Adaptive Partitioning Algorithm Research Articles

DeepZoning: Re-accelerate CNN Inference with Zoning Graph for Heterogeneous Edge Cluster

Parallelizing CNN inference on heterogeneous edge clusters with data parallelism has gained popularity as a way to meet real-time requirements without sacrificing model accuracy. However, existing algorithms struggle to find optimal parallel granularity for complex CNNS, the structure of which is a directed acyclic graph (DAG) rather than a chain, and the parallel dimension is inflexible. To distribute the workload of modern CNNs on heterogeneous devices is also proven as NP-hard problem. In this paper, we introduce DeepZoning , a versatile and cooperative inference framework that combines both model and data parallelism to accelerate CNN inference. DeepZoning employs two algorithms at different levels: (1) a low-level Adaptive Workload Partition algorithm that uses linear programming and takes spatial and channel dimensions into optimization during the search for feature map distribution on heterogeneous devices, and (2) a high-level Model Partition algorithm that finds the optimal model granularity and organizes complex CNNs into sequential zones to balance communication and computation during execution. Our experimental evaluations show that DeepZoning is effective, achieving up to a 3.02 × speed improvement on our experimental prototype compared to state-of-the-art algorithms.

Fast Depth Map Coding Algorithm for 3D-HEVC Based on Gradient Boosting Machine

Three-Dimensional High-Efficiency Video Coding (3D-HEVC) has been extensively researched due to its efficient compression and deep image representation, but encoding complexity continues to pose a difficulty. This is mainly attributed to redundancy in the coding unit (CU) recursive partitioning process and rate–distortion (RD) cost calculation, resulting in a complex encoding process. Therefore, enhancing encoding efficiency and reducing redundant computations are key objectives for optimizing 3D-HEVC. This paper introduces a fast-encoding method for 3D-HEVC, comprising an adaptive CU partitioning algorithm and a rapid rate–distortion-optimization (RDO) algorithm. Based on the ALV features extracted from each coding unit, a Gradient Boosting Machine (GBM) model is constructed to obtain the corresponding CU thresholds. These thresholds are compared with the ALV to further decide whether to continue dividing the coding unit. The RDO algorithm is used to optimize the RD cost calculation process, selecting the optimal prediction mode as much as possible. The simulation results show that this method saves 52.49% of complexity while ensuring good video quality.

PArtNNer: Platform-Agnostic Adaptive Edge-Cloud DNN Partitioning for Minimizing End-to-End Latency

The last decade has seen the emergence of Deep Neural Networks (DNNs) as the de facto algorithm for various computer vision applications. In intelligent edge devices, sensor data streams acquired by the device are processed by a DNN application running on either the edge device itself or in the cloud. However, “edge-only” and “cloud-only” execution of State-of-the-Art DNNs may not meet an application’s latency requirements due to the limited compute, memory, and energy resources in edge devices, dynamically varying bandwidth of edge-cloud connectivity networks, and temporal variations in the computational load of cloud servers. This work investigates distributed (partitioned) inference across edge devices (mobile/end device) and cloud servers to minimize end-to-end DNN inference latency. We study the impact of temporally varying operating conditions and the underlying compute and communication architecture on the decision of whether to run the inference solely on the edge, entirely in the cloud, or by partitioning the DNN model execution among the two. Leveraging the insights gained from this study and the wide variation in the capabilities of various edge platforms that run DNN inference, we propose PArtNNer , a platform-agnostic adaptive DNN partitioning algorithm that finds the optimal partitioning point in DNNs to minimize inference latency. PArtNNer can adapt to dynamic variations in communication bandwidth and cloud server load without requiring pre-characterization of underlying platforms. Experimental results for six image classification and object detection DNNs on a set of five commercial off-the-shelf compute platforms and three communication standards indicate that PArtNNer results in 10.2× and 3.2× (on average) and up to 21.1× and 6.7× improvements in end-to-end inference latency compared to execution of the DNN entirely on the edge device or entirely on a cloud server, respectively. Compared to pre-characterization-based partitioning approaches, PArtNNer converges to the optimal partitioning point 17.6× faster.

Chemical bonding in lead anionic clusters

Lead anion clusters, as examples of Zintl anions, have captivated researchers due to their intriguing structures and unexpected stoichiometries. Despite extensive experimental and theoretical investigations, the understanding of the chemical bonding in these clusters remains incomplete and subject to interpretation. In this study, we focused on several widely studied homoatomic and heteroatomic lead anionic clusters and employed the Adaptive Natural Density Partitioning (AdNDP) algorithm to decipher their bonding patterns. Our aim was to provide chemically intuitive descriptions of the bonding in these clusters. Our findings reveal the presence of multiple delocalized bonds in most of the studied lead clusters. Remarkably, these bonding patterns exhibit characteristics of aromatic compounds. This work represents a step towards a more comprehensive understanding of the bonding in lead anion clusters. By shedding light on their chemical bonding patterns, we hope to inspire further research in the design and synthesis of novel lead homoatomic and heteroatomic Zintl clusters, fostering advancements in their properties and potential applications.

Kendall transfer entropy: a novel measure for estimating information transfer in complex systems

Objective. Transfer entropy (TE) has been widely used to infer causal relationships among dynamical systems, especially in neuroscience. Kendall transformation provides a novel quantization method for estimating information-theoretic measures and shows potential advantages for small-sample neural signals. But it has yet to be introduced into the framework of TE estimation, which commonly suffers from the limitation of small sample sizes. This paper aims to introduce the idea of Kendall correlation into TE estimation and verify its effect. Approach. We proposed the Kendall TE (KTE) which combines the improved Kendall transformation and the TE estimation. To confirm its effectiveness, we compared KTE with two common TE estimation techniques: the adaptive partitioning algorithm (D-V partitioning) and the symbolic TE. Their performances were estimated by simulation experiments which included linear, nonlinear, linear + nonlinear models and neural mass models. Moreover, the KTE was also applied to real electroencephalography (EEG) recordings to quantify the directional connectivity between frontal and parietal regions with propofol-induced general anesthesia. Main results. The simulation results showed that the KTE outperformed the other two methods by many measures: (1) identifying the coupling direction under a small sample size; (2) the sensitivity to coupling strength; (3) noise resistance; and (4) the sensitivity to time-dependent coupling changes. For real EEG recordings, the KTE clearly detected the disrupted frontal-to-parietal connectivity in propofol-induced unconsciousness, which is in agreement with previous findings. Significance. We reveal that the proposed KTE method is a robust and powerful tool for estimating TE, and is particularly suitable for small sample sizes. The KTE also provides an innovative form of quantizing continuous time series for information-theoretic measures.

Probability Density Estimation through Nonparametric Adaptive Partitioning and Stitching

We present a novel nonparametric adaptive partitioning and stitching (NAPS) algorithm to estimate a probability density function (PDF) of a single variable. Sampled data is partitioned into blocks using a branching tree algorithm that minimizes deviations from a uniform density within blocks of various sample sizes arranged in a staggered format. The block sizes are constructed to balance the load in parallel computing as the PDF for each block is independently estimated using the nonparametric maximum entropy method (NMEM) previously developed for automated high throughput analysis. Once all block PDFs are calculated, they are stitched together to provide a smooth estimate throughout the sample range. Each stitch is an averaging process over weight factors based on the estimated cumulative distribution function (CDF) and a complementary CDF that characterize how data from flanking blocks overlap. Benchmarks on synthetic data show that our PDF estimates are fast and accurate for sample sizes ranging from 29 to 227, across a diverse set of distributions that account for single and multi-modal distributions with heavy tails or singularities. We also generate estimates by replacing NMEM with kernel density estimation (KDE) within blocks. Our results indicate that NAPS(NMEM) is the best-performing method overall, while NAPS(KDE) improves estimates near boundaries compared to standard KDE.

PROADAPT: Proactive framework for adaptive partitioning for big data warehouses

Parallel DBMSs have become more and more mature and getting several success stories in the industry. This situation has been reached by powerful data partitioning and data allocation techniques and algorithms. By analyzing these findings closely, we figure out that they are quickly stressed by the workload changes, which represents the usual case of business analytics applications. To deal with this challenge in the context of big data warehouses, several studies proposed to move to another processing paradigm outside the DBMS realm such as Spark by proposing adaptive partitioning solutions to tackle the workload changes. The majority of approaches are offline and those that are online cause significant random disk I/O costs. This is because the correlation that may exist between jobs and data blocks read from the disk is not captured to refine the adaptive partitioning algorithms. This represents one of the major causes of providing high performance of dynamic workloads. To solve such limitations, we propose in this paper a proactive framework for query-aware adaptive partitioning (called PROADAPT) that uses an AI-inspired methodology that can be connected to any query optimizer managing partitioned data. This methodology uses a genetic algorithm to solve our formalized problem that considers the interaction that may exist among workload queries. PROADAPT intensively rewrites queries by exploiting dimension hierarchies to skip irrelevant data and then improves I/O performance. Different technical modules of our framework are discussed. Finally, we conduct intensive experiments on Postgres-XL and a Spark SQL parallel cluster to show the effectiveness and efficiency of our approach.

Adaptive non-uniform partition algorithm based on linear canonical transform

This paper proposes a new signal representation method of adaptive non-uniform partition algorithm based on linear canonical transform, which can partition the signal in independent regions to further improve the reconstruction quality of the signal thus extending its application range. That means the partial signal in each sub-region can be approximated by a determined linear canonical transform series after applying the least squares approximation. The partition process is guided by a set control error so that it can more effectively reflect the regional characteristics of the signal. Compared with the signal reconstructed effect by both the original adaptive non-uniform partition algorithm and Fourier transform through simulation experiments, the results prove that the proposed algorithm has better performance in signal representation and reconstruction under the same approximation condition. On the other hand, through the comparison of simulation experiments with wavelet transform and the original rectangular non-uniform partition algorithm for one-dimensional signal denoising, it can be concluded that the proposed method has certain potential in signal denoising.

Traffic Accident Detection Based on Deformable Frustum Proposal and Adaptive Space Segmentation

Road accident detection plays an important role in abnormal scene reconstruction for Intelligent Transportation Systems and abnormal events warning for autonomous driving. This paper presents a novel 3D object detector and adaptive space partitioning algorithm to infer traffic accidents quantitatively. Using 2D region proposals in an RGB image, this method generates deformable frustums based on point cloud for each 2D region proposal and then frustum-wisely extracts features based on the farthest point sampling network (FPS-Net) and feature extraction network (FE-Net). Subsequently, the encoder-decoder network (ED-Net) implements 3D-oriented bounding box (OBB) regression. Meanwhile, the adaptive least square regression (ALSR) method is proposed to split 3D OBB. Finally, the reduced OBB intersection test is carried out to detect traffic accidents via separating surface theorem (SST). In the experiments of KITTI benchmark, our proposed 3D object detector outperforms other state-of-the-art methods. Meanwhile, collision detection algorithm achieves the satisfactory performance of 91.8% accuracy on our SHTA dataset.

Information Classification Algorithm based on Project-based Learning Data-driven and Stochastic Grid

The adaptive partitioning algorithm of information set in simulation laboratory based on project-based learning data-driven and random grid is studied to effectively preprocess the information set and improve the adaptive partitioning effect of the information set. Using the improved fuzzy C-means clustering algorithm driven by project-based learning data, the fuzzy partition of information set in simulation laboratory is carried out to complete preprocessing of information set; The pre-processing information set space is roughly divided by the grid partitioning algorithm based on the data histogram; A random mesh generation algorithm based on uniformity is used to finely divide the coarse mesh cells; Taking the representative points of grid cells as the clustering center, the pre-processing information set is clustered by the density peak clustering algorithm to complete the adaptive partitioning of the information set in simulation laboratory. Experimental results show that this algorithm can effectively preprocess and adaptively partition the information set of simulation laboratory; For different dimension information sets, the evaluation index values of Rand index, Purity, standard mutual information, interval and Dunn index of the algorithm are all high, and the evaluation index values of compactness and Davidson's banding index are all low, so the algorithm has a high accuracy of adaptive partitioning of information sets.

Occurrence of Double Bond in π-Aromatic Rings: An Easy Way to Design Doubly Aromatic Carbon-Metal Structures.

A chemical bonding of several metallabenzenes and metallabenzynes was studied via an adaptive natural density partitioning (AdNDP) algorithm and the induced magnetic field analysis. A unique chemical bonding pattern was discovered where the M=C (M: Os, Re) double bond coexists with the delocalized 6c-2e π-bonding elements responsible for aromatic properties of the investigated complexes. In opposition to the previous description where 8 delocalized π-electrons were reported in metallabenzenes and metallabenzynes, we showed that only six delocalized π-electrons are present in those molecules. Thus, there is no deviation from Hückel’s aromaticity rule for metallabenzynes/metallabenzenes complexes. Based on the discovered bonding pattern, we propose two thermodynamically stable novel molecules that possess not only π-delocalization but also retain six σ-delocalized electrons, rendering them as doubly aromatic species. As a result, our investigation gives a new direction for the search for carbon-metal doubly aromatic molecules.

An efficient adaptive space partitioning algorithm for electromagnetic scattering calculation of complex 3D models

The space partitioning algorithm based on the rounding and addressing operations has been proved to be an efficient space partitioning algorithm with the potential for real-time calculation. An improvement on this kind of space partitioning algorithms for solving complex 3D models is presented. Numerical examples show that the efficiency of the improved algorithm is better than that of the original method. When the size of most target elements is smaller than the size of spatial grids, the efficiency of the improved method can be more than four times of that of the original method. An adaptive method of space partitioning based on the improved algorithm is developed by taking the surface element density or the curvature as the threshold for deep partitioning and conducting the deep partitioning using the octree method. A computer program implementation for applying the method in some typical applications is discussed, and the performance in terms of the efficiency, reliability, and resource use is evaluated. Application testing shows that the results of the adaptive spacing partitioning are more convenient for the follow-up use than that of the basic uniform space partitioning. Furthermore, when it is used to calculate the electromagnetic scattering of complex targets by the ray tracing (RT) method, the adaptive space partitioning algorithm can reduce the calculation time of the RT process by more than 40% compared with the uniform space segmentation algorithm.

Fusion algorithm of visible and infrared image based on anisotropic diffusion and image enhancement (capitalize only the first word in a title (or heading), the first word in a subtitle (or subheading), and any proper nouns).

Aiming at the situation that the existing visible and infrared images fusion algorithms only focus on highlighting infrared targets and neglect the performance of image details, and cannot take into account the characteristics of infrared and visible images, this paper proposes an image enhancement fusion algorithm combining Karhunen-Loeve transform and Laplacian pyramid fusion. The detail layer of the source image is obtained by anisotropic diffusion to get more abundant texture information. The infrared images adopt adaptive histogram partition and brightness correction enhancement algorithm to highlight thermal radiation targets. A novel power function enhancement algorithm that simulates illumination is proposed for visible images to improve the contrast of visible images and facilitate human observation. In order to improve the fusion quality of images, the source image and the enhanced images are transformed by Karhunen-Loeve to form new visible and infrared images. Laplacian pyramid fusion is performed on the new visible and infrared images, and superimposed with the detail layer images to obtain the fusion result. Experimental results show that the method in this paper is superior to several representative image fusion algorithms in subjective visual effects on public data sets. In terms of objective evaluation, the fusion result performed well on the 8 evaluation indicators, and its own quality was high.

A Partition Spatial Filtering Method for Acoustic Array Configuration

Acoustic array is a ubiquitous tool for locating and quantifying sound sources. However, its effectiveness depends greatly on the array configuration. This paper presents an array configuration method to enhance array performance, especially on the spatial resolution and the Doppler effect correction. The problem of array configuration is formulated into a position matrix determined by introducing partition spatial filtering. Irregular coaxial ring grid spacings and partition filtering conditions are suggested to control array spatial resolution. Geometrical parameters and performance indicators are constructed to quantify the relationships between the array configuration and performance. Based on these quantitative relations, the spatial variation of the array beam pattern and the Doppler effect has got adaptive adjustment. In particular, an adaptive partition algorithm is proposed to reduce computation time. The performance of the method is examined numerically and experimentally, which is compared with the other methods. The results provide the method to guide the design of a 64‐microphone optimized array with high performance (1.8° spatial angle resolution and 40% Doppler frequency correction over the bandwidth from 800 Hz to 3000 Hz) and fast computing speed (18 s array generated time for 2000 arrays). Furthermore, an unusual feature of the method is that it can be utilized in the case when the source moves at a nonconstant velocity.

Weighted Adaptive Partition for Heterogeneous IoT Data Stream

In the past, many researchers used sliding-window mechanism in Internet of Things (IoT) data stream mining, which can divide the incoming large volume data into small segments and improve the information processing rate. With the emergence of heterogeneous IoT data streams, data forms become richer and more diverse. The fusion analysis of multimodality information can improve the accuracy of information processing and network decision-making, because it can make up for the lack of some information in a single-mode data stream. However, the representation of information in different data sources is different, and the amount of information contained in it is distinct. Thus, a weighted adaptive partition algorithm is proposed in this article, which performs a dynamic weighted fusion of multisource heterogeneous IoT data in sliding windows. The article adopts the Gaussian mixture model (GMM) to adaptively determine the weights of each data source, which is calculated based on the judgement confidence of each modal data to the change in data stream. Besides, the probability distribution change detection mechanism based on multimodality restricted Boltzmann machine (multimodality RBM) is proposed, which measures the joint probability distribution by the weighted data’s energy value. Finally, according to the energy change rate of the data stream, the size of the window is adjusted adaptively to realize the partition for heterogeneous IoT data stream. The algorithm proposed in this article is verified on the CUAVE data set and the analysis of results shows that the data under adaptive weighted fusion has better performance.

A New Image Compression Algorithm Based on Non-Uniform Partition and U-System

JPEG lossy image compression is a still image compression algorithm model that is currently widely used in major network media. However, it is unsatisfactory in the quality of compressed images at low bit rates. The objective of this paper is to improve the quality of compressed images and suppress blocking artifacts by improving the JPEG image compression model at low bit rates. First, the image texture adaptive non-uniform rectangular partition (ITANRP) algorithm is proposed which partitions the image into <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$8\times 8$</tex-math></inline-formula> size image blocks with high texture complexity and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$16\times 16$</tex-math></inline-formula> size image blocks with low texture complexity. Then, a new transform coding based on the complete orthogonal U-system and all-phase digital filter (APDF) is proposed for coding image blocks with different sizes. Next, a flexible adaptive quantization scheme is designed to quantize image blocks with different sizes by considering the sensitivity of the human visual system (HVS) to different texture complexities. Finally, combining the above method with the JPEG model, a novel image compression algorithm model with low algorithm complexity is proposed to solve the problem in JPEG. The experimental results demonstrate that the performance of our algorithm model outperforms the JPEG image compression algorithms, the quality of the compressed image is greatly improved, and the blocking artifacts are also significantly suppressed.

NPC Three-Level Inverter Open-Circuit Fault Diagnosis Based on Adaptive Electrical Period Partition and Random Forest

Fault detection can increase the reliability and efficiency of power electronic converters employed in power systems. Among the converters in the power system, a Neutral Point Clamped (NPC) three-level inverter is most commonly used to drive electric motors. In this paper, a new approach for open-circuit fault detection and location of the NPC three-level inverter for a shifting process using a constant voltage-to-frequency ratio is proposed. In order to diagnose open-circuit fault in as short a time as possible, an adaptive electrical period partition (AEPP) algorithm is proposed to pick single electrical periods from real-time three-phase current signals. The Maximal Overlap Discrete Wavelet Transformation (MODWT) and Park’s Vector Modulus (PVM) are used for feature analysis and normalization of electrical period signals. The statistical characteristics of the electrical period signals are extracted, and a random forest model is constructed to realize the state classification. Compared with the traditional fault diagnosis method, the proposed algorithm finds fault locations quickly and accurately. The effectiveness and accuracy of the proposed algorithm are verified by experiments.

Scalable and Adaptive Fuzzy Grid Partitioning for Enhanced Rule Generation in Complex Decision-Making Systems

This research focuses on addressing the challenges of rule generation in complex decision-making systems by proposing a scalable and adaptive fuzzy grid partitioning approach. Traditional rule generation methods often struggle to handle large datasets and dynamic environments, leading to decreased accuracy and computational inefficiencies. In this study, we present a novel approach that integrates scalable and adaptive techniques to enhance the accuracy, efficiency, and interpretability of rule-based frameworks. The scalable fuzzy grid partitioning algorithm efficiently partitions the attribute space, allowing for the generation of rules in decision-making systems with a large number of data points. By incorporating data parallelization and dimensionality reduction techniques, the approach mitigates computational complexity while maintaining rule generation accuracy. Furthermore, the adaptive fuzzy grid partitioning algorithm dynamically adjusts the partitioning structure based on changing conditions, capturing evolving patterns and ensuring the relevancy and reliability of the generated rules over time. The generated rules are evaluated using fuzzy rule evaluation functions, which consider the degree of membership in the corresponding fuzzy grid cells. This evaluation process ranks and selects the rules based on their firing strengths, providing an interpretable decision-making framework for complex systems. The approach enhances the interpretability of the generated rules by capturing the uncertainties and complexities inherent in decision-making processes. To validate the effectiveness of the proposed approach, we conducted experiments using a credit risk assessment case example. The results demonstrate improved accuracy and efficiency compared to traditional rule generation methods. The generated rules offer transparency and insight into the factors influencing credit risk assessments, enabling informed decision-making. However, this research has some limitations, including potential dataset dependencies, the choice of fuzzy membership functions, computational complexity, and the need for further evaluation metrics and real-world implementation considerations. Future research should focus on addressing these limitations and exploring the applicability of the proposed approach in diverse domains. In conclusion, the scalable and adaptive fuzzy grid partitioning approach presented in this research offers a promising solution to the challenges of rule generation in complex decision-making systems. By addressing scalability, adaptability, and interpretability, this approach enhances the accuracy and efficiency of rule-based frameworks, paving the way for more effective decision support systems in various domains.

GraphA: Efficient Partitioning and Storage for Distributed Graph Computation

Distributed graph computation is central to applications ranging from language processing to social networks. However, natural graphs tend to have skewed power-law distributions where a small subset of the vertices have a large number of neighbors. Existing graph-parallel systems suffer from load imbalance, high communication cost, and inefficient processing. To address this problem, in this paper we present GraphA, an a daptive scheme for efficient large-scale graph computation. At the core of GraphA is an adaptive and uniform graph partitioning algorithm, which partitions the datasets by using an incremental number of mapping functions. GraphA further improves and leverages the ART index structure to realize fine-grained and low-cost graph storage. We have implemented GraphA both on Spark and on GraphLab. Extensive evaluation shows that GraphA significantly outperforms state-of-the-art graph-parallel systems (GraphX and PowerLyra) in ingress time, execution time and storage cost, for both real-world and synthetic graphs. GraphA achieves up to $7.1\times$ 7 . 1 × performance improvement over GraphX and 19.7 percent improvement over PowerLyra.

A novel adaptive fast partition algorithm based on CU complexity analysis in HEVC

High efficiency video coding (HEVC) is the latest video coding standard. Compared with H.264, the proposal of quad-tree structure not only improves the coding efficiency greatly but also increases complexity of coding. Therefore, this paper performed a fast-adaptive algorithm based on the complexity of images for intra prediction. Two values from micro and macro levels are considered for each coding unit block. We use entropy on the macroscopic level and texture contrast on the microscopic level. Relatively, large pictures always have more smooth blocks and there are more complex blocks in small pictures. To deal with this problem, this study uses adaptive process to make the values of entropy and texture contrast close to the ideal values of the current video, thereby making the division reasonable. Experimental results show that the proposed algorithm can reduce 34.6% coding time on average, with a loss of Bjontegaard delta rate of 0.8%.