Hierarchical Aggregation Research Articles

The high-speed networks of the Summit and Sierra supercomputers

Oak Ridge National Laboratory's Summit supercomputer and Lawrence Livermore National Laboratory's Sierra supercomputer utilize InfiniBand interconnect in a Fat-tree network topology, interconnecting all compute nodes, storage nodes, administration, and management nodes into one linearly scalable network. These networks are based on Mellanox 100-Gb/s EDR InfiniBand ConnectX-5 adapters and Switch-IB2 switches, with compute-rack packaging and cooling contributions from IBM. These devices support in-network computing acceleration engines such as Mellanox Scalable Hierarchical Aggregation and Reduction Protocol, graphics processor unit (GPU) Direct RDMA, advanced adaptive routing, Quality of Service, and other network and application acceleration. The overall IBM Spectrum Message Passing Interface (MPI) messaging software stack implements Open MPI, and was a collaboration between IBM, Mellanox, and NVIDIA to optimize direct communication between endpoints, whether compute nodes (with IBM POWER CPUs, NVIDIA GPUs, and flash memory devices), or POWER-hosted storage nodes. The Fat-tree network can isolate traffic among the compute partitions and to/from the storage subsystem, providing more predictable application performance. In addition, the high level of redundancy of this network and its reconfiguration capability ensures reliable high performance even after network component failures. This article details the hardware and software architecture and performance of the networks and describes a number of the high-performance computing (HPC) enhancements engineered into this generation of InfiniBand.

EHANet: An Effective Hierarchical Aggregation Network for Face Parsing

In recent years, benefiting from deep convolutional neural networks (DCNNs), face parsing has developed rapidly. However, it still has the following problems: (1) Existing state-of-the-art frameworks usually do not satisfy real-time while pursuing performance; (2) similar appearances cause incorrect pixel label assignments, especially in the boundary; (3) to promote multi-scale prediction, deep features and shallow features are used for fusion without considering the semantic gap between them. To overcome these drawbacks, we propose an effective and efficient hierarchical aggregation network called EHANet for fast and accurate face parsing. More specifically, we first propose a stage contextual attention mechanism (SCAM), which uses higher-level contextual information to re-encode the channel according to its importance. Secondly, a semantic gap compensation block (SGCB) is presented to ensure the effective aggregation of hierarchical information. Thirdly, the advantages of weighted boundary-aware loss effectively make up for the ambiguity of boundary semantics. Without any bells and whistles, combined with a lightweight backbone, we achieve outstanding results on both CelebAMask-HQ (78.19% mIoU) and Helen datasets (90.7% F1-score). Furthermore, our model can achieve 55 FPS on a single GTX 1080Ti card with 640 × 640 input and further reach over 300 FPS with a resolution of 256 × 256, which is suitable for real-world applications.

In situ morphological characterization of soursop (Annona muricata L.) plants in Manabí, Ecuador

Soursop (Annona muricata L.) germplasm, from the central region of the Ecuadorian littoral, has diverse phenotypic characteristics, which have been little studied. The main objective of this research was to characterize in situ the morphological variability of 60 accessions of soursop. The morphological traits were evaluated using 20 quantitative and 10 qualitative variables of the plant and the fruit. Multivariate statistical methods were used as principal components, hierarchical conglomerate aggregation, discriminant and correlation analysis (Pearson) were performed. Descriptive analysis such as central tendency and dispersion (coefficient of variation, range, mean and standard deviation) were determined for the quantitative data; while frequencies were calculated for qualitative data. In terms of morphological traits, 74 % of the total variability was explained by the fourth component. Three conglomerates of similarity were formed, in which the height, diameter of canopy and fruit number per tree, were the characters of greater contribution for its conformation.

Quality-Aware Feature Aggregation Network for Robust RGBT Tracking

RGBT tracking becomes a popular computer vision task, and has a variety of applications in visual surveillance systems, self-driving cars and intelligent transportation system. This paper investigates how to perform robust visual tracking in adverse and challenging conditions using complementary visual and thermal infrared data (RGBT tracking). We propose a novel deep network architecture called quality-aware Feature Aggregation Network (FANet) for robust RGBT tracking. Unlike existing RGBT trackers, our FANet aggregates hierarchical deep features within each modality to dispose the challenge of significant changes in appearance which is triggered by low illumination,deformation, background clutter and occlusion. In particular, we employ the operations of max pooling to transform these hierarchical and multi-resolution features into uniform space with the same resolution, and use 1×1 convolution operation to compress feature dimensions to achieve more effective hierarchical feature aggregation. To model the interactions between RGB and thermal modalities, we elaborately design an adaptive aggregation subnetwork to integrate features from different modalities based on their reliabilities and thus are able to alleviate noise effects introduced by low-quality sources. The whole FANet is trained in an end-to-end manner. Extensive experiments on large-scale benchmark datasets demonstrate the high-accurate performance against other state-of-the-art RGBT tracking methods.

Assessment of aggregation strategies for machine-learning based short-term load forecasting

Effective short-term load forecasting (STLF) plays an important role in power system operations. It is challenging to identify an ML model that has the outperformance in all scenarios. Therefore, there are a number of aggregation strategies developed to improve STLF. However, the superiority of these aggregation strategies has not been assessed. In this paper, STLF with three aggregation strategies are developed, which are information aggregation (IA), model aggregation (MA), and hierarchy aggregation (HA). The IA, MA, and HA strategies aggregate inputs, models, and forecasts at the pre-forecasting, model-building, or post-forecasting stage, respectively. To verify the effectiveness of the three aggregation strategies, a set of 10 models based on 4 machine-learning algorithms are developed in each aggregation category to predict 1-hour-ahead load. Case studies show that: (i) STLF-IA presents superior performance than STLF with weather data and STLF with individual load data consistently, and the performance can be further enhanced by the recursive feature elemination (RFE) feature selection method; (ii) MA improves the STLF robustness by reducing the risk of unsatisfactory single-algorithm STLF models; and (iii) STLF-HA produces the most accurate forecasts with a 0.83% normalized mean absolute error and a 1.35% mean absolute percentage error, while keeping hierarchical aggregate consistency.

Joint Bottom-Up Method for Hierarchical Time-Series: Application to Australian Tourism

Many applications involve a hierarchy of time-series, where values at the bottom level aggregate to values at higher levels. Forecasts of such hierarchical data need to be accurate, probabilistic, and coherent in the sense of respecting hierarchical aggregation. While recent developments have explicitly modeled every time-series in the hierarchy, we show, under general conditions, that hierarchical data can be modeled jointly by considering only its bottom-level series and their contemporaneous covariance. Inspired by this result, we devise a Bayesian method that models bottom-level series jointly, takes into account their contemporaneous covariance, and performs automatic selection of lag terms, both within and across series. The model copes with high-dimensional data, and outputs both point and probabilistic forecasts. Additionally, it returns posterior distributions of all parameters, which can be used for inference. As a case study, we apply our method to make recommendations on planning and promotion of domestic tourism in Australia. Our model reveals the hidden spatio-temporal dynamics of different types of domestic tourism in Australia, and allows us to explore how promotional investments could be localized to develop tourism in accordance with the declared desiderata of the Australian government.

Interactive exploration of large time-dependent bipartite graphs

Abstract Bipartite graphs are typically visualized using linked lists or matrices, but these visualizations neither scale well nor do they convey temporal development. We present a new interactive exploration interface for large, time-dependent bipartite graphs. We use two clustering techniques to build a hierarchical aggregation supporting different exploration strategies. Aggregated nodes and edges are visualized as linked lists with nested time series. We demonstrate two use cases: finding advertising expenses of public authorities following similar temporal patterns and comparing author-keyword co-occurrences across time. Through a user study, we show that linked lists with hierarchical aggregation lead to more insights than without.

Hierarchical aggregation of energy performance indicators in continuous production processes

The mitigation of the climate change requires a significant reduction of the fossil energy consumption in all industrial sectors. The implementation of formalized management systems supports the industry to continuously improve the energy performance which is measured using so called “Energy Performance Indicators”. One essential requirement for the evaluation is the correction of these indicators and the corresponding baselines by the influences of external static or dynamic factors, e.g. the ambient conditions, the product spectrum or the plant load. This is in particular difficult for large integrated production sites as e.g. in the chemical industry. In this contribution, an aggregation method is proposed to exploit the analysis of the factors on lower hierarchical layers for the evaluation of the performance of an aggregated domain. Thereby, the resulting aggregated indicator and the corresponding baseline consider all the identified factors from the lower layers, which facilitates the analysis and allocation of possible savings potentials. The concept is applied exemplarily to an integrated chemical production site and the contributions of each plant to the deviation of the energy performance of the site are analyzed. The method facilitates the verification of the improvement of the energy performance as required by ISO 50001:2018 and helps decision makers to prioritize investments in energy efficiency projects. The results can be used for discussions with policy-makers, certification bodies and other stakeholders on energy efficiency targets.

Approximate dynamic programming for planning a ride-hailing system using autonomous fleets of electric vehicles

We address a comprehensive ride-hailing system taking into account many of the decisions required to operate it in reality. The ride-hailing system is formed of a centrally managed fleet of autonomous electric vehicles which is creating a transformative new technology with significant cost savings. This problem involves a dispatch problem for assigning riders to cars, a surge pricing problem for deciding on the price per trip and a planning problem for deciding on the fleet size. We use approximate dynamic programming to develop high-quality operational dispatch strategies to determine which car is best for a particular trip, when a car should be recharged, when it should be re-positioned to a different zone which offers a higher density of trips and when it should be parked. These decisions have to be made in the presence of a highly dynamic call-in process, and assignments have to take into consideration the spatial and temporal patterns in trip demand which are captured using value functions. We prove that the value functions are monotone in the battery and time dimensions and use hierarchical aggregation to get better estimates of the value functions with a small number of observations. Then, surge pricing is discussed using an adaptive learning approach to decide on the price for each trip. Finally, we discuss the fleet size problem.

MAP-Vis: A Distributed Spatio-Temporal Big Data Visualization Framework Based on a Multi-Dimensional Aggregation Pyramid Model

During the exploration and visualization of big spatio-temporal data, massive volume poses a number of challenges to the achievement of interactive visualization, including large memory consumption, high rendering delay, and poor visual effects. Research has shown that the development of distributed computing frameworks provides a feasible solution for big spatio-temporal data management and visualization. Accordingly, to address these challenges, this paper adopts a proprietary pre-processing visualization scheme and designs and implements a highly scalable distributed visual analysis framework, especially targeted at massive point-type datasets. Firstly, we propose a generic multi-dimensional aggregation pyramid (MAP) model based on two well-known graphics concepts, namely the Spatio-temporal Cube and 2D Tile Pyramid. The proposed MAP model can support the simultaneous hierarchical aggregation of time, space, and attributes, and also later transformation of the derived aggregates into discrete key-value pairs for scalable storage and efficient retrieval. Using the generated MAP datasets, we develop an open-source distributed visualization framework (MAP-Vis). In MAP-Vis, a high-performance Spark cluster is used as a parallel preprocessing platform, while distributed HBase is used as the massive storage for the generated MAP data. The client of MAP-Vis provides a variety of correlated visualization views, including heat map, time series, and attribute histogram. Four open datasets, with record numbers ranging from the millions to the tens of billions, are chosen for system demonstration and performance evaluation. The experimental results demonstrate that MAP-Vis can achieve millisecond-level query response and support efficient interactive visualization under different queries on the space, time, and attribute dimensions.

A decentralized application for the traceability process in the pharma industry

With the growth of a global market, new regulations aiming at preventing the fraud of counterfeiting drugs have been released all over the world. In the pharma industry, an innovative technology named serialization is becoming more and more popular as it allows to implement a packaging process based on a hierarchical aggregation. This guarantees that each box entering in the distribution network is marked with a unique identifier for an easy traceability by a central regulatory in charge to follow the life cycle of the product until given to the final patient.In this scenario, the blockchain might offer a breakthrough proposing a decentralized and immutable traceability mechanism able to increase the security of the data and to reduce the success of a fraud attempt. To demonstrate the effectiveness of this technology, in this paper, a DAPP based on the Ethereum blockchain has been coded and tested as prototype in a pharma industry.

Online Algorithms for Multilevel Aggregation

Online Algorithms for Hierarchical Aggregation Problems Data and inventory aggregation problems arise in multicasting, sensor networks, communication in organization hierarchies, and in supply chain management. These problems are naturally online, in the sense that aggregation decisions need to be made without information about future requests. We study these problems with a general tree structure of links that can be used for deliveries. This generalizes some well-studied optimization problems: trees of depth one capture the TCP acknowledgment problem, and trees of depth two capture the joint replenishment problem. For trees of depth one and two, constant-competitive online algorithms are known. We solve a major open problem by giving a constant-competitive algorithm for trees of arbitrary (fixed) depth. The algorithm works for arbitrary waiting cost functions, including the variant with deadlines.

Seed-Assisted Synthesis and Catalytic Performance of Nano-sized ZSM-5 Aggregates in a One-Step Crystallization Process

Hierarchical nano-sized ZSM-5 aggregates were successfully synthesized via a seed-assisted method in the presence of cetyltrimethylammonium bromide (CTAB) through a facile one-step crystallization process. Commercial ZSM-5 zeolites with a SiO2/Al2O3 ratio comparable to that of ZSM-5 products were treated with alkali and used as the seed particles. The influences of crystallization conditions were investigated, and the possible synthesis mechanism was proposed. ZSM-5 zeolites with different amounts of CTAB added were characterized using many techniques and evaluated in toluene alkylation with methanol. The results showed that a trace amount of CTAB significantly promoted the crystallization of ZSM-5 zeolite, with the morphology changing from hexagonal-shape crystals to uniform spherical aggregates. CTAB may act as the structure-directing agent and assemble the primary crystallites to generate hierarchical ZSM-5 aggregates. The ZSM-5 zeolite with the smallest primary particles of 50–80 nm exhibited large specific surface area, abundant mesopores, and the greatest microporosity. The hierarchical nano-sized ZSM-5 aggregate showed higher toluene conversion and a longer lifetime without compromising the selectivity to xylene and p-xylene in toluene alkylation with methanol.

Hierarchical approach to aggregate equilibria

The authors propose a thermodynamic model to describe hierarchical aggregation. They find that the degree of aggregation, volume, and number fraction at each level determine the change in free energy, enthalpy, and entropy of aggregation/dissociation as well as the particle size distributions at each level. This model enables the discussion of energetics for equilibrated polydisperse hierarchical materials

Remote Sensing Scene Classification by Gated Bidirectional Network

Remote sensing (RS) scene classification is a challenging task due to various land covers contained in RS scenes. Recent RS classification methods demonstrate that aggregating the multilayer convolutional features, which are extracted from different hierarchical layers of a convolutional neural network, can effectively improve classification accuracy. However, these methods treat the multilayer convolutional features as equally important and ignore the hierarchical structure of multilayer convolutional features. Multilayer convolutional features not only provide complementary information for classification but also bring some interference information (e.g., redundancy and mutual exclusion). In this paper, a gated bidirectional network is proposed to integrate the hierarchical feature aggregation and the interference information elimination into an end-to-end network. First, the performance of each convolutional feature is quantitatively analyzed and a superior combination of convolutional features is selected. Then, a bidirectional connection is proposed to hierarchically aggregate multilayer convolutional features. Both the top–down direction and the bottom–up direction are considered to aggregate multilayer convolutional features into the semantic-assist feature and appearance-assist feature, respectively, and a gated function is utilized to eliminate interference information in the bidirectional connection. Finally, the semantic-assist feature and appearance-assist feature are merged for classification. The proposed method can compete with the state-of-the-art methods on four RS scene classification data sets (AID, UC-Merced, WHU-RS19, and OPTIMAL-31).

A Novel Hierarchical Data Aggregation with Particle Swarm Optimization for Internet of Things

Due to small battery powered devices and inefficient utilization of resources, the sensor nodes in Internet of Things (IoT) may be lost prematurely. In order to extend the lifetime and avoid energy-hole problem, a novel hierarchical data aggregation with particle swarm optimization for Wireless Sensor Networks (HDA-PSO) is proposed. Firstly, the fitness function is designed from multiple relational matrices, including the residual energy, the average distance among adjacent nodes and the centroid degree of the covering region. Secondly, in order to effectively encircling the optimal solution with initial position of particles, we propose a population initialization method based on beta distribution according to the distribution characteristics of nodes in sensor networks. Next, based on differential evolution, a novel operator is introduced for velocity update which effectively balances the exploration and development of particle swarm optimization. The experimental results show that the proposed algorithm can effectively balance the energy consumption of nodes under different node’s density, improve the energy efficiency and prolong the lifetime of network significantly.

Hierarchical multi-view aggregation network for sensor-based human activity recognition.

Sensor-based human activity recognition aims at detecting various physical activities performed by people with ubiquitous sensors. Different from existing deep learning-based method which mainly extracting black-box features from the raw sensor data, we propose a hierarchical multi-view aggregation network based on multi-view feature spaces. Specifically, we first construct various views of feature spaces for each individual sensor in terms of white-box features and black-box features. Then our model learns a unified representation for multi-view features by aggregating views in a hierarchical context from the aspect of feature level, position level and modality level. We design three aggregation modules corresponding to each level aggregation respectively. Based on the idea of non-local operation and attention, our fusion method is able to capture the correlation between features and leverage the relationship across different sensor position and modality. We comprehensively evaluate our method on 12 human activity benchmark datasets and the resulting accuracy outperforms the state-of-the-art approaches.

Thermodynamics of Hierarchical Aggregation in Pigment Dispersions.

Many commercially and industrially important materials aggregate to form nanoscale mass-fractal structures. Unlike hard aggregates such as fumed silica, aqueous pigment-based inks consist of weakly bound nanoparticles stabilized by a surfactant. These soft aggregates can easily break apart and re-form balancing mixing energy and the reduction in surface energy with clustering or aggregation. Rapid thermal motion of small elemental crystallites leads to dense clusters or primary particles. The larger primary particles have slower thermal motion and aggregate into ramified mass fractals to form a dual-level hierarchical structure. It is proposed that the hierarchical structure relies on subtle and competitive equilibria between the different hierarchical structural levels. A new hierarchical thermodynamics model by Vogtt is used. Pigment yellow 14 and pigment blue 15:3 as surfactant-stabilized aqueous dispersions were employed to explore the thermodynamics of nanoparticle hierarchical equilibria. It was demonstrated that reversible nanoparticle aggregation can be described solely by the change in free energy of dissociation and the change in free energy of mixing in the context of a subunit being removed from a cluster. The hierarchical thermodynamics is dominated by the solubility of the dispersing surfactant. At the cloud point for the surfactant, primary particles approach the size of an elemental particle and the degree of aggregation becomes very large. The results indicate that subtle and reproducible control over pigment hierarchical structure and size is possible through thermal equilibration, manipulation of the surfactant properties, and elemental crystallite size.

Multi-Core Processing Cloud Eclat Growth

Data mining is a lively process used in many leading technologies of this information era. Eclat growth is one of the best performance data mining algorithms. This work is indented to create a suave interface for Eclat growth algorithm to run in multi-core processor-based cloud computing environments. Recent improvements in processor manufacturing technology make it possible to create multi-core high performance Central Processing Units (CPUs) and Graphics Processing Units (GPUs). Many cloud services are already providing accessibility to these high-power processor virtual machines. The process of blending these technologies with Eclat Growth is proposed here in the name of “Multi-core Processing Cloud Eclat Growth” (MPCEG) to achieve higher processing speeds without compromising the standard data mining metrics such as Accuracy, Precision, Recall and F1-Score. New procedures for Cloud Parallel Processing, GPU Utilization, Annihilation of floating point arithmetic errors by fixed point replacement in GPUs and Hierarchical offloading aggregation are introduced in the construction process of proposed MPCEG.

Monocular depth estimation with geometrical guidance using a multi-level convolutional neural network

Depth estimation using monocular images is highly challenging but is a considerable topic in understanding scene structure. This paper proposed a multi-level convolutional neural network (CNN), in which the low-level and high-level features were well-integrated in order to estimate the depth values from a single image. To estimate the depth values, a fully convolutional architecture which used a structure improvement strategy was applied to correct the depth values using low-level features of the shallow layers. A hierarchical context aggregation scheme was proposed according to the dilated convolutional operators that integrate the global and local contexts in a progressive way to recover the local details. In addition, a rectifying block was used to subtract the existed fitting residuals of integrated multi-level features. In the second level, a modifier network was provided to improve the estimated depth values, in particular in the object boundaries. It is possible to determine the objective function in the modifier networks by considering the geometrical features directing the network to achieve the best results. The proposed framework was evaluated using the computer vision (Make3D, NYU, and SUN datasets) and remote sensing (Vaihingen and Potsdam datasets) datasets that specified the significant performance of the proposed framework.