesome-pruning that serves as a comprehensive resource for neural network pruning papers and corresponding open-source codes. We will keep updating this repository to include the latest advancements in the field.

Data-driven reduced order surrogate modeling for coronary in-stent restenosis

Background:The intricate process of coronary in-stent restenosis (ISR) involves the interplay between different mediators, including platelet-derived growth factor, transforming growth factor-β, extracellular matrix, smooth muscle cells, endothelial cells, and drug elution from the stent. Modeling such complex multiphysics phenomena demands extensive computational resources and time. Methods:This paper proposes a novel non-intrusive data-driven reduced order modeling approach for the underlying multiphysics time-dependent parametrized problem. In the offline phase, a 3D convolutional autoencoder, comprising an encoder and decoder, is trained to achieve dimensionality reduction. The encoder condenses the full-order solution into a lower-dimensional latent space, while the decoder facilitates the reconstruction of the full solution from the latent space. To deal with the 5D input datasets (3D geometry + time series + multiple output channels), two ingredients are explored. The first approach incorporates time as an additional parameter and applies 3D convolution on individual time steps, encoding a distinct latent variable for each parameter instance within each time step. The second approach reshapes the 3D geometry into a 2D plane along a less interactive axis and stacks all time steps in the third direction for each parameter instance. This rearrangement generates a larger and complete dataset for one parameter instance, resulting in a singular latent variable across the entire discrete time-series. In both approaches, the multiple outputs are considered automatically in the convolutions. Moreover, Gaussian process regression is applied to establish correlations between the latent variable and the input parameter. Results:The constitutive model reveals a significant acceleration in neointimal growth between 30−60 days post percutaneous coronary intervention (PCI). The surrogate models applying both approaches exhibit high accuracy in pointwise error, with the first approach showcasing smaller errors across the entire evaluation period for all outputs. The parameter study on drug dosage against ISR rates provides noteworthy insights of neointimal growth, where the nonlinear dependence of ISR rates on the peak drug flux exhibits intriguing periodic patterns. Applying the trained model, the rate of ISR is effectively evaluated, and the optimal parameter range for drug dosage is identified. Conclusion:The demonstrated non-intrusive reduced order surrogate model proves to be a powerful tool for predicting ISR outcomes. Moreover, the proposed method lays the foundation for real-time simulations and optimization of PCI parameters.

Fast-Vid2Vid++: Spatial-Temporal Distillation for Real-Time Video-to-Video Synthesis.

Video-to-Video synthesis (Vid2Vid) gains remarkable performance in generating a photo-realistic video from a sequence of semantic maps, such as segmentation, sketch and pose. However, this pipeline is heavily limited to high computational cost and long inference latency, mainly attributed to two essential factors: 1) network architecture parameters, 2) sequential data stream. Recently, the parameters of image-based generative models have been significantly reduced via more efficient network architectures. Existing methods mainly focus on slimming network architectures but ignore the size of the sequential data stream. Moreover, due to the lack of temporal coherence, image-based compression is not sufficient for the compression of the video task. In this paper, we present a spatial-temporal hybrid distillation compression framework, Fast-Vid2Vid++, which focuses on knowledge distillation of the teacher network and the data stream of generative models on both space and time. Fast-Vid2Vid++ makes the first attempt at time dimension to transfer hierarchical features and time coherence knowledge to reduce computational resources and accelerate inference. Specifically, we compress the data stream spatially and reduce the temporal redundancy. We distill the knowledge of the hierarchical features and the final response from the teacher network to the student network in high-resolution and full-time domains. We transfer the long-term dependencies of the features and video frames to the student model. After the proposed spatial-temporal hybrid knowledge distillation (Spatial-Temporal-HKD), our model can synthesize high-resolution key-frames using the low-resolution data stream. Finally, Fast-Vid2Vid++ interpolates intermediate frames by motion compensation with slight latency and generates full-length sequences with motion-aware inference (MAI). On standard benchmarks, Fast-Vid2Vid++ achieves a real-time performance of 30-59 FPS and saves 28-35× computational cost on a single V100 GPU.

An effective surrogate-assisted rank method for evolutionary neural architecture search

Evolutionary neural architecture search (ENAS) is able to automatically design high-performed architectures of deep neural networks (DNNs) for specific tasks. In recent years, surrogate models have gained significant traction because they can estimate the performance of neural architectures, avoiding excessive computational costs for training. However, most existing surrogate models primarily predict the performance of architectures directly or predict pairwise comparison relationships, which makes it challenging to obtain the rank of a group of architectures when training samples are limited. To address this problem, we propose TCMR-ENAS, an effective triple-competition model-assisted rank method for ENAS. TCMR-ENAS employs a novel triple-competition surrogate model combined with a score-based fitness evaluation method to predict group performance rank. Moreover, a progressive online learning method is proposed to enhance the predictive performance of the triple-competition surrogate model in the framework of modified genetic search. To validate the effectiveness of TCMR-ENAS, we conducted a series of experiments on NAS-Bench-101, NAS-Bench-201, NATS-Bench and NAS-Bench-301, respectively. Experimental results show that TCMR-ENAS can achieve better performance with lower computational resources. The accuracies of searched architectures achieve the best results compared with those of the state-of-the-art methods with limited training samples. In addition, the factors that may influence the effectiveness of TCMR-ENAS are explored in the ablation studies.

Efficient Asynchronous Multi-Participant Vertical Federated Learning

Vertical Federated Learning (VFL) is a private-preserving distributed machine learning paradigm that collaboratively trains machine learning models with participants whose local data overlap largely in the sample space, but not so in the feature space. Existing VFL methods are mainly based on synchronous computation and homomorphic encryption (HE). Due to the differences in the communication and computation resources of the participants, straggling participants can cause delays during synchronous VFL model training, resulting in low computational efficiency. In addition, HE incurs high computation and communication costs. Moreover, it is difficult to establish a VFL coordinator (a.k.a. server) that all participants can trust. To address these problems, we propose an efficient Asynchronous Multi-participant Vertical Federated Learning method (AMVFL). AMVFL leverages asynchronous training which reduces waiting time. At the same time, secret sharing is used instead of HE for privacy protection, which further reduces the computational cost. In addition, AMVFL does not require a trusted entity to serve as the VFL coordinator. Experimental results based on real-world and synthetic datasets demonstrate that AMVFL can significantly reduce computational cost and improve the accuracy of the model compared to five state-of-the-art VFL methods.

Rotating machinery flow field prediction based on hybrid neural network

ABSTRACT The flow phenomena in fluid machinery such as pumps, wind turbines and turbines are complex. For large-scale engineering problems, the computational resources required to solve the Navier-Stokes(N-S) equation are large. In order to solve this problem, a fast and accurate solution method based on hybrid neural network (Convolutional Neural Network – Long Short-Term Memory) is proposed. The nonlinear mapping relationship between the flow field characteristics of rotating machinery and the eddy viscosity coefficient is constructed. The deep learning model is used to replace the RANS standard k-ϵ model, and the eddy viscosity coefficient distribution cloud diagram of rotating machinery is obtained. The results show that the predicted value of the eddy viscosity coefficient of the hybrid neural network model on different slices is in good agreement with the original value and the error is small. Compared with traditional machine learning models such as random forest model, hybrid neural network model takes up only 11.4% of memory, and the accuracy is much better than the random forest model. The hybrid neural network model proposed in this paper has great potential in the prediction of the flow field of rotating machinery.

Gradient Inversion Attacks: Impact Factors Analyses and Privacy Enhancement.

Gradient inversion attacks (GIAs) have posed significant challenges to the emerging paradigm of distributed learning, which aims to reconstruct the private training data of clients (participating parties in distributed training) through the shared parameters. For counteracting GIAs, a large number of privacy-preserving methods for distributed learning scenario have emerged. However, these methods have significant limitations, either compromising the usability of global model or consuming substantial additional computational resources. Furthermore, despite the extensive efforts dedicated to defense methods, the underlying causes of data leakage in distributed learning still have not been thoroughly investigated. Therefore, this paper tries to reveal the potential reasons behind the successful implementation of existing GIAs, explore variations in the robustness of models against GIAs during the training process, and investigate the impact of different model structures on attack performance. After these explorations and analyses, this paper propose a plug-and-play GIAs defense method, which augments the training data by a designed vicinal distribution. Sufficient empirical experiments demonstrate that this easy-to-implement method can ensure the basic level of privacy without compromising the usability of global model.

Fast interactive simulations of cardiac electrical activity in anatomically accurate heart structures by compressing sparse uniform cartesian grids

Background and Objective:Numerical simulations are valuable tools for studying cardiac arrhythmias. Not only do they complement experimental studies, but there is also an increasing expectation for their use in clinical applications to guide patient-specific procedures. However, numerical studies that solve the reaction–diffusion equations describing cardiac electrical activity remain challenging to set up, are time-consuming, and in many cases, are prohibitively computationally expensive for long studies. The computational cost of cardiac simulations of complex models on anatomically accurate structures necessitates parallel computing. Graphics processing units (GPUs), which have thousands of cores, have been introduced as a viable technology for carrying out fast cardiac simulations, sometimes including real-time interactivity. Our main objective is to increase the performance and accuracy of such GPU implementations while conserving computational resources. Methods:In this work, we present a compression algorithm that can be used to conserve GPU memory and improve efficiency by managing the sparsity that is inherent in using Cartesian grids to represent cardiac structures directly obtained from high-resolution MRI and mCT scans. Furthermore, we present a discretization scheme that includes the cross-diagonal terms in the computational cell to increase numerical accuracy, which is especially important for simulating thin tissue sections without the need for costly mesh refinement. Results:Interactive WebGL simulations of atrial/ventricular structures (on PCs, laptops, tablets, and phones) demonstrate the algorithm’s ability to reduce memory demand by an order of magnitude and achieve calculations up to 20x faster. We further showcase its superiority in slender tissues and validate results against experiments performed in live explanted human hearts. Conclusions:In this work, we present a compression algorithm that accelerates electrical activity simulations on realistic anatomies by an order of magnitude (up to 20x), thereby allowing the use of finer grid resolutions while conserving GPU memory. Additionally, improved accuracy is achieved through cross-diagonal terms, which are essential for thin tissues, often found in heart structures such as pectinate muscles and trabeculae, as well as Purkinje fibers. Our method enables interactive simulations with even interactive domain boundary manipulation (unlike finite element/volume methods). Finally, agreement with experiments and ease of mesh import into WebGL paves the way for virtual cohorts and digital twins, aiding arrhythmia analysis and personalized therapies.

Integrated impacts of surrounding building density and gap permeability on pollutant dispersion in four-way street networks.

Gaps between buildings facing the streets can effectively accelerate the natural removal of pollutants in street canyons by enhancing ventilation and diffusion processes. This removal process is closely related to gap permeability (Pg, ratio of gap width to street length) and building density (λn) surrounding the streets. However, the integrated effects of these two parameters on pollutant dispersion remain unclear, because of the limited computational resources and the difficulty of traditional modeling methods to discrete the numerous buildings. Numerical simulations coupled with the porous media approach are performed for this purpose. Results indicate that for isolated streets surrounded by empty land (λn = 0), the high-speed airflow through the gaps causes significant flushing effects on pollutants in the street, and thus larger Pg induces better street air quality than the case without gaps (Pg = 0). Conversely, when the streets are surrounded by buildings (λn ≠ 0), the air quality is good for both smaller Pg = 0 and larger Pg = 52 %, and the worst case occurs in Pg = 24 %. Moreover, because of the blocking effect of surrounding buildings, the percentage of pollutant removal by the gaps decreases with increasing λn, which can reach up to 50 % for λn = 0, decreasing to 30 % for medium density (λn = 0.22). Notably, for high density (λn = 0.44), only when Pg > 52 % the pollutant concentration can be reduced compared to Pg = 0. The present findings can serve as principles for determining appropriate building gap layouts at different building densities, thereby improving the urban air quality more effectively.

Hierarchical Federated Deep Reinforcement Learning based Joint Communication and Computation for UAV Situation Awareness

The computation-intensive situational awareness (SA) task of unmanned aerial vehicle (UAV) is greatly affected by its limited power and computing capability. To solve this challenge, we consider the joint communication and computation (JCC) design for UAV network in this paper. Firstly, a multi-objective optimization (MOO) model, which can optimize UAV computation offloading, transmit power, and local computation resources simultaneously, is built to minimize energy consumption and task execution delay. Then, we develop Thompson sampling based double-DQN (TDDQN) learning algorithm which allows the agent to explore more deeply and effectively, and propose a joint optimization algorithm that combines TDDQN and sequential least squares quadratic programming (SLSQP) to handle the MOO problem. Finally, to enhance the training speed and quality, we incorporate federated learning (FL) into the presented joint optimization algorithm and propose hierarchical federated TDDQN with SLSQP (HF TDDQN-S) to implement the JCC design. Simulation results show that the introduced HF TDDQN-S can efficiently learn the best JCC strategy and minimize the average cost contrasted with the DDQN with SLSQP (DDQN-S) and TDDQN with SLSPQ (TDDQN-S) approach, and achieve the low average delay SA with power efficient.

Topologic-thermal synergism analysis for wedge-shaped channels leveraging data mining and self-organization geometries

Heat transfer enhancement is particularly difficult to achieve within narrow space and turning channels due to strong structural constraints plus flow recirculations. Traditional designs usually follow periodic topology layout of simple geometric features, such as pin-fin arrays, to distribute cooling for narrow turning channels, lacking alignment with specific design objectives and the constraints. This is attributed to three main issues: the absence of control methods for complex structures, the lack of data for variable topologies, and insufficient understandings of Topologic-thermal Synergism. To address these challenges, the integration of self-organization geometry and the constrained Bayesian optimization method is employed. The channel is divided into 15 regions: the left region is subjected to solid presence control and anisotropy determination, while other regions are assigned control parameters, enabling self-organizing parameterization in design region. The parameters then are optimized using a constrained Bayesian optimization approach, aimed at maximizing the thermal performance factor (TPF) while constraining the area of low heat transfer regions. The optimization process begins with 216 samples to build the initial surrogate model, followed by 30 optimization iterations, resulting in the creation of high-performance complex structures and the establishment of a topology database. Furthermore, the SHAP method is utilized to extract topology design guidelines based on the most influential parameters and their beneficial variation directions. The guideline serves to guide the topology layout of other cooling structures in similar design scenarios. Results demonstrate that optimized self-organized structure enhanced the overall thermal parameter by 70% and decreased the low heat transfer zones by 74% when compared with pin fin structures. Data mining identified material orientation and generation on the left-side region as critical factors in topological structures, significantly influencing overall heat transfer performance. Leveraging the obtained design guidelines, a topological layout for a guiding pin fin structure is redesigned in wedge-shaped channels with varying contraction ratios, achieving performance improvements without the need for additional computational resources. The outcomes of this work hold scientific significance and industrial application value in enabling rapid design of high performance heat transfer structures.

A soft scanning electron microscopy for efficient segmentation of alloy microstructures based on a new self-supervised pre-training deep learning network

To provide an on-site metallographic segmentation using only optical microscopy images, sSEM-Net, a soft scanning electron microscopy network, is developed based on a self-supervised pre-training deep learning framework. During model training, only a sparse collection of SEM images is necessary for annotation assistance. By integrating CNN and Transformer, sSEM-Net efficiently utilizes global context information while mitigating data dependency and computational resource constraints. Using only readily available optical microscopy images as input, sSEM-Net achieves metallographic segmentation comparable to SEM images, catering to rapid and cost-effective industrial needs. This methodology leverages non-destructive inspection attributes, catering to rapid and cost-sensitive industrial requirements. The efficacy of the proposed sSEM-Net is demonstrated through metallographic structure analysis of TC4 titanium alloy, with potential extensions to other alloy types.

DEM simulation of the compression of crushable sand: does the initial particle shape matter?

Advances in DEM modeling, combined with high-resolution X-ray tomography, opened the way for computer models based on virtual replicas of the particles to preserve nearly all facets of their geometry. This leads to simulation advantages, but also high computational costs. Here we tackle a question stemming from this trend: how accurate should particle models be to ensure accuracy? We address this question for the case of the compression of crushable sand. LS-DEM was used to generate three models of Ottawa sand (exact replicas, ellipsoids, and spheres) from digital images of its grains. Compression-induced crushing was simulated for all sets by tracking evolving size and shape distribution. The results confirm that exact replicas provide the closest match of the measurements. However, intermediate degrees of rendering (e.g., ellipsoids preserving volume and aspect ratio of the real grains) led to satisfactory results only marginally different from those of exact replicas. These findings provide an example of the protocols that may be followed to identify the optimal degree of particle approximation which should be regarded as mandatory to achieve a conscious, sustainable use of computational resources.