检索结果-内蒙古大学图书馆

作者： Craven, Michael P. University of Nottingham

学位级别：博士

As advances have been made in semiconductor processing technology, the number of transistors on a chip has increased out of step with the number of input/output pins, which has introduced a communications ’bottle-neck’ in the design of computer architectures. This is a major issue in the hardware design of parallel structures implemented in either digital or analogue VLSI, and is particularly relevant to the design of neural networks which need to be highly interconnected. This work reviews hardware implementations of neural networks, with an emphasis on analogue implementations, and proposes a new method for overcoming connectivity constraints, by the use of Frequency Division Multiplexing (FDM) for the inter-chip communications. In this FDM scheme, multiple analogue signals are transmitted between chips on a single wire by modulating them at different frequencies. The main theoretical work examines the number of signals which can be packed into an FDM channel, depending on the quality factors of the filters used for the demultiplexing, and a fractional overlap parameter which was defined to take into account the inevitable overlapping of filter frequency responses. It is seen that by increasing the amount of permissible overlap, it is possible to communicate a larger number of signals in a given bandwidth. Alternatively, the quality factors of the filters can be reduced, which is advantageous for hardware implementation. Therefore, it was found necessary to determine the amount of overlap which might be permissible in a neural network implementation utilising FDM communications. A software simulator is described, which was designed to test the effects of overlap on Multilayer Perceptron neural networks. Results are presented for networks trained with the backpropagation algorithm, and with the alternative weight perturbation algorithm. These were carried out using both floating point and quantised weights to examine the combined effects of overlap and weight q

关键词： neural network learning algorithm multiplexing backpropagation weight perturbation simulation filter oscillator VLSI ANN FDM text-to-speech

来源：评论

学校读者我要写书评

暂无评论

RADIAL BASIS FUNCTION NETWORKS FOR POWER-SYSTEM DYNAMIC LOAD MODELING

RADIAL BASIS FUNCTION NETWORKS FOR POWER-SYSTEM DYNAMIC LOAD...

引用

TENCON 93 - 1993 IEEE Region 10 Conference on Computer, Communication, Control and Power Engineering

作者： CHEN, HL SHEN, SD ZHU, SZ TSING HUA UNIV DEPT ELECT ENGNBEIJING 100084PEOPLES R CHINA

ISBN: (纸本)780003285X

The importance of electric load models in power system transient stability studies has long been recognized. In this paper the radial basis function network (RBFN) is presented for dynamic load modeling. The learning algorithm for RBFN is based on first choosing the RBF centers using the K-means clustering method and then using singular value decomposition to obtain the parameters. Its fast training procedure and high precision makes it more appropriate for power system dynamic load modeling. The simulation results of the field and laboratory tests demonstrate that the application of the RBFN is promising.

关键词： power system analysis computing load (electric) feedforward neural nets digital simulation learning (artificial intelligence) parameter estimation power system dynamic load modeling radial basis function networks power system transient stability learning algorithm radial basis function centers K-means clustering method singular value decomposition parameters estimation training computer simulation

来源：评论

学校读者我要写书评

暂无评论

The Finite Element Neural Network Application to Plant Factory

引用

IFAC Proceedings Volumes 1993年第2期26卷 1155-1158页

作者： H. Murase S. Koyama Department of Agricultural Engineering University of Osaka Prefecture 1-1 Gakuen Sakai Osaka 593 Japan

A spatial neural network (FENN) has been developed. The FENN is structured in a physical space unlike other conventional artificial neural networks of which structure represent only mathematical connections of neural elements. The FENN structure is formed by the finite clement used in the finite element method. The governing differential equation (Poisson's equation) was utilized for neural information processing. Kalman filter was used as a learning algorithm of the FENN. In this paper, a practical aspect of the FENN in the field of plant factory is discussed in relation to the sensor fusion technique for environmental control

关键词： Neural nets Finite element method learning algorithm Kalman filters Sensor fusion

来源：评论

学校读者我要写书评

暂无评论

Performance of an Asynchronous Computing Machine for Supervised learning

Performance of an Asynchronous Computing Machine for Supervi...

引用

作者： Kim, Sung Tae Drexel University

学位级别：M.S., Master of Science/Master of Surgery

Theories and applications of artificial neural classifiers are developed, increasing the demand for efficient implementation schemes. In this thesis, reprogrammable dataflow neural classifiers are proposed as an alternative to traditional implementations. In general, these classifiers are based on functional languages, neural-dataflow transformations, dataflow algorithmic transformations and dataflow multiprocessors. An experimental approach is used to investigate the performance of a large-scale, fine-grained dataflow classifier architecture. In this study, the functional descriptions of high-level data dependency of a supervised learning algorithm are transformed into a machine-executable low-level dataflow graph. The tagged token dataflow algorithmic transformation is applied to exploit the parallelism. Dataflow neural classifiers are used to implement the learning algorithm. No attempt is made to optimize the granularity of the high-level language programming blocks to balance the computation and communication. The proposed classifier architecture is more versatile than other existing architectures. Performance results show the effectiveness of dataflow neural classifiers.

关键词： Computing Machines Neural-dataflow learning algorithm Parallelism algorithmic transformation

来源：评论

学校读者我要写书评

暂无评论

A STOCHASTIC REINFORCEMENT learning algorithm FOR learning REAL-VALUED FUNCTIONS

引用

NEURAL NETWORKS 1990年第6期3卷 671-692页

作者： GULLAPALLI, V Department of Computer and Information Science University of Massachusetts USA

Most of the research in reinforcement learning has been on problems with discrete action spaces. However, many control problems require the application of continuous control signals. In this paper, we present a stochastic reinforcement learning algorithm for learning functions with continuous outputs using a connectionist network. We define stochastic units that compute their real-valued outputs as a function of random activations generated using the normal distribution. learning takes place by using our algorithm to adjust the two parameters of the normal distribution so as to increase the probability of producing the optimal real value for each input pattern. The performance of the algorithm is studied by using it to learn tasks of varying levels of difficulty. Further, as an example of a potential application, we present a network incorporating these stochastic real-valued units that learns to perform an underconstrained positioning task using a simulated 3 degree-of-freedom robot arm.

关键词： Neural networks Associative reinforcement learning learning algorithm Stochastic automata Real-valued functions Shaping Robotics Neurocontrol

来源：评论

学校读者我要写书评

暂无评论

Neural networks and computing

引用

Future Generation Computer Systems 1991年第1期7卷 69-77页

作者： Fogelman Soulié, Françoise Mimetics 92 298 Chatenay Malabry 5 Centrale Parc avenue Sully Prudhomme France

In this paper, we give a general presentation of neural networks, showing their links and differences with Artificial Intelligence and neurosciences. We provide the general formalism of neural networks and describe two neural networks learning algorithms: gradient backpropagation and learning vector quantization. We then illustrate the behavior of these algorithms on a particular application in speech processing. We then discuss what a neural networks programming environment should provide and present some of the existing commercial products. We end the paper by the presentation of MimeNice, which is an example of such an environment developed within the Esprit Pygmalion-Galatea projects. © 1991.

关键词： learning algorithm Neural networks neural networks programming environment neurocomputer neurocomputing

来源：评论

学校读者我要写书评

暂无评论

EXPLORATIONS OF THE MEAN FIELD-THEORY learning algorithm

引用

NEURAL NETWORKS 1989年第6期2卷 475-494页

作者： PETERSON, C HARTMAN, E MICROELECTR & COMP TECHNOL CORP AUSTINTX

The mean field theory (MFT) learning algorithm is elaborated and explored with respect to a variety of tasks. MFT is benchmarked against the back-propagation learning algorithm (BP) on two different feature recognition problems: two-dimensional mirror symmetry and multidimensional statistical pattern classification. We find that while the two algorithms are very similar with respect to generalization properties, MFT normally requires a substantially smaller number of training epochs than BP. Since the MFT model is bidirectional, rather than feed-forward, its use can be extended naturally from purely functional mappings to a content addressable memory. A network with N visible and N hidden units can store up to approximately 4 N patterns with good content-addressability. We stress an implementational advantage for MFT: it is natural for VLSI circuitry.

关键词： Neural network Bidirectional Generalization Content addressable memory Mean field theory learning algorithm

来源：评论

学校读者我要写书评

暂无评论

learning BY PARALLEL FORWARD PROPAGATION

LEARNING BY PARALLEL FORWARD PROPAGATION

引用

INTERNATIONAL JOINT CONF ON NEURAL NETWORKS ( IJCNM-90 )

作者： ABE, S Hitachi Ltd Japan

The back-propagation algorithm is widely used for learning weights of multilayered neural networks. The major drawbacks however, are the slow convergence and lack of a proper way to set the number of hidden neurons. The author proposes a learning algorithm which solves the above problems. The weights between two layers are successively calculated, fixing other weights so that the error function, which is the square sum of the difference between the training data and the network outputs is minimised. Since the calculation results in solving a set of linearized equations, redundancy of the hidden neurons is judged by the singularity of the corresponding coefficient matrix. For the exclusive-OR and parity check circuits, excellence convergence characteristics are obtained, and the redundancy of the hidden neurons is checked by the singularity of the matrix

关键词： parallel forward propagation weights multilayered neural networks learning algorithm error function training data linearized equations redundancy hidden neurons singularity coefficient matrix exclusive-OR parity check circuits convergence characteristics learning systems neural nets parallel algorithms parallel architectures

来源：评论

学校读者我要写书评

暂无评论

NEURAL NETWORK DATA FUSION CONCEPTS AND APPLICATION

NEURAL NETWORK DATA FUSION CONCEPTS AND APPLICATION

引用

INTERNATIONAL JOINT CONF ON NEURAL NETWORKS ( IJCNM-90 )

作者： EGGERS, M KHUON, T MIT Lincoln Lab Lexington MA USA

A neural network data fusion decision system and the accompanying experimental results obtained from joint sensor data are presented. The decisions include detection and correct classification of space object maneuvers simultaneously observed by two radars of different aspect, frequency, and resolution. The system consists of a statistically-based adaptive preprocessor for each sensor, followed by a highly parallel neural network for associating the preprocessor outputs with the appropriate decisions. The preprocessing approach, supported by a signal decomposition theorem, recursively models the detrended sensor data as an autoregressive process of sufficiently high order. This approach also accommodates nonstationary data by incorporating an information-theoretic transition detector which identifies the segments of near-stationary data. Together, feature vectors are produced over near-stationary segments of data which are scale invariant, translation invariant, and normalized and represent sufficient statistics. Subsequently, the feature vectors arising from the sensor preprocessors are collectively associated with the correct output decision. The association is conducted by a multilayer perceptron neural network associative memory employing a modified learning algorithm which converges at a rate comparable to that of conventional algorithms, yet requires less computation.

关键词： detection decisions classification decisions scale invariance translation invariance normalization convergence rate neural network data fusion decision system space object maneuvers radars statistically-based adaptive preprocessor signal decomposition detrended sensor data autoregressive process nonstationary data information-theoretic transition detector feature vectors near-stationary segments multilayer perceptron neural network associative memory learning algorithm aerospace computing classification content-addressable storage data acquisition electric sensing devices information theory learning systems neural nets radar applications statistics

来源：评论

学校读者我要写书评

暂无评论

A learning algorithm FOR BOLTZMANN MACHINES

引用

COGNITIVE SCIENCE 1985年第1期9卷 147-169页

作者： ACKLEY, DH HINTON, GE SEJNOWSKI, TJ JOHNS HOPKINS UNIV DEPT BIOPHYSBALTIMOREMD 21218

The computotionol power of massively parallel networks of simple processing elements resides in the communication bandwidth provided by the hardware connections between elements. These connections con allow a significant fraction of the knowledge of the system to be applied to an instance of a prob- lem in o very short time. One kind of computation for which massively porollel networks appear to be well suited is large constraint satisfaction searches, but to use the connections efficiently two conditions must be met: First, a search technique that is suitable for parallel networks must be found. Second, there must be some way of choosing internal representations which allow the preexisting hardware connections to be used efficiently for encoding the con- straints in the domain being searched. We describe a generol parallel search method, based on statistical mechanics, and we show how it leads to a gen- eral learning rule for modifying the connection strengths so as to incorporate knowledge obout o task domain in on efficient way. We describe some simple examples in which the learning algorithm creates internal representations thot ore demonstrobly the most efficient way of using the preexisting connec- tivity structure.

关键词： .describe examples allow representations Connections suitable strengths efficiently connectionist hardware learning algorithm satisfaction Carnegie-Mellon

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：