Pruning Algorithm Research Articles

A Parameters optimization framework for pose estimation algorithm based on point cloud

Pose estimation using point cloud data is pivotal in robotics. Despite the myriad of algorithms developed for this task, their efficiency is often contingent on optimal parameter settings. Parameters adjustment is a process that traditionally relies on extensive experience and deep understanding of the underlying algorithm. This manual, experience-driven process can hamper the adaptability and performance of pose estimation algorithms. Addressing this limitation, this paper proposes a novel framework for parameter optimization tailored to pose estimation algorithms. The framework introduces an objective function based on pose error, designed to enhance both the algorithm’s number of results and the accuracy. The methodology requires the object model and several scene point clouds with real poses of the objects.By synergistically integrating this objective function with various sampling and pruning algorithms to optimize the pose estimate algorithm, obtaining a set of superior parameters to replace default values becomes a straightforward process. Experimental assessments were conducted on the pose estimation algorithm implemented in Halcon, considering two types of objects within the Industrial 3D Object Detection Dataset (ITODD). The experimental results shown that the number and accuracy of matching results for optimized parameters are better than those proposed by algorithm developers. This not only underscores the framework’s potential as an alternative to conventional manual parameter tuning but also its utility as a foundational configuration for further refinements. Ultimately, the methodology augments the foundational efficiency and versatility of pose estimation algorithms, paving the way for more adept robotic interactions with a myriad of objects.

Whole transcriptome sequencing identifies a competitive endogenous RNA network that regulates the immunity of bladder cancer

Several types of non-coding RNAs such as circRNAs, lncRNAs, and miRNAs have been identified to regulate mRNAs through the mechanism known as the competitive endogenous RNA (ceRNA) network. To explore the role of the ceRNA regulatory network in the immune microenvironment of bladder cancer, whole-transcriptome sequencing of bladder tumor and its peritumoral tissues from 38 bladder cancer patients, with a total of 63 samples, was performed to screen differentially expressed circ-, lnc-, mi-, and mRNAs to construct a circ/lnc-mi-mRNA regulatory network with pruning algorithms. We excavated a key immune-related gene BDNF to build the final ceRNA network as hsa-miR-107 sponged by hsa-circ-000211, AC108488.1, and LINC00163. Finally, a meta-analysis of 7 public datasets demonstrated that low expression of BDNF and high expression of hsa-miR-107 were associated with longer survival. Our study identified a ceRNA regulatory network as a potentially new prognostic marker and molecular therapeutic target of bladder cancer.

Noninvasive Assessment of hiPSC Differentiation toward Cardiomyocytes Using Pretrained Convolutional Neural Networks and the Channel Pruning Algorithm.

Human induced pluripotent stem cell (hiPSC)-derived cardiomyocytes (hiPSC-CMs) offer versatile applications in tissue engineering and drug screening. To facilitate the monitoring of hiPSC cardiac differentiation, a noninvasive approach using convolutional neural networks (CNNs) was explored. HiPSCs were differentiated into cardiomyocytes and analyzed using the quantitative real-time polymerase chain reaction (qRT-PCR). The bright-field images of the cells at different time points were captured to create the dataset. Six pretrained models (AlexNet, GoogleNet, ResNet 18, ResNet 50, DenseNet 121, VGG 19-BN) were employed to identify different stages in differentiation. VGG 19-BN outperformed the other five CNNs and exhibited remarkable performance with 99.2% accuracy, recall, precision, and F1 score and 99.8% specificity. The pruning process was then applied to the optimal model, resulting in a significant reduction of model parameters while maintaining high accuracy. Finally, an automation application using the pruned VGG 19-BN model was developed, facilitating users in assessing the cell status during the myocardial differentiation of hiPSCs.

UniADS: Universal Architecture-Distiller Search for Distillation Gap

In this paper, we present UniADS, the first Universal Architecture-Distiller Search framework for co-optimizing student architecture and distillation policies. Teacher-student distillation gap limits the distillation gains. Previous approaches seek to discover the ideal student architecture while ignoring distillation settings. In UniADS, we construct a comprehensive search space encompassing an architectural search for student models, knowledge transformations in distillation strategies, distance functions, loss weights, and other vital settings. To efficiently explore the search space, we utilize the NSGA-II genetic algorithm for better crossover and mutation configurations and employ the Successive Halving algorithm for search space pruning, resulting in improved search efficiency and promising results. Extensive experiments are performed on different teacher-student pairs using CIFAR-100 and ImageNet datasets. The experimental results consistently demonstrate the superiority of our method over existing approaches. Furthermore, we provide a detailed analysis of the search results, examining the impact of each variable and extracting valuable insights and practical guidance for distillation design and implementation.

Neural network relief: a pruning algorithm based on neural activity

Neural network relief: a pruning algorithm based on neural activity

Lightweight Meter Pointer Recognition Method Based on Improved YOLOv5.

In substation lightning rod meter reading data taking, the classical object detection model is not suitable for deployment in substation monitoring hardware devices due to its large size, large number of parameters, and slow detection speed, while is difficult to balance detection accuracy and real-time requirements with the existing lightweight object detection model. To address this problem, this paper constructs a lightweight object detection algorithm, YOLOv5-Meter Reading Lighting (YOLOv5-MRL), based on the improved YOLOv5 model's speed while maintaining accuracy. Then, the YOLOv5s are pruned based on the convolutional kernel channel soft pruning algorithm, which greatly reduces the number of parameters in the YOLOv5-MRL model while maintaining a certain accuracy loss. Finally, in order to facilitate the dial reading, the dial external circle fitting method is proposed to calculate the dial reading using the circular angle algorithm. The experimental results on the self-built dataset show that the YOLOv5-MRL object detection model achieves a mean average precision of 96.9%, a detection speed of 5 ms/frame, and a model weight size of 5.5 MB, making it better than other advanced dial reading models.

Optimization of Nonlinear Convolutional Neural Networks based on Improved Chameleon Group Algorithm

In order to solve the most difficult problem of the architectural model established by CNN in solving specific problems, which results in parameter overflow and inefficient training, an optimization algorithm for nonlinear convolutional neural networks based on improved chameleon swarm algorithm is proposed. This article mainly introduces the use of Chameleon Swarm Optimization (PSO) algorithm to research the parameters of CNN architecture, solve them, and achieve the optimization of the optimization model.Although the number of parameters that need to be set up in CNN is very large, this method can find better testing space for Alexnet samples with 5 different images. In order to improve the performance of the improved pruning algorithms, two candidate pruning algorithms are also proposed. The experimental results show that compared with the traditional Alexnet model, the improved pruning method improves the image recognition ability of the Caffe primary parameter set from 1.3% to 5.7%. This method has wide applicability and can be applied to most neural networks which do not require any special functional modules of the Alexnet network model.

Hybrid Integration of Bagging and Decision Tree Algorithms for Landslide Susceptibility Mapping

Landslides represent a significant global natural hazard, threatening human settlements and the natural environment. The primary objective of the study was to develop a landslide susceptibility modeling approach that enhances prediction accuracy and informs land-use planning decisions. The study utilized a hybrid ensemble-based methodology to improve prediction accuracy and effectively capture the complexity of landslide susceptibility patterns. This approach harnessed the power of ensemble models, employing a bagging algorithm with base learners, including the reduced error pruning decision tree (REPTree) and functional tree (FT) models. Ensemble models are particularly valuable because they combine the strengths of multiple models, enhancing the overall performance and robustness of the landslide susceptibility prediction. The study focused on Yanchuan County, situated within the hilly and gully region of China’s Loess Plateau, known for its susceptibility to landslides, using sixteen critical landslide conditioning factors, encompassing topographic, environmental, and geospatial variables, namely elevation, slope, aspect, proximity to rivers and roads, rainfall, the normalized difference vegetation index, soil composition, land use, and more. Model performances were evaluated and verified using a range of metrics, including receiver operating characteristic (ROC) curves, trade-off statistical metrics, and chi-square analysis. The results demonstrated the superiority of the integrated models, particularly the bagging FT (BFT) model, in accurately predicting landslide susceptibility, as evidenced by its high area under the curve area (AUC) value (0.895), compared to the other models. The model excelled in both positive predictive rate (0.847) and negative predictive rate (0.886), indicating its efficacy in identifying landslide and non-landslide areas and also in the F-score metric with a value of 0.869. The study contributes to the field of landslide risk assessment, offering a significant investigation tool for managing and mitigating landslide hazards in Yanchuan County and similar regions worldwide.

Multi-objective compression for CNNs via evolutionary algorithm

Filter pruning has been successfully applied in convolutional neural networks (CNN) to reduce model redundancy and speed up inference. Current pruning algorithms require setting a target value for a pruned model, which is not conducive to the rapid implementation in practical application. Firstly, for a target model, it may be necessary to pay attention to multiple performances at the same time, but it is hard to keep the best trade-off among indicators. Secondly, it is time-consuming to find a suitable model by pruning in a trial-and-error way. Therefore, this paper attempts to combine the most concerning indicators into a search target, rather than producing a model that only contains some pre-defined constraint. Specifically, the multi-object differential evolution algorithm is introduced into the search process, which is used to obtain a set of sparse structures that have taken all useful indicators into account. Then select proper sub-network to retrain from search results according to actual requirements. Experiments on multiple advanced CNN networks show that our algorithm can compress different indicators while keeping comparable or better performance than prior works. On ImageNet, pruning based on ResNet50 reduces FLOPs by 60.4%, while the accuracy of top-1 and top-5 only loses 1.09% and 0.57%, respectively.

Decision tree analysis for age estimation in living individuals: integrating cervical and dental radiographic evaluations within a South African population.

Age estimation in living individuals around the age of 18years is medico-legally important in undocumented migrant cases and in countries like South Africa where many individuals are devoid of identification documents. Establishing whether an individual is younger than 18years largely influences the legal procedure that should be followed in dealing with an undocumented individual. The aim of this study was to combine dental third molar and anterior inferior apophysis ossification data for purposes of age estimation, by applying a decision tree analysis. A sample comprising of 871 black South African individuals (n = 446 males, 425 = females) with ages ranging between 15 and 24years was analyzed using panoramic and cephalometric radiographs. Variables related to the left upper and lower third molars and cervical vertebral ring apophysis ossification of C2, C3, and C4 vertebrae analyzed in previous studies were combined in a multifactorial approach. The data were analyzed using a pruned decision tree function for classification. Male and female groups were handled separately as a statistically significant difference was found between the sexes in the original studies. A test sample of 30 individuals was used to determine if this approach could be used with confidence in estimating age of living individuals. The outcomes obtained from the test sample indicated a close correlation between the actual ages (in years and months) and the predicted ages (in years only), demonstrating an average age difference of 0.47years between the corresponding values. This method showed that the application of decision tree analysis using the combination of third molar and cervical vertebral development is usable and potentially valuable in this application.

A study of the current status of the quality of teacher-child interactions in preschools based on mathematical statistics multiple regression modeling

Abstract In group teaching activities, teacher-child interaction occupies a dominant position, and its quality has a significant impact on children’s interpersonal communication, language expression, and problem-handling. This study first designed a relevant questionnaire to assess the current quality of teacher-child interactions in preschools and then developed a stepwise regression algorithm based on decision trees to investigate the factors that influence the quality of these interactions. Firstly, we remove a certain number of irrelevant variables from all the input variables to reduce interference. Next, we establish stepwise regression equations between the input variables and output variables selected by decision tree pruning. Finally, we eliminate the variables with covariance. The study found that the average score of the quality of teacher-child interactions in preschools is 4.2381, which is relatively low, and the highest score in the dimension of “classroom organization” is only 4.8908. The highest score in the dimension of “classroom organization” was only 4.8908. The average scores of “emotional support”, “classroom management,” and “educational support” in the dimensions of small class and life activities were 4.76746, 4.10562, 2.8455, and 5.562, respectively, 2.8455) and (5.01564, 4.99567, 3.75402) respectively. Teacher-child interaction quality = 4.012 + 0.145 × emotional expression, i.e., emotional expression has a significant positive effect on the quality of teacher-student interaction, and the independent variable, emotional expression, can significantly and positively predict the dependent variable teacher-child interaction quality. This paper provides a useful exploration for improving the professional abilities level of kindergarten teachers and the quality of their interactions with children.

An Investigation of Frequency-Domain Pruning Algorithms for Accelerating Human Activity Recognition Tasks Based on Sensor Data

An Investigation of Frequency-Domain Pruning Algorithms for Accelerating Human Activity Recognition Tasks Based on Sensor Data

SARGap: A Full-Link General Decoupling Automatic Pruning Algorithm for Deep Learning-Based SAR Target Detectors

SARGap: A Full-Link General Decoupling Automatic Pruning Algorithm for Deep Learning-Based SAR Target Detectors

Real-time litchi detection in complex orchard environments: A portable, low-energy edge computing approach for enhanced automated harvesting

Litchi, a succulent and perishable fruit, presents a narrow annual harvest window of under two weeks. The advent of smart agriculture has driven the adoption of visually-guided, automated litchi harvesting techniques. However, conventional approaches typically rely on laboratory-based, high-performance computing equipment, which presents challenges in terms of size, energy consumption, and practical application within litchi orchards. To address these limitations, we propose a real-time litchi detection methodology for complex environments, utilizing portable, low-energy edge computing devices. Initially, the litchi orchard imagery is collected to enhance data generalization. Subsequently, a convolutional neural network (CNN)-based single-stage detector, YOLOx, is constructed to accurately pinpoint litchi fruit locations within the images. To facilitate deployment on portable, low-energy edge devices, we employed channel pruning and layer pruning algorithms to compress the trained model, reducing its size and parameters. Additionally, the knowledge distillation technique is harnessed to fine-tune the network. Experimental findings demonstrated that our proposed method achieved a 97.1% compression rate, yielding a compact litchi detection model of a mere 6.9 MB, while maintaining 94.9% average precision and 97.2% average recall. Processing 99 frames per second (FPS), the method exhibited a 1.8-fold increase in speed compared to the unprocessed model. Consequently, our approach can be readily integrated into portable, low-computational automatic harvesting equipment, ensuring real-time, precise litchi detection within orchard settings.

Scalable and Efficient Approach for High Temporal Fuzzy Utility Pattern Mining.

Fuzzy utility (FU) pattern mining with an advantage in human reasoning has become one of the interesting topics in studies of knowledge discovery. The discovered information in FU pattern mining from real-life quantitative databases with item profits is suitable for interpreting data from a human perspective because it is not expressed using numerical values but linguistic terms which consist of natural languages. State-of-the-art approaches in this literature provide extended results by considering temporal factors, such as seasons, which can be influential in real-life situations. However, they still suffer from scalability issues because they are based on level-wise approaches which generate a number of candidates. In this article, we propose a scalable and efficient approach with a novel data structure for mining high temporal FU patterns without generating candidates. Efficient pruning techniques and algorithms are presented to improve the performance of the proposed approach. Performance experiments on both real and synthetic datasets show that the suggested algorithm has better performance than the state-of-the-art algorithms in terms of runtime, memory usage, and scalability.

Complex business ecosystem intelligence using AI-powered visual analytics

Business ecosystems are complex, dynamic systems characterized by a multitude of entities, including companies, ventures, and technologies, as well as activities and trends. Understanding the state of business ecosystems is an increasingly critical strategic imperative for many decision makers, but it is a resource-intensive activity as relevant information sources are dispersed, often highly unstructured, and not integrated or curated to deliver actionable insights. In this research, we present the design and implementation of an interactive visual analytic system that integrates artificial intelligence and graph visualization techniques to augment decision makers’ understanding of the complex public narrative associated with business ecosystems entities. Our system is driven by a real-time content engine of 100,000+ global data sources including press releases, news articles, industry reports, analyst blogs in multiple languages organized across several domain-specific repositories. Following a user-specified query, the engine extracts both domain-agnostic and domain-specific entities and concepts for each document in the result set. We then model and visualize the resulting data as a dynamic, multipartite network and implement graph pruning algorithms and interactive data controls to enable users to interactively explore and discover the underlying business ecosystem from multiple perspectives. We illustrate and discuss the value of our system using representative use cases. Our study makes multiple contributions to visual decision support theory and practice, including mining unstructured data, constructing and interacting with knowledge graphs, and designing visual analytic tools for ecosystem intelligence. We conclude the study with implications and future research opportunities.

A spanning tree algorithm with adaptive pruning with low redundancy coverage rate

In reality, when UAV is used in region reconnaissance, the shape of region is usually irregular. Selecting an algorithm to find a complete coverage path with smaller coverage redundancy rate, lower path motion complexity and theoretical energy consumption has certain research significance for the complete coverage path planning (CCPP). In existing algorithms, there is always a high redundant coverage rate under irregularly shaped regions. The motion complexity and the theoretical energy consumption of the path also increase. In this paper, a kind of spanning tree with adaptive pruning (STAP) algorithm based on the grid map is proposed to solve the above problems: on the one hand, it uses a greedy growth strategy to determine the branching growth direction, and proposes an adaptive pruning method for self-optimization of the initial path; on the other hand, a kind of nearest neighbor sub-region connection rule of adjacent paths is proposed to reduce the complexity and the coverage redundancy rate. The STAP algorithm searches for all grid centers, grows, and optimizes the spanning tree by creating paths between the centers. At the end of the optimization, it creates a complete coverage path based on the distribution of the spanning tree. Experiments show that when the STAP algorithm is applied to unobstructed regions, it plans the path with the lowest redundant coverage rate, which reaches 2.6296% and 4.9402%; when it is applied to the regions include obstacles, the advantages of the algorithm are more significant: On the one hand, the path planned by STAP has the least total length, the lowest number of turning and turning back. On the other hand, the redundant coverage rate of the path is still the lowest, reaching 4.1362% and 2.7765%, respectively. The STAP is primarily suitable for situations about CCPP under grid maps formed by irregularly shaped regions, and it holds certain significance for ground reconnaissance of UAVs.

Skeleton Ground Truth Extraction: Methodology, Annotation Tool and Benchmarks

AbstractSkeleton Ground Truth (GT) is critical to the success of supervised skeleton extraction methods, especially with the popularity of deep learning techniques. Furthermore, we see skeleton GTs used not only for training skeleton detectors with Convolutional Neural Networks (CNN), but also for evaluating skeleton-related pruning and matching algorithms. However, most existing shape and image datasets suffer from the lack of skeleton GT and inconsistency of GT standards. As a result, it is difficult to evaluate and reproduce CNN-based skeleton detectors and algorithms on a fair basis. In this paper, we present a heuristic strategy for object skeleton GT extraction in binary shapes and natural images. Our strategy is built on an extended theory of diagnosticity hypothesis, which enables encoding human-in-the-loop GT extraction based on clues from the target’s context, simplicity, and completeness. Using this strategy, we developed a tool, SkeView, to generate skeleton GT of 17 existing shape and image datasets. The GTs are then structurally evaluated with representative methods to build viable baselines for fair comparisons. Experiments demonstrate that GTs generated by our strategy yield promising quality with respect to standard consistency, and also provide a balance between simplicity and completeness.

Personal Information Protection in Government Data Openness Using Decision Tree Model

To address the issues of personal privacy and information security in the current digital age, this study first collects and analyzes relevant government data openness policies and personal information protection laws and regulations to understand the current policy and legal environment. Secondly, a complexity pruning decision tree model is constructed, which can identify and evaluate potential personal information protection risks in government data openness. Using the Singapore government open dataset, this decision tree model is applied for empirical analysis, and its accuracy and effectiveness are evaluated. The research results demonstrate that the complexity pruning decision tree model performs well in terms of accuracy, recall rate, F1 score, and the area under the ROC Curve (AUC). The model achieves an accuracy of 0.85 on the training and 0.8 on the test sets, indicating its high performance in personal information protection in government data openness.

Assessing tuberculosis in the skeleton with the use of decision tree analysis.

Diagnosis of specific infectious diseases in the skeleton is often difficult and relies on expert opinion. Statistics is not often used as a tool to assist in such diagnoses, and therefore this study aimed at employing data mining and machine learning in the form of decision tree analysis to aid in recognizing tuberculosis (TB) in skeletal remains and find patterns of skeletal involvement. The sample included 387 modern South African individuals (n = 207 individuals known to have died of TB and n = 180 as a control group) which were scored for the presence or absence of 21 skeletal lesions documented to be associated with TB. A pruned decision tree classification analysis was done to detect significant patterns and associations between variables which produced a model with a moderate classification rate based on four of the variables. As expected, vertebral changes were selected first, followed by rib, acetabular and lastly cranial changes. As a proof of concept, it was shown that machine learning was able to identify patterns of changes in TB skeletons versus a control group. However, further investigation into the use of machine learning in assessing skeletal changes associated with specific diseases is needed.