Lightweight Strategy Research Articles

Dual-grained Lightweight Strategy.

Removing redundant parameters and computations before the model training has attracted a great interest as it can effectively reduce the storage space of the model, speed up the training and inference of the model, and save energy consumption during the running of the model. In addition, the simplification of deep neural network models can enable high-performance network models to be deployed to resource-constrained edge devices, thus promoting the development of the intelligent world. However, current pruning at initialization methods exhibit poor performance at extreme sparsity. In order to improve the performance of the model under extreme sparsity, this paper proposes a dual-grained lightweight strategy-TEDEPR. This is the first time that TEDEPR has used tensor theory in the pruning at initialization method to optimize the structure of a sparse sub-network model and improve its performance. Specifically, firstly, at the coarse-grained level, we represent the weight matrix or weight tensor of the model as a low-rank tensor decomposition form and use multi-step chain operations to enhance the feature extraction capability of the base module to construct a low-rank compact network model. Secondly, unimportant weights are pruned at a fine-grained level based on the trainability of the weights in the low-rank model before the training of the model, resulting in the final compressed model. To evaluate the superiority of TEDEPR, we conducted extensive experiments on MNIST, UCF11, CIFAR-10, CIFAR-100, Tiny-ImageNet and ImageNet datasets with LeNet, LSTM, VGGNet, ResNet and Transformer architectures, and compared with state-of-the-art methods. The experimental results show that TEDEPR has higher accuracy, faster training and inference, and less storage space than other pruning at initialization methods under extreme sparsity.

Sustainable Engineered Designs and Manufacturing of Waste Derived Graphenes Reinforced Polypropylene Composite for Automotive Interior Parts.

The automotive sector is actively pursuing a lightweighting strategy as a means to urgently decrease greenhouse gas emissions, which are a significant driver of climate change. The development of lightweight composite structures has been identified as crucial for enhancing part performance while mitigating negative environmental impacts and adopting energy-efficient manufacturing methods. This comprehensive study aimed to decrease the main reinforcement content of talc in commercial compounds while integrating graphene derived from waste polypropylene (PP) grown on talc and graphene nanoplatelet obtained from waste tires by upcycling processes into the PP compound. The entire value chain of interior automotive part production, from compound development and scaling up with a high-shear mixer, to injection molding of the part and performance tests, was investigated with a focus on sustainability considerations. The successful integration of 4 wt % micron talc, together with 1 wt % graphene nanoparticles and 1 wt % hybrid additive into the blended HomoPP/CopoPP matrix resulted in a 10% weight reduction compared to the conventional part. Moreover, significant improvements in flexural and tensile strength were observed, with enhancements of 52 and 38%, respectively. The uniform dispersion of additives and improved interfacial adhesion between the PP matrix and additives facilitated efficient stress transfer, contributing to enhanced mechanical properties. Furthermore, a systematic life cycle assessment study demonstrated the positive impact of waste PP incorporation on CO2 reduction, achieving a remarkable 95% reduction compared to virgin PP. The developed compound also demonstrated favorable processability and flow properties, supporting its potential for mass production. Overall, this study presents a sustainable and effective approach for lightweight automotive interior part production using a synergistically designed PP compound meeting the requirements of the automotive industry.

Introducing Security Mechanisms in OpenFog-Compliant Smart Buildings

Designing smart building IoT applications is a complex task. It requires efficiently integrating a broad number of heterogeneous, low-resource devices that adopt lightweight strategies. IoT frameworks, especially if they are standard-based, may help designers to scaffold the applications. OpenFog, established as IEEE 1934 standard, promotes the use of free open source (FOS) technologies and has been identified for use in smart buildings. However, smart building systems may present vulnerabilities, which can put their integrity at risk. Adopting state-of-the-art security mechanisms in this domain is critical but not trivial. It complicates the design and operation of the applications, increasing the cost of the deployed systems. In addition, difficulties may arise in finding qualified cybersecurity personnel. OpenFog identifies the security requirements of the applications, although it does not describe clearly how to implement them. This article presents a scalable architecture, based on the OpenFog reference architecture, to provide security by design in buildings of different sizes. It adopts FOS technologies over low-cost IoT devices. Moreover, it presents guidelines to help developers create secure applications, even if they are not security experts. It also proposes a selection of technologies in different layers to achieve the security dimensions defined in the X.805 ITU-T recommendation. A proof-of-concept Indoor Environment Quality (IEQ) system, based on low-cost smart nodes, was deployed in the Faculty of Engineering of Vitoria-Gasteiz to illustrate the implementation of the presented approach. The operation of the IEQ system was analyzed using software tools frequently used to find vulnerabilities in IoT applications. The use of state-of-the-art security mechanisms such as encryption, certificates, protocol selection and network partitioning/configuration in the OpenFog-based architecture improves smart building security.

Enhancing Earth data analysis in 5G satellite networks: A novel lightweight approach integrating improved deep learning

Efficiently handling huge data amounts and enabling processing-intensive applications to run in faraway areas simultaneously is the ultimate objective of 5G networks. Currently, in order to distribute computing tasks, ongoing studies are exploring the incorporation of fog-cloud servers onto satellites, presenting a promising solution to enhance connectivity in remote areas. Nevertheless, analyzing the copious amounts of data produced by scattered sensors remains a challenging endeavor. The conventional strategy of transmitting this data to a central server for analysis can be costly. In contrast to centralized learning methods, distributed machine learning (ML) provides an alternative approach, albeit with notable drawbacks. This paper addresses the comparative learning expenses of centralized and distributed learning systems to tackle these challenges directly. It proposes the creation of an integrated system that harmoniously merges cloud servers with satellite network structures, leveraging the strengths of each system. This integration could represent a major breakthrough in satellite-based networking technology by streamlining data processing from remote nodes and cutting down on expenses. The core of this approach lies in the adaptive tailoring of learning techniques for individual entities based on their specific contextual nuances. The experimental findings underscore the prowess of the innovative lightweight strategy, LMAED2L (Enhanced Deep Learning for Earth Data Analysis), across a spectrum of machine learning assignments, showcasing remarkable and consistent performance under diverse operational conditions. Through a strategic fusion of centralized and distributed learning frameworks, the LMAED2L method emerges as a dynamic and effective remedy for the intricate data analysis challenges encountered within satellite networks interfaced with cloud servers. The empirical findings reveal a significant performance boost of our novel approach over traditional methods, with an average increase in reward (4.1 %), task completion rate (3.9 %), and delivered packets (3.4 %). This report suggests that these advancements will catalyze the integration of cutting-edge machine learning algorithms within future networks, elevating responsiveness, efficiency, and resource utilization to new heights.

Semantic Aware Stitching for Panorama.

The most critical aspect of panorama generation is maintaining local semantic consistency. Objects may be projected from different depths in the captured image. When warping the image to a unified canvas, pixels at the semantic boundaries of the different views are significantly misaligned. We propose two lightweight strategies to address this challenge efficiently. First, the original image is segmented as superpixels rather than regular grids to preserve the structure of each cell. We propose effective cost functions to generate the warp matrix for each superpixel. The warp matrix varies progressively for smooth projection, which contributes to a more faithful reconstruction of object structures. Second, to deal with artifacts introduced by stitching, we use a seam line method tailored to superpixels. The algorithm takes into account the feature similarity of neighborhood superpixels, including color difference, structure and entropy. We also consider the semantic information to avoid semantic misalignment. The optimal solution constrained by the cost functions is obtained under a graph model. The resulting stitched images exhibit improved naturalness. Extensive testing on common panorama stitching datasets is performed on the algorithm. Experimental results show that the proposed algorithm effectively mitigates artifacts, preserves the completeness of semantics and produces panoramic images with a subjective quality that is superior to that of alternative methods.

Progress in automobile body processing technology: multi-material and lightweight strategies for saving energy and reducing emissions

Progress in automobile body processing technology: multi-material and lightweight strategies for saving energy and reducing emissions

Research on multi-object detection technology for road scenes based on SDG-YOLOv5

To resolve the challenges of low detection accuracy and inadequate real-time performance in road scene detection, this article introduces the enhanced algorithm SDG-YOLOv5. The algorithm incorporates the SIoU Loss function to accurately predict the angle loss of bounding boxes, ensuring their directionality during regression and improving both regression accuracy and convergence speed. A novel lightweight decoupled heads (DHs) approach is employed to separate the classification and regression tasks, thereby avoiding conflicts between their focus areas. Moreover, the Global Attention Mechanism Group Convolution (GAMGC), a lightweight strategy, is utilized to enhance the network’s capability to process additional contextual information, thereby improving the detection of small targets. Extensive experimental analysis on datasets from Udacity Self Driving Car, BDD100K, and KITTI demonstrates that the proposed algorithm achieves improvements in mAP@.5 of 2.2%, 3.4%, and 1.0% over the original YOLOv5, with a detection speed of 30.3 FPS. These results illustrate that the SDG-YOLOv5 algorithm effectively addresses both detection accuracy and real-time performance in road scene detection.

Defect detection in automotive glass based on modified YOLOv5 with multi-scale feature fusion and dual lightweight strategy

Defect detection in automotive glass based on modified YOLOv5 with multi-scale feature fusion and dual lightweight strategy

Analyzing the impact of geographic variability on emissions reduction and cost efficiency in automotive lightweighting strategies

Lightweight design is crucial in automotive design, especially for battery electric vehicles (BEVs), as it can extend driving range or minimize battery size for a set distance. However, the choice of materials like high-strength steels, while beneficial for vehicle weight, often carries greater environmental costs during production. A comprehensive analysis considering the entire lifecycle and electricity sources is essential to identify truly eco-friendly materials. The market introduction of innovative materials like green steel and recycled products further complicates the selection process due to increased competition. Current assessments of lightweight strategies are complex and challenging to compare, highlighting a research gap in easy-to-understand and actionable evaluations. This study aims to investigate automotive lightweighting across various propulsion systems and regions using prospective Life Cycle Assessment (LCA). It introduces a novel approach by examining how the mix of electricity generation affects the environmental benefits of lightweight materials. A new Key Performance Indicator (KPI) is proposed to help those with limited LCA expertise evaluate the environmental impact of different structural materials. The research also provides visual tools and analyzes cost savings from lower lifecycle greenhouse gas emissions, presenting a rounded view of the economic and environmental advantages of diverse lightweight strategies.

SALI: A Scalable Adaptive Learned Index Framework based on Probability Models

The growth in data storage capacity and the increasing demands for high performance have created several challenges for concurrent indexing structures. One promising solution is the learned index, which uses a learning-based approach to fit the distribution of stored data and predictively locate target keys, significantly improving lookup performance. Despite their advantages, prevailing learned indexes exhibit constraints and encounter issues of scalability on multi-core data storage. This paper introduces SALI, the Scalable Adaptive Learned Index framework, which incorporates two strategies aimed at achieving high scalability, improving efficiency, and enhancing the robustness of the learned index. Firstly, a set of node-evolving strategies is defined to enable the learned index to adapt to various workload skews and enhance its concurrency performance in such scenarios. Secondly, a lightweight strategy is proposed to maintain statistical information within the learned index, with the goal of further improving the scalability of the index. Furthermore, to validate their effectiveness, SALI applied the two strategies mentioned above to the learned index structure that utilizes fine-grained write locks, known as LIPP. The experimental results have demonstrated that SALI significantly enhances the insertion throughput with 64 threads by an average of 2.04x compared to the second-best learned index. Furthermore, SALI accomplishes a lookup throughput similar to that of LIPP+.

Weakly supervised anomaly detection in the Milky Way

ABSTRACT Large-scale astrophysics data sets present an opportunity for new machine learning techniques to identify regions of interest that might otherwise be overlooked by traditional searches. To this end, we demonstrate how Classification Without Labels (CWoLa), a weakly supervised anomaly detection method, can help identify cold stellar streams within the more than one billion Milky Way stars observed by the Gaia satellite. CWoLa operates without the use of labelled streams or knowledge of astrophysical principles. Instead, it uses a classifier to distinguish between mixed samples for which the proportions of signal and background samples are unknown. As a proof of concept, we demonstrate that this computationally lightweight strategy is able to detect both simulated streams and the known stream GD-1 in data. Originally designed for high-energy collider physics, this technique may have broad applicability within astrophysics as well as other domains interested in identifying localized anomalies.

A lightweight speech recognition method with target-swap knowledge distillation for Mandarin air traffic control communications

Miscommunications between air traffic controllers (ATCOs) and pilots in air traffic control (ATC) may lead to catastrophic aviation accidents. Thanks to advances in speech and language processing, automatic speech recognition (ASR) is an appealing approach to prevent misunderstandings. To allow ATCOs and pilots sufficient time to respond instantly and effectively, the ASR systems for ATC must have both superior recognition performance and low transcription latency. However, most existing ASR works for ATC are primarily concerned with recognition performance while paying little attention to recognition speed, which motivates the research in this article. To address this issue, this article introduces knowledge distillation into the ASR for Mandarin ATC communications to enhance the generalization performance of the light model. Specifically, we propose a simple yet effective lightweight strategy, named Target-Swap Knowledge Distillation (TSKD), which swaps the logit output of the teacher and student models for the target class. It can mitigate the potential overconfidence of the teacher model regarding the target class and enable the student model to concentrate on the distillation of knowledge from non-target classes. Extensive experiments are conducted to demonstrate the effectiveness of the proposed TSKD in homogeneous and heterogeneous architectures. The experimental results reveal that the generated lightweight ASR model achieves a balance between recognition accuracy and transcription latency.

Optimizing Road Safety: Advancements in Lightweight YOLOv8 Models and GhostC2f Design for Real-Time Distracted Driving Detection.

The rapid detection of distracted driving behaviors is crucial for enhancing road safety and preventing traffic accidents. Compared with the traditional methods of distracted-driving-behavior detection, the YOLOv8 model has been proven to possess powerful capabilities, enabling it to perceive global information more swiftly. Currently, the successful application of GhostConv in edge computing and embedded systems further validates the advantages of lightweight design in real-time detection using large models. Effectively integrating lightweight strategies into YOLOv8 models and reducing their impact on model performance has become a focal point in the field of real-time distracted driving detection based on deep learning. Inspired by GhostConv, this paper presents an innovative GhostC2f design, aiming to integrate the idea of linear transformation to generate more feature maps without additional computation into YOLOv8 for real-time distracted-driving-detection tasks. The goal is to reduce model parameters and computational load. Additionally, enhancements have been made to the path aggregation network (PAN) to amplify multi-level feature fusion and contextual information propagation. Furthermore, simple attention mechanisms (SimAMs) are introduced to perform self-normalization on each feature map, emphasizing feature maps with valuable information and suppressing redundant information interference in complex backgrounds. Lastly, the nine distinct distracted driving types in the publicly available SFDDD dataset were expanded to 14 categories, and nighttime scenarios were introduced. The results indicate a 5.1% improvement in model accuracy, with model weight size and computational load reduced by 36.7% and 34.6%, respectively. During 30 real vehicle tests, the distracted-driving-detection accuracy reached 91.9% during daylight and 90.3% at night, affirming the exceptional performance of the proposed model in assisting distracted driving detection when driving and contributing to accident-risk reduction.

Environmental drawbacks of lightweight design algorithms in material extrusion additive manufacturing: a case study

Lightweight design is often assumed to be the leading strategy to improve the sustainability of parts produced by additive manufacturing. The present study confutes such an assumption by a cradle-to-gate life cycle assessment of different lightweight strategies applied to a specific case study in the medical field. In particular, a patient-specific finger splint made of polyamide is redesigned by means of generative design, topology optimization and lattice structures. The analysis investigates two markedly different deposition processes, namely Arburg plastic freeforming and fused filament fabrication. The former is carried out on an industrial-grade machine, while a desktop printer is used for the latter. This allows for observing the impact of the redesign in two quite distinct scenarios. Findings demonstrate that, since environmental impacts are mainly driven by building time, the adoption of automated design algorithms can be detrimental to the sustainability of the process. On the other hand, relevant benefits on environmental impacts were achieved by reducing the infill percentage of parts. The results of this work highlight the most relevant aspects which must be considered to limit environmental impacts when designing parts for deposition-based additive manufacturing. This information can be used by designers to drive weight reduction towards sustainability.

Automatic counting of lettuce using an improved YOLOv5s with multiple lightweight strategies

Automatic counting of lettuce using an improved YOLOv5s with multiple lightweight strategies

Public or private? Optimal organization for incentive-based travel demand management

Incentive-Based Travel Demand Management (IBTDM) strategies have received more attention in recent years as they are more appealing to commuters than penalty-based strategies and thus result in less public aversion. In IBTDM, fully time-variant positive incentives are endowed to the traveling public to encourage off-peak trips. We theoretically demonstrate the effectiveness of IBTDM in managing morning commuters with elastic demand based on Vickrey's bottleneck model. Due to the decentralization of IBTDM, incentive providers are no longer limited to public organizations, as private organizations can also administrate IBTDM programs. Therefore, from the perspective of public and private organizations, maximizing social surplus and private profit can be treated as objective functions. In addition, IBTDM is usually not mandatory but voluntary. Thus, we also considered the constraint of penetration rate. We found that when commuters are relatively sensitive to changes in their actual travel costs, a low penetration rate and a small incentive budget are enough for IBTDM to reach an optimal state (for both public and private administrators), and an excessive budget actually creates negative effects. Compared with public organizations, incentive schemes administrated by private organizations are a more lightweight strategy. However, public organizations are overwhelmingly dominant in the case of high market penetration and low traveler sensitivity. Therefore, one of the most effective modes of operation is to have private organizations implement IBTDM in the early stages of promotion, and then all private-owned businesses would be integrated into public organizations when they reach a certain scale.

On the Minimal Adversarial Perturbation for Deep Neural Networks With Provable Estimation Error.

Although Deep Neural Networks (DNNs) have shown incredible performance in perceptive and control tasks, several trustworthy issues are still open. One of the most discussed topics is the existence of adversarial perturbations, which has opened an interesting research line on provable techniques capable of quantifying the robustness of a given input. In this regard, the euclidean distance of the input from the classification boundary denotes a well-proved robustness assessment as the minimal affordable adversarial perturbation. Unfortunately, computing such a distance is highly complex due the non-convex nature of DNNs. Despite several methods have been proposed to address this issue, to the best of our knowledge, no provable results have been presented to estimate and bound the error committed. This paper addresses this issue by proposing two lightweight strategies to find the minimal adversarial perturbation. Differently from the state-of-the-art, the proposed approach allows formulating an error estimation theory of the approximate distance with respect to the theoretical one. Finally, a substantial set of experiments is reported to evaluate the performance of the algorithms and support the theoretical findings. The obtained results show that the proposed strategies approximate the theoretical distance for samples close to the classification boundary, leading to provable robustness guarantees against any adversarial attacks.

Designing and learning a lightweight network for infrared small target detection via dilated pyramid and semantic distillation

Large-scale transformer networks can achieve state-of-the-art infrared small target detection accuracy, but the high computational resource consumption makes their inference speed unsatisfactory. Existing lightweight networks can certainly achieve real-time detection of infrared small targets, but no matter how carefully designed the lightweight strategy is, there is still a gap in the accuracy of these networks compared to large-scale networks. To address these problems, in this paper, we propose a lightweight infrared small target detection network capable of high-speed inference and a knowledge distillation method for learning higher-order semantic information from the transformer network. Specifically, based on depthwise separable dilated convolution, we design each stage in the detection network, called multi-scale dilated pyramid network (MDPNet), as a multi-branch parallel pyramid. This design can enlarge the receptive field and enhance the ability to extract contextual features of the network. Furthermore, we utilize knowledge distillation to bridge the detection performance gap with the transformer. Based on the self-attention mechanism, a semantic distillation sub-network is constructed between the teacher transformer and student convolution network, which enables a more efficient cross-model transfer of knowledge about higher-order semantic feature extraction between networks with different mechanisms and increases the detection accuracy without computation load. We demonstrate the rationality and effectiveness of the overall network design and learning approach through exhaustive experiments. On the widely accepted public datasets SIRST and NUDT-SIRST, nIoU reaches 74.88 and 75.10, and PD reaches 99.08 and 97.88, outperforming other networks. With the number of parameters of only 0.23 M, the network achieves an inference speed of 137 FPS for 320 × 320 infrared images.

Intelligent electromagnetic mapping via physics driven and neural networks on frequency selective surfaces

The high mapping efficiency between various structures and electromagnetic (EM) properties of frequency selective surfaces (FSSs) is the state-of-the-art in the EM community. The most straightforward approaches for beam analysis depend on measurements and conventional EM calculation methods, which are inefficient and time-consuming. Equivalent circuit models (ECMs) with excellent intuitiveness and simplicity have been put forward extensively. Despite several applications, bottlenecks in ECM still exist, i.e. the application scope is restricted to narrow bands and specific structures, which is triggered by the ignorance of EM nonlinear coupling. In this study, for the first time, a lightweight physical model based on neural network (ECM-NN) is proposed , which exhibits great physical interpretability and spatial generalization abilities. The nonlinear mapping relationship between structure and beam behavior is interpreted by corresponding simulations. Specifically, two deep parametric factors obtained by multi-layer perceptron networks are introduced to serve as the core of lightweight strategies and compensate for the absence of nonlinearity. Experimental results of single square loop (SL) and double SL indicate that compared with related works, better agreements of the frequency responses and resonant frequencies are achieved with ECM-NN in broadband (0–30 GHz) as well as oblique incident angles (0°–60°). The average accuracy of the mapping is higher than 98.6%. The findings of this study provide a novel strategy for further studies of complex FSSs.

A Review on Processing–Microstructure–Property Relationships of Al-Si Alloys: Recent Advances in Deformation Behavior

While research on lightweight materials has been carried out for decades, it has become intensified with recent climate action initiatives leading pathways to net zero. Aluminum alloys are at the pinnacle of the light metal world, especially in the automotive and aerospace industries. This review intends to highlight recent developments in the processing, structure, and mechanical properties of structural Al-Si alloys to solve various pressing environmental issues via lightweighting strategies. With the excellent castability of Al-Si alloys, advancements in emerging casting methods and additive manufacturing processes have been summarized in relation to varying chemical compositions. Improvements in thermal stability and electrical conductivity, along with superior mechanical strength and fatigue resistance, are analyzed for advanced Al-Si alloys with the addition of other alloying elements. The role of Si morphology modification, along with particle distribution, size, and precipitation sequencing, is discussed in connection with the improvement of static and dynamic mechanical properties of the alloys. The physics-based damage mechanisms of fatigue failure under high-cycle and low-cycle fatigue loading are further elaborated for Al-Si alloys. The defect, porosity, and surface topography related to manufacturing processes and chemical compositions are also reviewed. Based on the gaps identified here, future research directions are suggested, including the usage of computational modeling of microstructures and the integration of artificial intelligence to produce mass-efficient and cost-effective solutions for the manufacturing of Al-Si alloys.