Progressively Balanced Multi-class Neural Trees

In traditional neural trees (NTs), each internal node is designed as a neural network (NN), such as single- or two-layer neural networks, to determine which branch should be followed for an input sample. Because each NN contained in the internal nodes is designed separately, the produced NT does not consider overall effectiveness. Thus, the designed NT is usually not an optimal NT. In this study, the tree-structured multilayer neural network (TSMLNN) is proposed for classification. The TSMLNN is similar to an NT, which is the result of dividing a deep multilayer NN into many small sub-networks. The TSMLNN has the advantages of both a multilayer NN and an NT. In addition, the split method is proposed to determine how to split the network in the TSMLNN. The genetic algorithm is proposed to automatically search for the weights, activation threshold of each node and the proper number of nodes at each layer according to both the computing complexity and classification error rate in the TSMLNN, and the proposed TSMLNN tends to be optimal. A heuristic method is also proposed to help users to decide which TSMLNN is the best within the classification error rate range. Finally, the performance of the proposed TSMLNN is compared with that of state-of-the-art neural networks in experiments.

In the TBFC (Tree-Based Fog Computing) and TWTBFC (Two-Way TBFC) models the electric energy consumed by fog nodes and servers can be reduced in the fog computing (FC) model. Here, fog nodes are hierarchically structured in a height-balanced tree, where a root node is a cloud of servers, leaf nodes are edge nodes which communicate with devices, and each node receives data from child nodes and sends the processed data to a parent node. In the TWTBFC model, nodes send processed data to not only a parent node but also each child node. In order to reduce the network traffic in the TWTBFC model, only aggregate nodes at some level collect the output data of every other aggregate node, i.e. aggregate data. Since only target actuators are to be activated, the aggregate data has to be only delivered to target actuators. Nodes whose descendant actuators are target ones are relay nodes. On receipt of aggregate data, only relay nodes forward the aggregate data to the child nodes. We evaluate the new TWTBFC model in terms of energy consumption of nodes and number of messages transmitted to deliver aggregate data to edge nodes.

In routing protocol (RPL), a child node exclusively makes the decision in selecting a parent based on the Rank value. In this paper, we make the parent selection mechanism, a mutual decision between the potential parent node and the child node, which significantly improves the network throughput under a heavy and highly dynamic load. To demonstrate the effectiveness of our parent selection mechanism, we used it to balance the load to avoid congestion by preventing the child node from using the congested parent node. The proposed parent selection mechanism is implemented as an add-on feature in RPL and hence is called enhanced RPL (EN-RPL). Furthermore, we introduce a new objective function for efficient routing (OF-ER) that is based on the newly defined composite efficient routing (CER) metric. CER metric takes important metrics, namely, the link quality, queue utilization, lifetime of node, delay, and number of bottleneck nodes, into account for optimal parent selection. The simulation results demonstrate that EN-RPL lowers the packet loss by up to 37% and reduces energy consumption by up to 46.5% compared with the standard RPL. The evaluation results of OF-ER outperform the state-of-the-art load balancing-based OF with respect to the network stability, reliability, and energy efficiency.

We show that distributed source coding of multi-view images using adaptive modules-operation can come close to the Slepian-Wolf bound. In the systematic scenario considered, two parent nodes (PN1 and PN2) on sides and child nodes (CNs) are located between PNs, which are statistically depended. A PN sends the whole image whereas a CN only partially, using an adaptive distributed source coding at a rate close to H(CN|PN1, PN2). The proposed scheme allows independent encoding and jointly decoding of views. Experimental results show performance close to the information-theoretic limit. Furthermore, good performance of the proposed architecture with adaptive scheme shows significant improvement over previous work.

In order to carry out a comprehensive design description of the specific architectural model of AI, the auxiliary model of AI and architectural spatial intelligence is deeply integrated, and flexible design is carried out according to the actual situation. AI assists in the generation of architectural intention and architectural form, mainly supporting academic and working theoretical models, promoting technological innovation, and thus improving the design efficiency of the architectural design industry. AI-aided architectural design enables every designer to achieve design freedom. At the same time, with the help of AI, architectural design can complete the corresponding work faster and more efficiently. With the help of AI technology, through the adjustment and optimization of keywords, AI automatically generates a batch of architectural space design schemes. Against this background, the auxiliary model of architectural space design is established through the literature research of the AI model, the architectural space intelligent auxiliary model, and the semantic network and the internal structure analysis of architectural space. Secondly, to ensure compliance with the three-dimensional characteristics of the architectural space from the data source, based on the analysis of the overall function and structure of space design, the intelligent design of the architectural space auxiliary by Deep Learning is carried out. Finally, it takes the 3D model selected in the UrbanScene3D data set as the research object, and the auxiliary performance of AI’s architectural space intelligent model is tested. The research results show that with the increasing number of network nodes, the model fitting degree on the test data set and training data set is decreasing. The fitting curve of the comprehensive model shows that the intelligent design scheme of architectural space based on AI is superior to the traditional architectural design scheme. As the number of nodes in the network connection layer increases, the intelligent score of space temperature and humidity will continue to rise. The model can achieve the optimal intelligent auxiliary effect of architectural space. The research has practical application value for promoting the intelligent and digital transformation of architectural space design.

Wireless sensor networks (WSNs) are used to collect large amounts of data over a wide area. However, the sensor nodes used in WSNs have a short lifespan because they are generally battery-operated. Research is actively underway to increase the lifespan of WSNs via techniques such as energy harvesting and wireless power transfer (WPT). This study proposes a scheme for clustering and adjusting the WSN data collection rate to alleviate the hotspot problem and improve network connectivity. The proposed scheme is targeted at WSNs in which the sensor nodes harvest solar energy. In addition, an unmanned aerial vehicle (UAV) traverses along a given path and delivers energy to sensor nodes using WPT, while collecting data as a sink node. In this scheme, the number of cluster heads is determined by considering the number of nodes at each hop distance and the maximum amount of data that can be transferred with the available energy. Further, sensor nodes limit the amount of data they collect to reduce the relay load on intermediate nodes. To achieve this, nodes consider the amount of data transferred by parent nodes, especially nodes in the hotspot, and the number of child nodes. The sink node traverses along a given path, and collects the data accumulated in the cluster head while supplying its remaining energy to the cluster head. The amount of energy it transfers is determined by considering the number of cluster heads. Consequently, nodes in the hotspot are prevented from becoming blacked out from a lack of energy, and the energy harvesting efficiency is increased. Simulation results indicate that the proposed scheme reduces the number of blackout nodes in the hotspot areas near cluster heads. Furthermore, the data collection and monitoring performance increases with the increase in network connectivity.

With the widespread application of wireless sensor networks, localization issue has attracted much attention. It is a major challenge for many sensor network tasks to locate a large number of asynchronous nodes in an unknown environment without external references. In this article, we present an asynchronous collaborative localization system (ACLS) to address the localization challenge for large-capacity sensor networks (LCSNs). ACLS exploits a hierarchical architecture, under which the wireless sensor nodes in the network are divided into parent nodes that can communicate with each other through wireless signals and child nodes that can only passively receive signals. Specific protocols and nonlinear distance estimators for this broadcast communication ranging technique are proposed. These characteristics are verified through theoretical analyses to have a strong suppression effect on local clock errors and are not sensitive to measurement noise. The simulation experiments further illustrate that the proposed ACLS can achieve high-rate and high-precision ranging and localization for asynchronous LCSN without preinstalled infrastructures.

We consider t-ary trees characterized by their numbers of nodes and their total path length. When t=2 these are called binary trees, and in such trees a parent node may have up to t child nodes. We give asymptotic expansions for the total number of trees with nodes and path length p, when n and p are large. We consider several different ranges of n and p. For n→∞ and p=O(n^{3/2}) we recover the Airy distribution for the path length in trees with many nodes, and also obtain higher order asymptotic results. For p→∞ and an appropriate range of n we obtain a limiting Gaussian distribution for the number of nodes in trees with large path lengths. The mean and variance are expressed in terms of the maximal root of the Airy function. Singular perturbation methods, such as asymptotic matching and WKB type expansions, are used throughout, and they are combined with more standard methods of analytic combinatorics, such as generating functions, singularity analysis, saddle point method, etc. The results are applicable to problems in information theory, that involve data compression schemes which parse long sequence into shorter phrases. Numerical studies show the accuracy of the various asymptotic approximations. Key Words: Trees; Universal Types; Asymptotics; Path Length; Singular Perturbations

Non-Volatile Memories (NVMs) require security mechanisms, e.g., counter mode encryption and integrity tree verification, which are important to protect systems in terms of encryption and data integrity. These security mechanisms heavily rely on extra security metadata that need to be efficiently and accurately recovered after system crashes or power off. Established SGX integrity tree (SIT) becomes efficient to protect system integrity and however fails to be restored from leaves, since the computations of SIT nodes need their parent nodes as inputs. To recover the security metadata with low write overhead and short recovery time, we propose an efficient and instantaneous persistence scheme, called STAR, which instantly persists the modifications of security metadata without extra memory writes. STAR is motivated by our observation that the parent nodes in cache are modified due to persisting their child nodes. STAR stores the modifications of parent nodes in their child nodes and persists them just using one atomic memory write. To eliminate the overhead of persisting the modifications, STAR coalesces the modifications and MACs in the evicted metadata. For fast recovery and verification of the metadata, STAR uses bitmap lines in asynchronous DRAM refresh (ADR) to indicate the locations of stale metadata, and constructs a cached merkle tree to verify the correctness of the recovery process. Our evaluation results show that compared with state-of-the-art work, our proposed STAR delivers high performance, low write traffic, low energy consumption and short recovery time.

In recent years, program evolution algorithms based on the estimation of distribution algorithm (EDA) have been proposed to improve search ability of genetic programming (GP) and to overcome GP-hard problems. One such method is the probabilistic prototype tree (PPT) based algorithm. The PPT based method explores the optimal tree structure by using the full tree whose number of child nodes is maximum among possible trees. This algorithm, however, suffers from problems arising from function nodes having different number of child nodes. These function nodes cause intron nodes, which do not affect the fitness function. Moreover, the function nodes having many child nodes increase the search space and the number of samples necessary for properly constructing the probabilistic model. In order to solve this problem, we propose binary encoding for PPT. In this article, we convert each function node to a subtree of binary nodes where the converted tree is correct in grammar. Our method reduces ineffectual search space, and the binary encoded tree is able to express the same tree structures as the original method. The effectiveness of the proposed method is demonstrated through the use of two computational experiments.

Transmission links in wireless sensor networks (WSNs) are prone to failure because of breakdown, movement, and energy depletion of nodes. Aiming at node-intensive environment, this paper proposes a highly reliable forwarding model based on virtual nodes (HRVN) in tree-based topology networks. A virtual node (VN) consists of several neighboring physical sensor nodes, which have common parent node and children nodes. Normally, the physical sensor nodes in the same VN forward packets in a load-balancing scheme. When a sensor node in a VN is in failure, the others would take over its forwarding work immediately. Moreover, the internal changes of a VN are transparent to its parent and children nodes. The experiments prove the model can provide load-balance and fast recovery from failure with low cost.

Abstract—This work proposes a novel congestion management protocol for constrained delay traffic and optimized rate control for HWSNs. The proposed protocol consists of congestion avoidance and control. We distinguish between high and low priority traffics which are serviced upon their priority and QoS requirements. For high priority traffic, we control end to end delay constrains with node output scheduling weights. For low priority traffics firstly we avoid congestion using a new Active Queue Management (AQM) algorithm that uses distinct virtual queue's situation for a single physical queue in order to decide accepting or dropping the received packets from child nodes. If acceptance occurs for the incoming packet, congestion is detected by the proposed protocol using a three-state machine as well as a virtual queue status. Afterward, child’s sending rate is adjusted using an optimization function. Results of simulation indicate goal achievement for the protocol proposed.

In order to increase the performance of the IoT (Internet of Things), the FC (Fog Computing) model is proposed. Here, subprocesses of an application process to handle sensor data are performed on fog nodes in addition to servers. In this paper, we newly introduce join subprocesses with multiple input parameters. If a node gets faulty, the child nodes are disconnected and have to be reconnected to new parent nodes. In our previous studies, new parent nodes are selected at the same level as the faulty node. In this paper, we newly propose a GTBFC (General TBFC) algorithm by which disconnected nodes are reconnected to new parent nodes so that the data obtained from the output data of every disconnected node is processed by an ancestor join node of the faulty node. In the evaluation, we show the energy consumption and execution time of a new parent node selected by the GTBFC algorithm.

The information of the Gene Ontology annotation is helpful in the explanation of life science phenomena, and can provide great support for the research of the biomedical field. The use of the Gene Ontology is gradually affecting the way people store and understand bioinformatic data. To facilitate the prediction of gene functions with the aid of text mining methods and existing resources, we transform it into a multi-label top-down classification problem and develop a method that uses the hierarchical relationships in the Gene Ontology structure to relieve the quantitative imbalance of positive and negative training samples. Meanwhile the method enhances the discriminating ability of classifiers by retaining and highlighting the key training samples. Additionally, the top-down classifier based on a tree structure takes the relationship of target classes into consideration and thus solves the incompatibility between the classification results and the Gene Ontology structure. Our experiment on the Gene Ontology annotation corpus achieves an F-value performance of 50.7% (precision: 52.7% recall: 48.9%). The experimental results demonstrate that when the size of training set is small, it can be expanded via topological propagation of associated documents between the parent and child nodes in the tree structure. The top-down classification model applies to the set of texts in an ontology structure or with a hierarchical relationship.

In this paper, a hierarchical learning algorithm based on the Bayesian Neural Network classifier with backtracking is proposed to support large-scale image classification, where a Visual Confusion Label Tree is established for constructing a hierarchical structure for large numbers of categories in image datasets and determining the hierarchical learning tasks automatically. Specifically, the Visual Confusion Label Tree is established based on outputs of convolution neural network models. One parent node on the Visual Confusion Label Tree contains a set of sibling coarse-grained categories, and child nodes have several sets of fine-grained categories which are partitions of categories on the parent node. The proposed Hierarchical Bayesian Neural Network with backtracking algorithm can benefit from the hierarchical structure of the Visual Confusion Label Tree. Focusing on those confusion subsets instead of the entire set of categories makes the classification ability of the tree classifier stronger. The backtracking algorithm can utilize the uncertainty information captured from the Bayesian Neural Network to make a second classification to re-correct samples that were classified incorrectly in the previous classification process. Experiments on four large-scale datasets show that our tree classifier obtains a significant improvement over the state-of-the-art tree classifier, which have demonstrated the discriminative hierarchical structure of our Visual Confusion Label Tree and the effectiveness of our Hierarchical Bayesian Neural Network with backtracking algorithm.