Global Computation Research Articles

Lightweight skin cancer detection IP hardware implementation using cycle expansion and optimal computation arrays methods

Skin cancer is recognized as one of the most perilous diseases globally. In the field of medical image classification, precise identification of early-stage skin lesions is imperative for accurate diagnosis. However, deploying these algorithms on low-cost devices and attaining high-efficiency operation with minimal energy consumption poses a formidable challenge due to their intricate computational demands. This study proposes a lightweight hardware design based on a convolutional neural network (CNN) for real-time processing of skin disease classifiers on portable devices. Our skin cancer recognition processor utilizes an optimally parallel designed processing engine (PE) for global computation, which greatly reduces hardware resource utilization by multiplexing of computational unit circuits. In addition, a design approach that provides loop unrolling effectively reduces the number of data accesses, thereby reducing computational complexity and logic resource requirements. The hardware circuits in this design perform data inference in convolutional, pooling, and fully connected layers based on 16-bit floating-point numbers. Evaluation of the HAM10000 database dataset shows that the architecture achieves an average classification accuracy of 97.8 %. We are the first to implement an all-hardware FPGA-based skin cancer detection platform that offers a 3.5x speedup in recognition compared to existing skin cancer accelerators at 50 MHz while consuming only 0.48 W of power. The implementation of this hardware architecture meets the major constraints of portable devices, featuring low resource utilization, low power consumption, and cost-effectiveness, while still providing efficient classification and high accuracy results.

Electromagnetic force calculation within finite volume framework

ABSTRACT In this research, the comparison of force calculation methods in magnetostatic simulations within a cell-centered finite volume framework is presented. Various procedures for postprocessing of the magnetic vector potential field are laid out to obtain the surface and volume integrals of Maxwell’s stress tensor and the Lorentz force density. Two different magnetostatic problems with multiple operating points were modeled with the AVL FIRETM M. The force results have been verified by comparing them to the Ansys Maxwell and were validated on a benchmark case from a Testing Electromagnetic Analysis Methods (TEAM) workshop. The interpolation of the magnetic flux density on a multi-material interface is presented, and it is used in conjunction with the surface integral of Maxwell’s stress tensor for an efficient global force computation. For all observed cases, the calculated results correspond well to the results found in the literature. It was found that the accuracy of the volume integral of Maxwell’s stress tensor is dependent on the mesh quality and gradient calculation scheme. Employing the semi-structured mesh with Gauss’s gradient scheme results in force discrepancies of less than 3% between the surface and volume integral of Maxwell’s stress tensor. The convergence of the iterative solver is observed to be slower in the case of the materials with permeability jump, where discontinuities between the tangential component of the magnetic flux density and the normal component of the magnetic field intensity occur.

E-commerce and Mobile Application Development in the Sports Industry

In order to achieve user recommendations that best match their current contextual needs, the author proposes a mobile service QoS (Quality of Service) hybrid recommendation model based on sports user situational awareness. Cluster the users and service items covered by mobile users based on their location contextual information according to the classification principle of autonomous systems, forming a collaborative filtering and recommendation mechanism for mobile user service items; In response to the cold start problem of new users and new projects in traditional QoS recommendation methods, prediction and recommendation of missing QoS attribute preference values are based on User based and Item based CF; In response to the problem of difficult determination of mixed recommendation weights caused by massive data and uneven distribution of service QoS attribute values in mobile network environments, MapReduce based ant colony neural network weight training is used for CF mixed recommendation. Experimental results have shown that the Hadoop sports industry has improved the operational efficiency of algorithms and reduced the time for global QoS preference prediction; And by comparing the operation results of the 10% and 100% sub datasets, it can also be seen that the acceleration ratio of the algorithm will continuously improve with the increase of data volume, thereby improving the operational efficiency of the recommendation algorithm. It has been proven that the MapReduce based ant colony neural network weight training method significantly reduces the global computation time of the algorithm and improves the operational efficiency of the recommendation system.

LGIT: local–global interaction transformer for low-light image denoising

Transformer-based methods effectively capture global dependencies in images, demonstrating outstanding performance in multiple visual tasks. However, existing Transformers cannot effectively denoise large noisy images captured under low-light conditions owing to (1) the global self-attention mechanism causing high computational complexity in the spatial dimension owing to a quadratic increase in computation with the number of tokens; (2) the channel-wise self-attention computation unable to optimise the spatial correlations in images. We propose a local–global interaction Transformer (LGIT) that employs an adaptive strategy to select relevant patches for global interaction, achieving low computational complexity in global self-attention computation. A top-N patch cross-attention model (TPCA) is designed based on superpixel segmentation guidance. TPCA selects top-N patches most similar to the target image patch and applies cross attention to aggregate information from them into the target patch, effectively enhancing the utilisation of the image’s nonlocal self-similarity. A mixed-scale dual-gated feedforward network (MDGFF) is introduced for the effective extraction of multiscale local correlations. TPCA and MDGFF were combined to construct a hierarchical encoder-decoder network, LGIT, to compute self-attention within and across patches at different scales. Extensive experiments using real-world image-denoising datasets demonstrated that LGIT outperformed state-of-the-art (SOTA) convolutional neural network (CNN) and Transformer-based methods in qualitative and quantitative results.

Efficient full-aperture mirror polishing method with variable orbital radius in computer-controlled optical surfacing.

Optical systems in astronomy have extremely high requirements on the full-aperture surface precision and fabrication efficiency of aspherical mirrors. However, the current full-aperture optics fabrication method suffers from both fabrication and computation inefficiency. The former is caused by the isolated polishing strategy for the inner and edge regions of the mirror with different tools, while the latter is caused by the global computation strategy for the two regions. In this paper, a full-aperture mirror polishing method with the reversed strategy is proposed to solve this problem. Firstly, the dwell time of inner/edge regions are respectively calculated by the deconvolution method and the linear equations method based on the space-variant tool influence function. Therefore, both the computation cost and the edge polishing error can be reduced. Then, a fused tool path is developed to achieve variable orbital radius in the inner/edge region so the full aperture can be polished in one round. Simulations and experiments using a SiC aspherical mirror and large segmented mirrors demonstrate that the surface error can converge quickly in the full aperture. As a consequence, the full-aperture fabrication accuracy and efficiency can be greatly improved through the proposed method.

Localized and global representation of prior value, sensory evidence, and choice in male mouse cerebral cortex.

Adaptive behavior requires integrating prior knowledge of action outcomes and sensory evidence for making decisions while maintaining prior knowledge for future actions. As outcome- and sensory-based decisions are often tested separately, it is unclear how these processes are integrated in the brain. In a tone frequency discrimination task with two sound durations and asymmetric reward blocks, we found that neurons in the medial prefrontal cortex of male mice represented the additive combination of prior reward expectations and choices. The sensory inputs and choices were selectively decoded from the auditory cortex irrespective of reward priors and the secondary motor cortex, respectively, suggesting localized computations of task variables are required within single trials. In contrast, all the recorded regions represented prior values that needed to be maintained across trials. We propose localized and global computations of task variables in different time scales in the cerebral cortex.

On the circuit model of two quantum adiabatic search algorithms

The early work of Wei et al. has shown that it is possible to do the computation in constant time for quantum search in quantum local adiabatic evolution framework if the system Hamiltonian has an extra enlarged factor proportional to the square-root of the size of the problem. Also in our prior work, we have shown that it is possible to apply a linear interpolating path in the system Hamiltonian to achieve an O(1) time complexity for a quantum global adiabatic computation. In this paper, we discuss the issues when implementing these two quantum algorithms on the quantum circuit model, and find that to our surprise, the time slices produced does not equal to the time complexity of the algorithm in each case. This is in contrast to the previous related results, because there these two quantities always coincide. When taking into account of the whole energy-level increase of the system compared with that of the usual quantum adiabatic evolutions and the corresponding time complexity simultaneously, we find that the time slices needed always consist of the two quantities multiplying together. This result is new, and may be hopeful to find application in designing quantum adiabatic algorithm for problems beyond quantum search.

An efficient energy management strategy based on heuristic dynamic programming specialized for hybrid electric unmanned delivery aerial vehicles

Large-sized unmanned delivery aerial vehicles (UDAVs) stand out as prominent vertical take-off and landing aircraft, benefiting from integrating a hybrid electric propulsion system (HEPS) to enhance the power-to-weight ratio. Despite this advancement, UDAVs encounter challenges regarding range limitations and the need for swift computational solutions. Addressing these issues necessitates a computationally efficient energy management strategy (EMS) that optimizes power distribution to properly plan engine working points, ultimately aimed at enhancing fuel economy to increase flight range. This study proposes a heuristic dynamic programming (HDP)-based EMS tailored for the HEPS of a large-sized UDAV. In this strategy, the HDP framework is adopted to improve computational efficiency by iteratively updating control variables. To account for the specific flight characteristics, a flight state identifier incorporating selected flight command flags applicable to UDAVs is designed. This identifier aids in predicting velocity changes for different flight states by amalgamating corresponding prediction networks into an integrated velocity prediction network (IVPN). Substituting the model network with IVPN in the HDP framework significantly enhances prediction accuracy and guides power distribution towards optimal results, signifying an efficient EMS specialized for UDAVs. The proposed strategy is evaluated against baseline strategies in designed flight scenarios, highlighting a remarkable 51.37% improvement in prediction accuracy, a noteworthy 6.04% reduction in equivalent fuel consumption, and an impressive 95.81% reduction in global computation time. Finally, the actual applicability of the proposed strategy is validated in a hardware-in-loop test. Consequently, the proposed strategy can provide theoretical support for the efficient energy management of diverse large-sized UDAVs.

Locally Abstract, Globally Concrete Semantics of Concurrent Programming Languages

Formal, mathematically rigorous programming language semantics are the essential prerequisite for the design of logics and calculi that permit automated reasoning about concurrent programs. We propose a novel modular semantics designed to align smoothly with program logics used in deductive verification and formal specification of concurrent programs. Our semantics separates local evaluation of expressions and statements performed in an abstract, symbolic environment from their composition into global computations, at which point they are concretised. This makes incremental addition of new language concepts possible, without the need to revise the framework. The basis is a generalisation of the notion of a program trace as a sequence of evolving states that we enrich with event descriptors and trailing continuation markers. This allows to postpone scheduling constraints from the level of local evaluation to the global composition stage, where well-formedness predicates over the event structure declaratively characterise a wide range of concurrency models. We also illustrate how a sound program logic and calculus can be defined for this semantics.

A modified atmospheric scattering model and degradation image clarification algorithm for haze environments

Addressing the limitation of existing image dehazing algorithms based on atmospheric scattering models, which often overlook the underlying physical causes of color distortion in hazy images, resulting in suboptimal algorithm robustness and image clarity when faced with varying levels of haze pollution, this study introduces a refined atmospheric scattering model. It also presents a haze image clarity algorithm based on this modified atmospheric scattering model by analyzing the fundamental factors contributing to image degradation and distortion caused by haze. Initially, the hazy image undergoes conversion to a new HSI color space. The hazy image is aligned more closely with human visual perception characteristics by implementing an adaptive saturation enhancement technique and adjusting contribution coefficients for the three RGB channels to the luminance component. Subsequently, the hazy image is partitioned into dense haze and non-dense haze regions based on haze distribution characteristics within the image. A global light intensity computation method is devised for the hazy image utilizing a quad-tree segmentation algorithm. Following this, calculations are performed to determine the atmospheric light value and transmittance for dense and non-dense haze regions separately. The global atmospheric light and transmittance values are then derived based on the probability density function of the dense haze region using Alpha Fusion. Finally, corrections are applied to the global atmospheric light values using an atmospheric light correction vector, while denoising the global transmittance is achieved through a fast non-local mean filtering algorithm. Experimental findings demonstrate that the proposed haze image clarity algorithm, grounded in the modified atmospheric scattering model, exhibits adaptability across various degrees and types of haze images. Comparative analysis against other prominent image dehazing algorithms reveals its efficacy in addressing severe color distortion in dense haze scenarios, achieving a more natural clarity effect, and showcasing superior performance across all numerical evaluation metrics.

Learning Stiffness Tensors in Self-Activated Solids via a Local Rule.

Mechanical metamaterials are often designed with particular properties for specific load-bearing functions. Alternatively, this study aims to create a class of active lattice metamaterials, dubbed self-activated solids, that can learn desired stiffness tensors from the elastic deformations they experienced, a crucial feature to improve the performance, efficiency, and functionality of materials. Artificial adaptive matters that combine sensory, control, and actuation elements can offer appealing solutions. However, challenges still remain: The designs will rely on accurate off-line and global computations, as well as intricate coordination among individual elements. Here, a simple online and local learning strategy is initiated based on contrastive Hebbian learning to gradually guide self-activated solids to possess sought-after stiffness tensors autonomously and reversibly. During learning, the bond stiffness of the active lattice varies depending only on its local strain. The numerical tests show that the self-activated solid can not only achieve the desired bulk, shear, and coupling moduli but also manifest uni-mode and bi-mode extremal materials by itself after experiencing the corresponding elastic deformations. Further, the self-activated solid can also achieve the desired time-varying moduli when exposed to temporally different loads. The design is applicable to any lattice geometries and is resistant to damage and instabilities. The material design approach and the physical learning strategy suggested can benefit the design of autonomous materials, physical learning machines, and adaptiverobots.

FedTKD: A Trustworthy Heterogeneous Federated Learning Based on Adaptive Knowledge Distillation.

Federated learning allows multiple parties to train models while jointly protecting user privacy. However, traditional federated learning requires each client to have the same model structure to fuse the global model. In real-world scenarios, each client may need to develop personalized models based on its environment, making it difficult to perform federated learning in a heterogeneous model environment. Some knowledge distillation methods address the problem of heterogeneous model fusion to some extent. However, these methods assume that each client is trustworthy. Some clients may produce malicious or low-quality knowledge, making it difficult to aggregate trustworthy knowledge in a heterogeneous environment. To address these challenges, we propose a trustworthy heterogeneous federated learning framework (FedTKD) to achieve client identification and trustworthy knowledge fusion. Firstly, we propose a malicious client identification method based on client logit features, which can exclude malicious information in fusing global logit. Then, we propose a selectivity knowledge fusion method to achieve high-quality global logit computation. Additionally, we propose an adaptive knowledge distillation method to improve the accuracy of knowledge transfer from the server side to the client side. Finally, we design different attack and data distribution scenarios to validate our method. The experiment shows that our method outperforms the baseline methods, showing stable performance in all attack scenarios and achieving an accuracy improvement of 2% to 3% in different data distributions.

Ore image segmentation Based on Multiscale Parallel Efficient Channel Attention U-Network

Accurately detecting ore particle size and applying them to the control of semi-autogenous grinding mills are important for improving grinding productivity. Due to the wide range of ore size distribution and mutual adhesion of ores when entering the mill, the current methods based on deep learning image segmentation suffer from significant limitations and poor segmentation effectiveness in practical application. To address this issue, we propose a multi-scale parallel efficient channel attention U-net (MPECA-Unet) segmentation model. The proposed method introduces two-dimensional global attention computation in both spatial and channel domains. It also integrates multiscale fusion information with the encoder's varied scale features to improve segmentation performance for ore image. Experimental results show that the segmentation performance of the model is superior, with smaller error rate when compared to the on-site manual screening results.

Benchmarking Supervised and Self-Supervised Learning Methods in A Large Ultrasound Muti-task Images Dataset.

Deep learning in ultrasound(US) imaging aims to construct foundational models that accurately reflect the modality's unique characteristics. Nevertheless, the limited datasets and narrow task types have restricted this field in recent years. To address these challenges, we introduce US-MTD120K, a multi-task ultrasound dataset with 120,354 real-world two-dimensional images. This dataset covers three standard plane recognition and two diagnostic tasks in ultrasound imaging, providing a rich basis for model training and evaluation. We detail the data collection, distribution, and labelling processes, ensuring a thorough understanding of the dataset's structure. Furthermore, we conduct extensive benchmark tests on 27 state-of-the-art methods from both supervised and self-supervised learning(SSL) perspectives. In the realm of supervised learning, we analyze the sensitivity of two main feature computation methods to ultrasound images at the representational level, highlighting that models which judiciously constrain global feature computation could potentially serve as a viable analytical approach for US image analysis. In the context of self-supervised learning, we delved into the modelling process of self-supervised learning models for medical images and proposed an improvement strategy, named MoCo-US, a solution that addresses the excessive reliance on pretext task design from the input side. It achieves competitive performance with minimal pretext task design and enhances other SSL methods simply. The dataset and the code will be available at https://github.com/JsongZhang/CDOA-for-UMTD.

Evolution of wireless communication networks from 5G to 6G future perspective

Industry 4.0 relies heavily on artificial-intelligence(AI) and cloud-computing(CC), both of which have been greatly aided by the 5th-generation mobile-network (5G). The arrival of 5G, however, is seen as a watershed moment that will radically alter the current global trends in wireless communication practices and the lives of the masses. 5G envisions a future where the digital and physical worlds merge. The 6th-generation (6G) wireless communication network will likely unite terrestrial, aerial, and maritime communications into a single, unified system that is both more stable and faster, and can accommodate a far larger number of devices with ultra-low latency needs. This research hopes to foresee a scenario in which 6G supersedes 5G as the dominant standard for wireless communication in the years to come. Several advances have been made, but the utopian period of instantaneous global communication, instantaneous global computation, and no latency has not yet arrived. This essay investigates the most significant obstacles and difficulties that the transition from 5G to 6G may encounter on the way to realizing these loftier goals. The vital goal of "technology for humanity" is to improve service to the world's most disadvantaged people, and this article lays out a plan for 6G that includes the enabling technology infrastructures, obstacles, and research directions that will get us there.

Floating bridge global responses with hydrodynamic interaction

Hydrodynamic interaction between pontoons is studied for the global response computation of a floating bridge. Different issues related to bridge hydrodynamic interaction modeling are discussed. The pontoon's hydrodynamic coefficients become highly oscillating in the wind wave region under interaction. Wave trapping was observed in the numerical solution and proved by a decay test of the mid-pontoon in a three-pontoon model test. Model test results were compared with computations applying different free surface damping levels, which proved no free surface damping was needed for the current design. To facilitate the time-domain simulation of a bridge with many pontoons, simplification is necessary. It is proposed that the mid-pontoon hydrodynamic coefficients be applied instead of the coupled full hydrodynamics matrices of all the pontoons. By modal analysis, it was shown that dense frequencies in the hydrodynamic-interaction-affected region are important for the identification of bridge modes. Computation was compared with model tests in the ocean basin, where one, three pontoons and a hydro-elastic bridge were tested. The proposed simplification strategy was validated. Further, the comparison shows that hydrodynamic interaction is important in the global analysis of the floating bridge in the study.

ADRLO: Adaptive deep reinforcement learning-based offloading for edge computing

Computation offloading, as an efficient and promising computing paradigm for mobile edge computing (MEC), provides optimal computation allocation between MEC servers and Internet of Things (IoT) devices. However, in modern wireless communications, most of the computational tasks are latency-sensitive, presenting an efficiency challenge for offloading services. Moreover, to make the best use of computation resources, it remains an issue to ensure stable optimal offloading for the allocation of computation resources. In this paper, a novel adaptive deep reinforcement learning-based offloading (ADRLO) framework is developed to achieve a stable optimal computation revenue with low execution latency for MEC offloading. Structurally, we implement a residual connected deep neural network (DNN) to generate a series of offloading strategies and use them to obtain the global system computation rate. Especially, the strategy that achieves the maximum computation rate is then selected as experience to support the DNN training, forming a closed-loop deep reinforcement learning (DRL) mechanism. To further enhance the efficiency and stability of the MEC offloading, an adaptive regulator is presented for the framework parameters including the size of action space and the DNN learning rate. Numerical results show that, compared to existing state-of-the-art approaches, our proposal achieves optimal computation revenue, exhibits better convergence performance, and significantly reduces the execution latency by almost 50%.

A return mapping algorithm based on the hyper dual step derivative approximation for elastoplastic models

Accurately evaluating derivatives poses a key challenge when numerically implementing complex constitutive models. This work presents an implicit stress update algorithm that utilizes the hyper dual step derivative approximation to address derivative evaluations in elastoplastic problems. Initially, the performance of various numerical differentiation methods is discussed and compared by examining their numerical errors in the representative example. Subsequently, the hyper dual step derivative approximation, without truncation and subtractive cancellation errors, is employed to compute the Jacobian matrix and consistent tangent operator, ensuring quadratic convergence in both local and global computations. The size of the Newton search step is optimized by the line search technique, thereby enhancing the convergence in solving nonlinear stress integral equations. Finally, the proposed stress update algorithm is used to implement the non-associated Mohr–Coulomb plastic model in the ABAQUS software using the UMAT subroutine. The stress update algorithm’s performance and its practical application in geotechnical engineering problems are demonstrated using five boundary value problems.

DPACFuse: Dual-Branch Progressive Learning for Infrared and Visible Image Fusion with Complementary Self-Attention and Convolution.

Infrared and visible image fusion aims to generate a single fused image that not only contains rich texture details and salient objects, but also facilitates downstream tasks. However, existing works mainly focus on learning different modality-specific or shared features, and ignore the importance of modeling cross-modality features. To address these challenges, we propose Dual-branch Progressive learning for infrared and visible image fusion with a complementary self-Attention and Convolution (DPACFuse) network. On the one hand, we propose Cross-Modality Feature Extraction (CMEF) to enhance information interaction and the extraction of common features across modalities. In addition, we introduce a high-frequency gradient convolution operation to extract fine-grained information and suppress high-frequency information loss. On the other hand, to alleviate the CNN issues of insufficient global information extraction and computation overheads of self-attention, we introduce the ACmix, which can fully extract local and global information in the source image with a smaller computational overhead than pure convolution or pure self-attention. Extensive experiments demonstrated that the fused images generated by DPACFuse not only contain rich texture information, but can also effectively highlight salient objects. Additionally, our method achieved approximately 3% improvement over the state-of-the-art methods in MI, Qabf, SF, and AG evaluation indicators. More importantly, our fused images enhanced object detection and semantic segmentation by approximately 10%, compared to using infrared and visible images separately.

A fast LiDAR place recognition and localization method by fusing local and global search

Place recognition is an important branch of Simultaneous Localization and Mapping (SLAM), which finds revisited places, thereby reducing error accumulations. Most mainstream methods only concentrate on the similarity of two LiDAR scans and cannot obtain the pose information. In this paper, we design a descriptor named Occupied Place Description (OPD) and propose an efficient map-aided place recognition and localization method. Our method can effectively detect loop events, obtain the vehicle’s poses, and overcome the overlap dependence of pairwise point cloud registration. It is composed of four modules. First, the map data generation is designed to store the map database as binary files. Then, a trajectory initialization consisting of global search, local search, and trajectory score computation is proposed to determine the vehicle’s initial poses. Next, an online localization method is developed by combining descriptor similarity and Kalman Filtering, which only takes less than 5ms per localization. Finally, loop events are detected based on pose information. Exhaustive evaluations on KITTI and MulRan datasets demonstrate the superiority of our method. The average absolute trajectory error (ATE) is 0.5 m and the average relative rotation error is 2.7 degrees. F1 score measures the success rate of place recognition and our average maximum F1 score achieves nearly 0.987. The source code will be available in https://github.com/ShiPC-AI/Occupied-Place-Description.