Computational Cost Research Articles

Semantic Guided Matting Net

Abstract Human matting refers to extracting human parts from natural images with high quality, including human detail information such as hair, glasses, hats, etc. This technology plays an essential role in image synthesis and visual effects in the film industry. When the green screen is not available, the existing human matting methods need the help of additional inputs (such as trimap, background image, etc.), or the model with high computational cost and complex network structure, which brings great difficulties to the application of human matting in practice. To alleviate such problems, we use a segmentation network as the foundation and use multiple branches to achieve human segmentation, contour detail extraction, and information fusion. We also propose a foreground probability map module, which uses the feature maps in the segmentation network to pre-estimate the foreground probabilities of each pixel and obtain Semantic Guided Matting Net. Under the condition that only a single image is needed as the input, the human matting task can be realized by making full use of the semantic information in the image. We validate our method on the P3M-10k dataset. Compared with the benchmark, our method has made significant improvements in various evaluation indicators.

Uncovering suggestions in MOOC discussion forums: a transformer-based approach

The field of natural language processing has experienced significant advances in recent years, but these advances have not yet resulted in improved analytics for instructors on MOOC platforms. Valuable information, such as suggestions, is generated in the comment forums of these courses, but due to their volume, manual processing is often impractical. This study examines the feasibility of fine-tuning and effectively utilizing state-of-the-art deep learning models to identify comments that contain suggestions in MOOC forums. The main challenges encountered are the lack of labeled datasets from the MOOC context for fine-tuning classification models and the soaring computational cost of this training. For this study, we manually collected and labeled 2228 comments in Spanish and English from 5 MOOCs and scraped 1.4 million MOOC reviews from 3 platforms. We fine-tuned and evaluated 4 pretrained models based on the transformer architecture and 3 traditional machine learning models to compare their effectiveness in the suggestion mining task in this domain. Transformer-based models proved to be highly effective in this task/domain combination, achieving performance levels that matched or exceeded those deemed appropriate in other contexts and were significantly greater than those achieved by traditional models. Domain adaptation led to improved linguistic understanding of the target domain; however, in this project, this approach did not translate into an observable improvement in suggestion mining. The automated identification of comments that can be labeled as suggestions can result in considerable time savings for instructors, especially considering that less than a quarter of the analyzed comments contain suggestions.

ROBUST DESIGN OPTIMIZATION OF A COMPRESSOR ROTOR USING RECURSIVE COKRIGING BASED MULTI-FIDELITY UNCERTAINTY QUANTIFICATION AND MULTI-FIDELITY OPTIMIZATION

Abstract This work focuses on the application of multi-fidelity methods for the robust design optimization of engine components. The robust design optimization approach yields geometric designs with high efficiency and low sensitivity to uncertainties from manufacturing and wear. The uncertainty quantification techniques required to evaluate the robustness are computationally expensive, which limits their use in robust optimization. Multi-fidelity methods offer a promising solution to reduce the computational cost while maintaining the accuracy in both uncertainty quantification and optimization. A Kriging and a multi-fidelity recursive Cokriging framework are developed and applied to a test function. Furthermore, a multi-fidelity super efficient global optimization algorithm is developed. The optimizer is surrogate based and handles constraints. The developed methods are then applied to a compressor test case of a high pressure compressor blade row with 9 uncertainty and 24 design parameters of the geometry. The 2.5 % quantile of the stage efficiency is used as a robustness measure and is optimized. Design bounds and performance constraints are applied. Various uncertainty quantification techniques are analyzed. A multi-fidelity uncertainty quantification approach is developed that combines simplified coarse-grid low-fidelity with high-fidelity results to reduce the computational cost while maintaining a high accuracy. Uncertainty quantification techniques of three fidelity levels are developed and used for the multi-fidelity approach in the design space. The robust design optimization is performed and the optimal designs obtained from the multi-fidelity approach show superior performance compared to existing robust design optima in the literature.

One- and two-photon excitation dynamics using semiclassical electron force field model.

We have extended the semiclassical-based electron force-field simulation by introducing field-electron interaction to enable us to describe linear and nonlinear electronic excitation dynamics of a condensed matter system with low computational cost. To verify the simulation method, as a first step, numerical examples of interaction dynamics of simple systems (H atom, SiH4 molecule, and Si crystalline solid) with applied short electric field pulse as well as the obtained absorbed energies by the one- and two-photon excitations have been reported along with comparison with quantum dynamics calculations as reference.

Understanding and Predicting Ligand Efficacy in the μ-Opioid Receptor through Quantitative Dynamical Analysis of Complex Structures.

The μ-opioid receptor (MOR) is a G-protein coupled receptor involved in nociception and the primary target of opioid drugs. Understanding the relationships among the ligand structure, receptor dynamics, and efficacy in activating MOR is crucial for drug discovery and development. Here, we use coarse-grained normal-mode analysis to predict ligand-induced changes in receptor dynamics with the Quantitative Dynamics Activity Relationship (QDAR) DynaSig-ML methodology, training a LASSO regression model on the entropic signatures (ESs) computed from ligand-receptor complexes. We train and validate the methodology using a data set of 179 MOR ligands with experimentally measured efficacies split into strictly chemically different cross-validation sets. By analyzing the coefficients of the ES LASSO model, we identified key residues involved in MOR activation, several of which have mutational data supporting their role in MOR activation. Additionally, we explored a contact-only LASSO model based on ligand-protein interactions. While the model showed predictive power, it failed at predicting efficacy for ligands with low structural similarity to the training set, emphasizing the importance of receptor dynamics for predicting ligand-induced receptor activation. Moreover, the low computational cost of our approach, at 3 CPU s per ligand-receptor complex, opens the door to its application in large-scale virtual screening contexts. Our work contributes to a better understanding of dynamics-function relationships in the μ-opioid receptor and provides a framework for predicting ligand efficacy based on ligand-induced changes in receptor dynamics.

The Significance of Artificial Intelligence and Machine Learning in the Identification of Immunotherapy Targets for Cancer: Advances, Challenges, and Future Directions

Cancer immunotherapy has revolutionized cancer treatment by leveraging the immune system to target malignant cells, yet resistance in many cancers highlights the need for novel therapeutic targets. Artificial intelligence (AI) and machine learning (ML) have emerged as transformative tools for identifying new immunotherapy targets by analyzing vast datasets from genomics, proteomics, and clinical studies. This review explores the role of AI and ML in advancing the discovery of cancer-specific immunotherapy targets, such as tumor antigens and immune pathways. Key advances include the integration of big data, neoantigen prediction, biomarker discovery, and single-cell analysis. Despite these advancements, challenges remain, including data quality and standardization, interpretability of AI models, computational costs, and the need for biological validation of AI-driven discoveries. As AI and ML technologies continue to evolve, they hold the potential to overcome these barriers, leading to personalized immunotherapy solutions. This review also discusses future directions for AI-driven immunotherapy, emphasizing the need for improved models, ethical considerations, and clinical integration. Keywords: Artificial Intelligence, Machine Learning, Immunotherapy, Cancer, Advances, Challenges, Future Directions

Demand management of plug-in electric vehicle charging station considering bidirectional power flow using deep reinforcement learning

The recent development in plug-in electric vehicle technology has increased its popularity. Plug-in electric vehicles are widely used for their environment-friendly nature and they contribute to the reduction in global warming. With the increasing number of plug-in electric vehicles, charging coordination becomes essential for managing the charging station demand. The random charging behaviour of plug-in electric vehicles makes it a difficult task. This paper proposes a novel demand management strategy of charging stations. The proposed strategy supports the grid and reduces its burden during peak load. Deep-Q network based deep reinforcement learning is used in the proposed strategy. It is a value based deep reinforcement learning algorithm that approximates the Q value function using deep neural network. The deep reinforcement schedules the charging and discharging of plug-in electric vehicles to optimize the cost and manage the charging station load. Deep reinforcement learning enhances charging coordination by dynamically optimizing charging schedules according to real-time conditions and user preferences, thereby increasing efficiency and better integration with the grid. The reward function in deep reinforcement learning is designed based on the power price and demand of the charging station. A discount factor is introduced based on the service time to make the charging coordination efficient. A case study using dynamic pricing is carried out to validate the proposed strategy. The results prove that the proposed strategy optimizes charging and discharging costs and manages charging station demand efficiently. It is also observed that the proposed strategy is fast and incurs less computational cost.

Advantages of base isolation in reducing the reliability sensitivity to structural uncertainties of buildings

Base isolation is among the most effective techniques for safeguarding buildings against earthquakes. Similar to non-isolated buildings, the seismic performance of a base-isolated building can vary not only due to the inherent record-to-record variability in earthquake ground motions but also the non-negligible influence of uncertainties associated with structural properties. Introducing an isolation layer of isolation bearings and energy dissipators to the building further adds to the source of uncertainties, raising the question regarding the reliability of base isolated buildings. This study proposed a methodology that employs the probability density evolution method (PDEM) with local time grid refinement to evaluate the reliability of a base-isolated structure and its non-isolated counterpart under earthquakes of various magnitudes and fault distances. It allows for simultaneously considering the uncertainties of various sources in the reliability analyses with affordable computational cost, including those of the ground motions, of the structural nonlinearity, and of the ultimate states that define the reliability. The results show that the base-isolated building exhibits much higher reliability than its non-isolated counterpart in all cases even if the ultimate behavior such as the isolator fracture or the impact to the retaining wall is considered. The added source of uncertainties in the isolation layer does not exhibit a notable impact on the reliability. In contrast, the seismic reliability of base-isolated buildings is much less sensitive to structural uncertainties than that of their non-isolated counterparts.

Analysis of Cauchy reaction-diffusion equations involving Atangana-Baleanu fractional operator

This study investigates the Cauchy reaction-diffusion equation (CRDE) with the Atangana-Baleanu differential operator. The existence and uniqueness of solutions to fractional starting value issues are begun using the fixed-point theorem and contraction principle, respectively. The proposed study uses the natural variation iteration technique (NVIM) to get an approximate solution for nonlinear fractional reaction-diffusion equations. This study's approximate answers are compared to other solutions found using known methodologies, and the results are discussed. The devised technique has benefits in terms of accuracy and computational cost efficiency, which may be used to solve nonlinear fractional reaction-diffusion equations.

L2T-DFM: Learning to Teach with Dynamic Fused Metric

The loss function plays a crucial role in the construction of machine learning algorithms. Employing a teacher model to set loss functions dynamically for student models has attracted attention. In existing works, (1) the characterization of the dynamic loss suffers from some inherent limitations, ie, the computational cost of loss networks and the restricted similarity measurement handcrafted loss functions; and (2) the states of the student model are provided to the teacher model directly without integration, causing the teacher model to underperform when trained on insufficient amounts of data. To alleviate the above-mentioned issues, in this paper, we select and weigh a set of similarity metrics by a confidence-based selection algorithm and a temporal teacher model to enhance the dynamic loss functions. Subsequently, to integrate the states of the student model, we employ statistics to quantify the information loss of the student model. Extensive experiments demonstrate that our approach can enhance student learning and improve the performance of various deep models on real-world tasks, including classification, object detection, and semantic segmentation scenarios.

Hydrodynamic Shape Optimization of a Naval Destroyer by Machine Learning Methods

This paper explores the integration of advanced machine learning (ML) techniques within simulation-based design optimization (SBDO) processes for naval applications, focusing on the hydrodynamic shape optimization of the DTMB 5415 destroyer model. The use of unsupervised learning for design-space dimensionality reduction, combined with supervised learning through active learning-based multi-fidelity surrogate modeling, allows for significant improvements in computational efficiency while addressing complex, high-dimensional design spaces. By applying these ML techniques to both single- and multi-objective optimizations, aimed at minimizing resistance and enhancing seakeeping performance, the proposed framework demonstrates its practical value in hydrodynamic design. This approach provides a scalable and efficient solution, reducing the reliance on high-fidelity simulations while accelerating the optimization process, without substantial modifications to existing toolchains. A design-space dimensionality reduction of approximately 70% is achieved, reducing the design variables from 22 to 7 while retaining 95% of the original geometric variance. Additionally, computational cost reductions of 65% to 98% are observed, compared to using the full design space and high-fidelity simulations only.

A learnable motion preserving pooling for action recognition

Using deep neural networks (DNN) for video understanding tasks is expensive in terms of computation cost. Pooling layers in DNN which are widely used in most vision tasks to resize the spatial dimensions play crucial roles in reducing the computation and memory cost. In video-related tasks, pooling layers are also applied, mostly in the spatial dimension only as the standard average pooling in the temporal domain can significantly reduce its performance. This is because conventional temporal pooling degrades the underlying important motion features in consecutive frames. Such a phenomenon is rarely investigated and most state-of-art methods simply do not adopt temporal pooling, leading to enormous computation costs. In this work, we propose a learnable motion-preserving pooling (MPPool) layer that is able to preserve the general motion progression after the pooling. This pooling layer first locates the frames with the strongest motion features and then keeps these crucial features during pooling. Our experiments demonstrate that MPPool not only reduces the computation cost for video data modeling, but also increases the final prediction accuracy on various motion-centric and appearance-centric datasets.

Multiobjective Hybrid Al-Biruni Earth Namib Beetle Optimization and Deep Learning based Task Scheduling in Cloud Computing

With the rapid development of computing networks, cloud computing (CC) enables the deployment of large-scale applications and meets the increased rate of computational demands. Moreover, task scheduling is an essential process in CC. The tasks must be effectually scheduled across the Virtual Machines (VMs) to increase resource usage and diminish the makespan. In this paper, the multi-objective optimization called Al-Biruni Earth Namib Beetle Optimization (BENBO) with the Bidirectional-Long Short-Term Memory (Bi-LSTM) named as BENBO+ Bi-LSTM is developed for Task scheduling. The user task is subjected to the multi-objective BENBO, in which parameters like makespan, computational cost, reliability, and predicted energy are used to schedule the task. Simultaneously, the user task is fed to Bi-LSTM-based task scheduling, in which the VM parameters like average computation cost, Earliest Starting Time (EST), task priority, and Earliest Finishing Time (EFT) as well as the task parameters like bandwidth and memory capacity are utilized to schedule the task. Moreover, the task scheduling outcomes from the multi-objective BENBO and Bi-LSTM are fused for obtaining the final scheduling with less makespan and resource usage. Moreover, the predicted energy, resource utilization and makespan are considered to validate the BENBO+ Bi-LSTM-based task scheduling, which offered the optimal values of 0.669J, 0.535 and 0.381.

Deep operator neural network applied to efficient computation of asteroid surface temperature and the Yarkovsky effect

Surface temperature distribution is crucial for thermal property-based studies about irregular asteroids in our Solar System. While direct numerical simulations could model surface temperatures with high fidelity, they often take a significant amount of computational time, especially for problems for which temperature distributions are required to be repeatedly calculated. To this end, the deep operator neural network (DeepONet) proves a powerful tool due to its high computational efficiency and generalization ability. In this work, we apply DeepONet to the modeling of asteroid surface temperatures. Results show that the trained network is able to predict temperature with an accuracy of ~1% on average, while the computational cost is five orders of magnitude lower, enabling thermal property analysis in a multidimensional parameter space. As a preliminary application, we analyzed the orbital evolution of asteroids through direct N- body simulations embedded with an instantaneous Yarkovsky effect inferred by DeepONet-based thermophysical modeling. Taking asteroids (3200) Phaethon and (89433) 2001 WM41 as examples, we show the efficacy and efficiency of our AI-based approach.

Relaxing strong compatibility at atomistic-continuum interface: consistent linear coupling method

In this paper, we develop a brand-new coupling method called consistent linear coupling (CLC) with a view to resolving the difficulty of scaling down of the interface elements in the strong compatibility coupling (SCC), employed in the most accurate multiscale methods such as quasicontinuum. To this end, the CLC determines some conditions required to be satisfied to reproduce the SCC rigorously, regardless of the size of the interface elements. Therefore, it can produce a solution as accurate as the SCC with two additional features: significantly less computational cost and easier implementation. The CLC-based schemes are compared with well-known coupling methods including the SCC in terms of solving quasi-static 3D nanoscale contact and a 2D crystal under uniaxial tension. Numerical results demonstrate the superiority of the CLC with regard to the accuracy and computational efficiency.

Enhanced local multi-windows attention network for lightweight image super-resolution

Since the global self-attention mechanism can capture long-distance dependencies well, Transformer-based methods have achieved remarkable performance in many vision tasks, including single-image super-resolution (SISR). However, there are strong local self-similarities in images, if the global self-attention mechanism is still used for image processing, it may lead to excessive use of computing resources on parts of the image with weak correlation. Especially in the high-resolution large-size image, the global self-attention will lead to a large number of redundant calculations. To solve this problem, we propose the Enhanced Local Multi-windows Attention Network (ELMA), which contains two main designs. First, different from the traditional self-attention based on square window partition, we propose a Multi-windows Self-Attention (M-WSA) which uses a new window partitioning mechanism to obtain different types of local long-distance dependencies. Compared with original self-attention mechanisms commonly used in other SR networks, M-WSA reduces computational complexity and achieves superior performance through analysis and experiments. Secondly, we propose a Spatial Gated Network (SGN) as a feed-forward network, which can effectively overcome the problem of intermediate channel redundancy in traditional MLP, thereby improving the parameter utilization and computational efficiency of the network. Meanwhile, SGN introduces spatial information into the feed-forward network that traditional MLP cannot obtain. It can better understand and use the spatial structure information in the image, and enhances the network performance and generalization ability. Extensive experiments show that ELMA achieves competitive performance compared to state-of-the-art methods while maintaining fewer parameters and computational costs.

A computationally efficient queue-based algorithm for simulating volume-controlled drainage under the influence of gravity on volumetric images of porous materials

Simulating non-wetting fluid invasion in volumetric images of porous materials is of broad interest in applications as diverse as electrochemical devices and CO2 sequestration. Among available methods, image-based algorithms offer much lower computational cost compared to direct numerical simulations. Recent work has extended image-based methods to incorporate more physics such as gravity and volume-controlled invasion. The present work combines these two developments to develop an image-based invasion percolation algorithm that incorporates the effect of gravity. Additionally, the presented algorithm was developed using a priority queue algorithm to drastically reduce the computational cost of the simulation. The priority queue-based method was validated against previous image-based methods both with and without the effect of gravity, showing identical results. It was also shown that the new method provides a speedup of 20X over the previous image-based methods. Finally, comparison with experimental results at three Bond numbers showed that the model can predict the real invasion process with a high accuracy with and without gravitational effects.

On the shortcut learning in multilingual neural machine translation

In this study, we revisit the commonly-cited off-target issue in multilingual neural machine translation (MNMT). By carefully designing experiments on different MNMT scenarios and models, we attribute the off-target issue to the overfitting of the shortcuts of (non-centric, centric) language mappings. Specifically, the learned shortcuts biases MNMT to mistakenly translate non-centric languages into the centric language instead of the expected non-centric language for zero-shot translation. Analyses on learning dynamics show that the shortcut learning generally occurs in the later stage of model training, and multilingual pretraining accelerates and aggravates the shortcut learning. Based on these observations, we propose a simple and effective training strategy to eliminate the shortcuts in MNMT models by leveraging the forgetting nature of model training. The only difference from the standard training is that we remove the training instances that may induce the shortcut learning in the later stage of model training. Without introducing any additional data and computational costs, our approach can consistently and significantly improve the zero-shot translation performance by alleviating the shortcut learning for different MNMT models and benchmarks.

Traffic Prevention and Security Enhancement in VANET Using Deep Learning With Trusted Routing Aided Blockchain Technology

ABSTRACTVehicular ad hoc networks (VANETs) in portable broadband networks are a revolutionary concept with enormous potential for developing safe and efficient transportation systems. Because VANETs are open networks that require regular information sharing, it might be difficult to ensure the security of data delivered through VANETs as well as driver privacy. This paper proposes a blockchain technology that supports trusted routing and deep learning for traffic prevention and security enhancement in VANETs. Initially, the proposed Feature Attention‐based Extended Convolutional Capsule Network (FA_ECCN) model predicts the driver's behaviors such as normal, drowsy, distracted, fatigued, aggressive, and impaired. Next, the Binary Fire Hawks‐based Optimized Link State Routing Protocol (BFH_OLSRP) is used to route traffic after trust values have been assessed. Furthermore, Binary Fire Hawks Optimization (BFHO) determines the best routing path based on criteria such as link stability and node stability degree. Finally, blockchain storage is supported by the Interplanetary File System (IPFS) technology to improve the security of VANET data. Additionally, the validation process is established by using Delegated Practical Byzantine Fault Tolerance (DPBFT). As a result, the proposed study employs the blockchain system to securely send data to neighboring vehicles via trust‐based routing, thereby accurately predicting the driver's behavior. The proposed method achieves a better outcome in terms of latency, packet delivery ratio (PDR), overhead packets, throughputs, end‐to‐end delay, transmission overhead, and computational cost. According to simulation results and efficiency evaluation, the proposed approach outperforms existing approaches and enhances vehicle communication security in an effective manner.

Discretely Nonlinearly Stable Weight-Adjusted Flux Reconstruction High-Order Method for Compressible Flows on Curvilinear Grids

To achieve genuine predictive capability, an algorithm must consistently deliver accurate results over prolonged temporal integration periods, avoiding the unwarranted growth of aliasing errors that compromise the discrete solution. Provable nonlinear stability bounds the discrete approximation and ensures that the discretization does not diverge. Nonlinear stability is accomplished by satisfying a secondary conservation law, namely for compressible flows; the second law of thermodynamics. For high-order methods, discrete nonlinear stability and entropy stability, have been successfully implemented for discontinuous Galerkin (DG) and residual distribution schemes, where the stability proofs depend on properties of L2-norms. In this paper, nonlinearly stable flux reconstruction (NSFR) schemes are developed for three-dimensional compressible flow in curvilinear coordinates. NSFR is derived by merging the energy stable flux reconstruction (ESFR) framework with entropy stable DG schemes. NSFR is demonstrated to use larger time-steps than DG due to the ESFR correction functions, at the cost of larger error levels at design order convergence for equivalent degrees of freedom, while preserving discrete nonlinear stability. NSFR differs from ESFR schemes in the literature since it incorporates the FR correction functions on the volume terms through the use of a modified mass matrix. We also prove that discrete kinetic energy stability cannot be preserved to machine precision for quadrature rules where the surface quadrature is not a subset of the volume quadrature. This result stems from the inverse mapping from the kinetic energy variables to the conservative variables not existing for the kinetic energy projected variables. This paper also presents the NSFR modified mass matrix in a weight-adjusted form. This form reduces the computational cost in curvilinear coordinates because the dense matrix inversion is approximated by a pre-computed projection operator and the inverse of a diagonal matrix on-the-fly and exploits the tensor product basis functions to utilize sum-factorization. The nonlinear stability properties of the scheme are verified on a nonsymmetric curvilinear grid for the inviscid Taylor-Green vortex problem and the correct orders of convergence were obtained on a curvilinear mesh for a manufactured solution. Lastly, we perform a computational cost comparison between conservative DG, overintegrated DG, and our proposed entropy conserving NSFR scheme, and find that our proposed entropy conserving NSFR scheme is computationally competitive with the conservative DG scheme.