Tree-structured Architecture Research Articles

Hierarchical mixture-of-experts models for count variables with excessive zeros

We propose a tree-structured hierarchical architecture for estimating count variables that exhibit both excessive zeros and over-dispersion, which is commonly observed in practice. The underlying statistical model is based on the Mixture-of-experts (MoE) model, a generalization of finite mixture regression models. With two levels of experts, the proposed model can efficiently model both excessive zeros and the long right-tail, controlled by two gating networks at different levels in the hierarchy. We develop the general form of the maximum likelihood estimator for the proposed model based on the EM algorithm with two latent variables, which allows a natural interpretation and convenient optimization for the likelihood function. As a numerical application, we apply the proposed model to a well-known real dataset and found that it shows superior performance and allows better interpretations compared to other existing alternatives.

Efficient VLSI Architecture for Lifting-Based Discrete Wavelet Packet Transform

This brief presents a novel very large-scale integration (VLSI) architecture for discrete wavelet packet transform (DWPT). By exploiting the in-place nature of the DWPT algorithm, this architecture has an efficient pipeline structure to implement high-throughput processing without any on-chip memory/first-in first out access. A folded architecture for lifting-based wavelet filters is proposed to compute the wavelet butterflies in different groups simultaneously at each decomposition level. According to the comparison results, the proposed VLSI architecture is more efficient than the previous proposed architectures in terms of memory access, hardware regularity and simplicity, and throughput. The folded architecture not only achieves a significant reduction in hardware cost but also maintains both the hardware utilization and high-throughput processing with comparison to the direct mapped tree-structured architecture

Specifying distributed multi-agent systems in chemical reaction metaphor

This paper presents an application of Chemical Reaction Metaphor (CRM) in distributed multi-agent systems (MAS). The suitability of using CRM to model multi-agent systems is justified by CRM's capacity in specifying dynamic features of multi-agent systems. A case study in an agent-based e-learning system (course material updating) demonstrates how the CRM based language, Gamma, can be used to specify the architectures of multi-agent systems. The effectiveness of specifying multi-agent systems in CRM from the view point of software engineering is further justified by introducing a transformational method for implementing the specified multi-agent systems. A computation model with a tree-structured architecture is proposed to base the design of the specified multi-agent system during the implementation phase. A module language based on the computation model is introduced as an intermediate language to facilitate the translation of the specification of multi-agent systems. The multicast networking technology pragmatizes the implementation of communications and synchronization among distributed agents. The paper also discusses the feasibility of implementing an automatic translation from the Gamma specification to a program in the module language.

Support vector machines experts for time series forecasting

This paper proposes using the support vector machines (SVMs) experts for time series forecasting. The generalized SVMs experts have a two-stage neural network architecture. In the first stage, self-organizing feature map (SOM) is used as a clustering algorithm to partition the whole input space into several disjointed regions. A tree-structured architecture is adopted in the partition to avoid the problem of predetermining the number of partitioned regions. Then, in the second stage, multiple SVMs, also called SVM experts, that best fit partitioned regions are constructed by finding the most appropriate kernel function and the optimal free parameters of SVMs. The sunspot data, Santa Fe data sets A, C and D, and the two building data sets are evaluated in the experiment. The simulation shows that the SVMs experts achieve significant improvement in the generalization performance in comparison with the single SVMs models. In addition, the SVMs experts also converge faster and use fewer support vectors.

Improved financial time series forecasting by combining Support Vector Machines with self-organizing feature map

A two-stage neural network architecture constructed by combining Support Vector Machines (SVMs) with self-organizing feature map (SOM) is proposed for financial time series forecasting. In the first stage, SOM is used as a clustering algorithm to partition the whole input space into several disjoint regions. A tree-structured architecture is adopted in the partition to avoid the problem of predetermining the number of partitioned regions. Then, in the second stage, multiple SVMs, also called SVM experts, that best fit each partitioned region are constructed by finding the most appropriate kernel function and the optimal learning parameters of SVMs. The Santa Fe exchange rate and five real futures contracts are used in the experiment. It is shown that the proposed method achieves both significantly higher prediction performance and faster convergence speed in comparison with a single SVM model.

Realisation of Hierarchical Look-Up Tables on Low-Cost Hardware

Numerous applications which require nonlinear models for control, supervision, or fault detection are characterized by real-time requirements and limited computer power. The most widely used type of nonlinear models in such time-critical applications are grid-based look-up tables. However, these models fully underlie the curse of dimensionality. Therefore, their use is restricted to at most two-dimensional input spaces. In this paper, an alternative look-up table concept based on a hierarchical input space decomposition is proposed. These look-up tables can be efficiently represented in a tree-structured architecture which allows the implementation on lowcost hardware.

Intelligent motion control

Intelligent motion control is surveyed. Two basic hierarchical architectures of the intelligent motion control, the tree-structured architecture and the nested architecture, are shown. The intelligent control methods with three categories of the model-based, the knowledge-based, and the soft computing methods are also shown. The intelligent motion controls are characterized by the problems of uncertain and inexact information, high complexity and ill structure, and real-time requirement. With the architecture and the control methods, the problems are shown to be solved successfully.

Hierarchical Mixtures of Experts and the EM Algorithm

We present a tree-structured architecture for supervised learning. The statistical model underlying the architecture is a hierarchical mixture model in which both the mixture coefficients and the mixture components are generalized linear models (GLIM's). Learning is treated as a maximum likelihood problem; in particular, we present an Expectation-Maximization (EM) algorithm for adjusting the parameters of the architecture. We also develop an on-line learning algorithm in which the parameters are updated incrementally. Comparative simulation results are presented in the robot dynamics domain.

A hierarchical representation for random fields

The conventional tree-structured hierarchical architectures, e.g. quadtrees, are defined for one frame of an image. Such a frame, however, is frequently a single realization from a family of images belonging to a random field. By allowing parent-children nodes to exhibit a continuum of strengths, as opposed to the on/off case for the conventional representation, a hierarchical description for random fields is proposed.

A Tree-structured Database Machine for large relational database systems

This paper proposes a Tree-structured Database Machine (TDM), which is designed for supporting large relational database systems. The TDM aims at totally improving three level performance bottlenecks in relational database systems, i.e. IO data transfer bottleneck, data processing bottleneck, and multiquery optimization bottleneck. For reducing these bottlenecks, the TDM provides the approaches of data partition and data distribution upon data storages for concurrent data accesses, the tree-structured multiprocessor architecture for parallel data processing, the technique of tightly-coupled-shared-memory for decreasing data movement, the mechanisms of indices and dictionaries for reducing data transfer from the data storages to the memory, and so on. These features show that the TDM not only improves the system performance, increases the parallelism and the data processing power, but also eliminates the performance bottlenecks in relational database systems.

Dado: A Tree-Structured Architecture for Artificial Intelligence Computation

Dado: A Tree-Structured Architecture for Artificial Intelligence Computation

Modular, object-oriented microcomputer architecture

The article proposes a multi-resource, distributed, matrix-structured, reconfigurable, fault-tolerant microcomputer system, particularly suited for efficient multiprocessing and multiprogramming operation. The system implements some ideas of the dynamic and the flexible architectures. The proposed system possess a high degree of regularity by implementing large amounts of identical units. The system implements a new Tree-Structured Architecture (TSA), proposed in this article. The TSA is object-oriented, implements the notions of capability-based-addressing and the single-level-store and it is particularly designed to narrow the semantic gap between HLL's and the machine. It encourages modular programming and directly supports the concept of tasks and intertask communication. The system hardware is modular and modules can be added or taken out, without change in system software.

A Tree Structured Architecture for semantic gap reduction

The article proposes a new Tree-Structured-Architecture (TSA). The TSA is object-oriented, implements the notions of capability-based-addressing and the single-level-store, and it is particularly designed to narrow the semantic gap. It encourages modular programming and directly supports the concepts of tasks and inter-task communication, making it particularly suitable for multiprocessing and multiprogramming implementation.The TSA is implemented on a multi-resource, distributed, matrix-structured, reconfigurable, fault-tolerant computing system, proposed in this article. The system implements some ideas of the dynamic and the flexible architecture. The system possesses a high degree of regularity by implementing large amounts of identical units. The hardware is modular, permitting the addition or removal of groups of units, without a change in the system's software.