基于半监督流形学习的无线传感器网络定位方法及其应用

提出了一种基于流形半监督学习的移动节点定位算法.该算法利用基于流形学习的半监督方法,通过一定量的有标签样本和无标签样本,获取隐含在节点接收信号强度信息中的流形结构,直接建立节点物理位置与接收信号强度之间的映射关系.算法不需要使用现有的理论或经验信号传播模型,避免了模型不准确带来的定位误差,而且允许网络中存在大量无标签样本,降低了数据采集难度,提高了算法实用性.冶金工业现场的实际应用结果表明,相对RADAR算法,本文算法具有较高的定位精度.     

Time dependencies in the occurrences of epileptic seizures

A new method of analysis, developed within the framework of nonlinear dynamics, is applied to patient recorded time series of the occurrence of epileptic seizures. These data exhibit broad band spectra and generally have no obvious structure. The goal is to detect hidden internal dependencies in the data without making any restrictive assumptions, such as linearity, about the structure of the underlying system. The basis of our approach is a conditional probabilistic analysis in a phase space reconstructed from the original data. The data, recorded from patients with intractable epilepsy over a period of 1-3 years, consist of the times of occurrences of hundreds of partial complex seizures. Although the epileptic events appear to occur independently, we show that the epileptic process is not consistent with the rules of a homogeneous Poisson process or generally with a random (IID) process. More specifically, our analysis reveals dependencies of the occurrence of seizures on the occurrence of preceding seizures. These dependencies can be detected in the interseizure interval data sets as well as in the rate of seizures per time period. We modeled patient's inaccuracy in recording seizure events by the addition of uniform white noise and found that the detected dependencies are persistent after addition of noise with standard deviation as great as 1/3 of the standard deviation of the original data set. A linear autoregressive analysis fails to capture these dependencies or produces spurious ones in most of the cases.     

Extracting rules from neural networks by pruning and hidden-unit splitting

An algorithm for extracting rules from a standard three-layer feedforward neural network is proposed. The trained network is first pruned not only to remove redundant connections in the network but, more important, to detect the relevant inputs. The algorithm generates rules from the pruned network by considering only a small number of activation values at the hidden units. If the number of inputs connected to a hidden unit is sufficiently small, then rules that describe how each of its activation values is obtained can be readily generated. Otherwise the hidden unit will be split and treated as output units, with each output unit corresponding to an activation value. A hidden layer is inserted and a new subnetwork is formed, trained, and pruned. This process is repeated until every hidden unit in the network has a relatively small number of input units connected to it. Examples on how the proposed algorithm works are shown using real-world data arising from molecular biology and signal processing. Our results show that for these complex problems, the algorithm can extract reasonably compact rule sets that have high predictive accuracy rates.     

A unifying review of linear gaussian models

Factor analysis, principal component analysis, mixtures of gaussian clusters, vector quantization, Kalman filter models, and hidden Markov models can all be unified as variations of unsupervised learning under a single basic generative model. This is achieved by collecting together disparate observations and derivations made by many previous authors and introducing a new way of linking discrete and continuous state models using a simple nonlinearity. Through the use of other nonlinearities, we show how independent component analysis is also a variation of the same basic generative model. We show that factor analysis and mixtures of gaussians can be implemented in autoencoder neural networks and learned using squared error plus the same regularization term. We introduce a new model for static data, known as sensible principal component analysis, as well as a novel concept of spatially adaptive observation noise. We also review some of the literature involving global and local mixtures of the basic models and provide pseudocode for inference and learning for all the basic models.     

基于人工免疫的RBF神经网络在钢筋性能预报中的应用

提出了一种基于免疫识别原理的径向基函数神经网络学习算法.该算法利用人工免疫系统的识别、记忆、学习等原理,将输入数据作为抗原,抗体为抗原的压缩映射作为径向基函数神经网络模型的隐层中心,输出采用最小二乘法确定权值.通过预报热轧带肋钢筋力学性能的仿真实验结果表明,与K-均值法选择中心点比较,该算法计算量较小,精度高.     

Slow stochastic Hebbian learning of classes of stimuli in a recurrent neural network

We study unsupervised Hebbian learning in a recurrent network in which synapses have a finite number of stable states. Stimuli received by the network are drawn at random at each presentation from a set of classes. Each class is defined as a cluster in stimulus space, centred on the class prototype. The presentation protocol is chosen to mimic the protocols of visual memory experiments in which a set of stimuli is presented repeatedly in a random way. The statistics of the input stream may be stationary, or changing. Each stimulus induces, in a stochastic way, transitions between stable synaptic states. Learning dynamics is studied analytically in the slow learning limit, in which a given stimulus has to be presented many times before it is memorized, i.e. before synaptic modifications enable a pattern of activity correlated with the stimulus to become an attractor of the recurrent network. We show that in this limit the synaptic matrix becomes more correlated with the class prototypes than with any of the instances of the class. We also show that the number of classes that can be learned increases sharply when the coding level decreases, and determine the speeds of learning and forgetting of classes in the case of changes in the statistics of the input stream.     

函数型数据分析与优化极限学习机结合的弹药传输机械臂参数辨识

为实现弹药传输机械臂中不可测参数的辨识,建立了机械臂的虚拟样机,并将其作为样本数据的来源;考虑到样本数据的连续性和平滑特性,使用函数型数据分析和函数型主成分分析对样本数据进行了特征提取,并利用提取的特征参数和待辨识参数作为训练样本对极限学习机(ELM)进行了训练.为提高极限学习机的辨识精度和泛化能力,利用粒子群算法对极限学习机的输入层与隐含层的连接权值和隐含层节点的阈值进行了优化.最后,分别利用仿真数据与测试数据对此方法进行了验证,仿真数据的辨识结果表明,优化后的极限学习机具有更高的辨识精度和泛化能力;同时,通过对比将测试数据的辨识结果代入模型中进行仿真得到的支臂角速度与测试角速度,验证了此方法的可行性和有效性.     

面向材料数据的主动回归学习方法

材料的生产环境和测量条件不同,导致用于机器学习的材料数据的噪声较大.对材料数据进行标注需要一定的专业知识和专业技能,因此标注成本也相对较高.这两方面的因素给机器学习应用于材料领域带来了巨大挑战.为应对这个挑战,提出了一个主动回归学习方法,由离群点检测模块、贪婪采样模块和最小变化采样模块组成.同其他主动学习方法相比,该方法整合了离群点检测机制,选取高质量样本的同时有效地排除了噪声数据的影响,避免了沉没成本.在公开数据集和非公开数据集上与最新的主动回归学习方法进行了对比实验,实验结果表明本文方法在相同的数据量下训练的任务模型性能指标相比于其他模型平均提高15%,且只需30%～40%的数据量作为训练集就可以达到甚至超过使用全部数据训练任务模型的精度.     

A distributed, developmental model of word recognition and naming.

The model described consists of sets of orthographic and phonological units and an interlevel of hidden units. Weights on connections between units were modified during a training phase using the back-propagation learning algorithm. The model simulates many aspects of human performance, including (a) differences between words in terms of processing difficulty, (b) pronunciation of novel items, (c) differences between readers in terms of word recognition skill, (d) transitions from beginning to skilled reading, and (e) differences in performance on lexical decisions and naming tasks. The model's behavior early in the learning phase corresponds to that of children acquiring word recognition skills. Training with a smaller number of hidden units produces output characteristic of many dyslexic readers. Naming is simulated without pronunciation rules, and lexical decisions are simulated without assessing word-level representations. The performance of the model is largely determined by three factors: the nature of the input, a significant fragment of written English; the learning rule, which encodes the implicit structure of the orthography in the weights on connections; and the architecture of the system, which influences the scope of what can be learned. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

An adaptive neural model for mapping invariant target position.

We perceive a constant target in space as constant even though the registration of that target on our senses is continuously shifting. This article derives and stimulates a neural network model that represents visual spot targets, invariant with respect to any combination of egocentric target measures. The model represents space in terms of signals used to move in that space. The model learns and maintains precise sensory-motor calibrations starting with only loosely defined relations. It is adaptive to physical changes of the eye and muscles as well as internal system parameters. Its performance is noise and fault tolerant. Computer simulations show that the average error in target orientation after learning is about 1% of the total visual field extent. The model maintains good accuracy with many different parameter choices. Its performance is most related to the function of the posterior parietal cortex. Testable predictions are made for the columnar topography and learning in that brain structure. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

A fuzzy adaptive learning control network with on-line structure and parameter learning

This paper addresses a general connectionist model, called Fuzzy Adaptive Learning Control Network (FALCON), for the realization of a fuzzy logic control system. An on-line supervised structure/parameter learning algorithm is proposed for constructing the FALCON dynamically. It combines the backpropagation learning scheme for parameter learning and the fuzzy ART algorithm for structure learning. The supervised learning algorithm has some important features. First of all, it partitions the input state space and output control space using irregular fuzzy hyperboxes according to the distribution of training data. In many existing fuzzy or neural fuzzy control systems, the input and output spaces are always partitioned into "grids". As the number of input/output variables increase, the number of partitioned grids will grow combinatorially. To avoid the problem of combinatorial growing of partitioned grids in some complex systems, the proposed learning algorithm partitions the input/output spaces in a flexible way based on the distribution of training data. Second, the proposed learning algorithm can create and train the FALCON in a highly autonomous way. In its initial form, there is no membership function, fuzzy partition, and fuzzy logic rule. They are created and begin to grow as the first training pattern arrives. The users thus need not give it any a priori knowledge or even any initial information on these. In some real-time applications, exact training data may be expensive or even impossible to obtain. To solve this problem, a Reinforcement Fuzzy Adaptive Learning Control Network (RFALCON) is further proposed. The proposed RFALCON is constructed by integrating two FALCONs, one FALCON as a critic network, and the other as an action network. By combining temporal difference techniques, stochastic exploration, and a proposed on-line supervised structure/parameter learning algorithm, a reinforcement structure/parameter learning algorithm is proposed, which can construct a RFALCON dynamically through a reward/penalty signal. The ball and beam balancing system is presented to illustrate the performance and applicability of the proposed models and learning algorithms.     

Neural Modeling of Square Surface Aerators

Applications of artificial neural networks in the field of aeration phenomena in surface aerators, which are not geometrically similar, are explored to predict reaeration rates under varying dynamic as well as geometric conditions. The primary network for prediction is a feed forward network with nonlinear elements. The network consists of an input layer, an output layer, a hidden layer, and the nonlinear transfer function in each processing element. The network requires supervised learning and the learning algorithm is the back-propagation. As back-propagation learning is affected by local minima, and to get over this aspect various other modifications have been suggested like Levenberg-Marquardt, quasi-Newton, conjugate-gradient, etc. The present study suggests that the Levenberg-Marquardt modification is a very efficient algorithm in comparison with others like quasi-Newton and conjugate-gradient. In the situations when the dimension of the input vector is large, and highly correlated, it is useful to reduce the dimension of the input vectors. An effective procedure for performing this operation is principal component analysis. The best prediction performance is achieved when the data are preprocessed using principal components analysis before they are fed to a back-propagated neural network, but at the cost of losing the physical significance of experimental data. The model thus developed can be used to predict the reaeration rate for different sizes of geometric elements (like rotor diameter, sizes of rotor, aerators’ geometry, water depth, etc.) under various dynamic conditions, i.e., the speed of the rotor.     

Estimating Evapotranspiration using Artificial Neural Network

This study investigates the utility of artificial neural networks (ANNs) for estimation of daily grass reference crop evapotranspiration (ETo) and compares the performance of ANNs with the conventional method (Penman–Monteith) used to estimate ETo. Several issues associated with the use of ANNs are examined, including different learning methods, number of processing elements in the hidden layer(s), and the number of hidden layers. Three learning methods, namely, the standard back-propagation with learning rates of 0.2 and 0.8, and backpropagation with momentum were considered. The best ANN architecture for estimation of daily ETo was obtained for two different data sets (Sets 1 and 2) for Davis, Calif. Using data of Set 1, the networks were trained with daily climatic data (solar radiation, maximum and minimum temperature, maximum and minimum relative humidity, and wind speed) as input and the Penman–Monteith (PM) estimated ETo as output. The best ANN architecture was selected on the basis of weighted standard error of estimate (WSEE) and minimal ANN architecture. The ANN architecture of 6-7-1, (six, seven, and one neuron(s) in the input, hidden, and output layers, respectively) gave the minimum WSEE (less than 0.3 mm/day) for all learning methods. This value was lower than the WSEE (0.74 mm/day) between the PM method and lysimeter measured ETo as reported by Jensen et al. in 1990. Similarly, ANNs were trained, validated, and tested using the lysimeter measured ETo and corresponding climatic data (Set 2). Again, all learning methods gave less WSEE (less than 0.60 mm/day) as compared to the PM method (0.97 mm/day). Based on these results, it can be concluded that the ANN can predict ETo better than the conventional method (PM) for Davis.     

Top-down learning of low-level vision tasks

Perceptual tasks such as edge detection, image segmentation, lightness computation and estimation of three-dimensional structure are considered to be low-level or mid-level vision problems and are traditionally approached in a bottom-up, generic and hard-wired way. An alternative to this would be to take a top-down, object-class-specific and example-based approach. In this paper, we present a simple computational model implementing the latter approach. The results generated by our model when tested on edge-detection and view-prediction tasks for three-dimensional objects are consistent with human perceptual expectations. The model's performance is highly tolerant to the problems of sensor noise and incomplete input image information. Results obtained with conventional bottom-up strategies show much less immunity to these problems. We interpret the encouraging performance of our computational model as evidence in support of the hypothesis that the human visual system may learn to perform supposedly low-level perceptual tasks in a top-down fashion.     

一种新型BP神经网络的生成算法

提出了一种新型的神经网络的训练方法，包括输入层节点的自动寻找技术和隐含层节点的自动生成技术．为了验证该算法的正确性，利用该算法对同向回转观机驱动振动系统进行了神经网络逼近分析，计算机模拟结果表明该算法可以准确地确定输入层和隐含层节点的数目．     

Bayesian Modal Updating using Complete Input and Incomplete Response Noisy Measurements

The problem of identification of the modal parameters of a structural model using complete input and incomplete response time histories is addressed. It is assumed that there exist both input error (due to input measurement noise) and output error (due to output measurement noise and modeling error). These errors are modeled by independent white noise processes, and contribute towards uncertainty in the identification of the modal parameters of the model. To explicitly treat these uncertainties, a Bayesian framework is adopted and a Bayesian time-domain methodology for modal updating based on an approximate conditional probability expansion is presented. The methodology allows one to obtain not only the optimal (most probable) values of the updated modal parameters but also their uncertainties, calculated from their joint probability distribution. Calculation of the uncertainties of the identified modal parameters is very important if one plans to proceed with the updating of a theoretical finite-element model based on these modal estimates. The proposed approach requires only one set of excitation and corresponding response data. It is found that the updated probability density function (PDF) can be well approximated by a Gaussian distribution centered at the optimal parameters at which the posterior PDF is maximized. Numerical examples using noisy simulated data are presented to illustrate the proposed method.     

Induction of decision trees and Bayesian classification applied to diagnosis of sport injuries

Machine learning techniques can be used to extract knowledge from data stored in medical databases. In our application, various machine learning algorithms were used to extract diagnostic knowledge which may be used to support the diagnosis of sport injuries. The applied methods include variants of the Assistant algorithm for top-down induction of decision trees, and variants of the Bayesian classifier. The available dataset was insufficient for reliable diagnosis of all sport injuries considered by the system. Consequently, expert-defined diagnostic rules were added and used as pre-classifiers or as generators of additional training instances for diagnoses for which only few training examples were available. Experimental results show that the classification accuracy and the explanation capability of the naive Bayesian classifier with the fuzzy discretization of numerical attributes were superior to other methods and estimated as the most appropriate for practical use.     

Cloud Neural Fuzzy PID Hybrid Integrated Algorithm of Flatness Control

In connection with the characteristics of multi-disturbance and nonlinearity of a system for flatness control in cold rolling process, a new intelligent PID control algorithm was proposed based on a cloud model, neural network and fuzzy integration. By indeterminacy artificial intelligence, the problem of fixing the membership functions of input variables and fuzzy rules was solved in an actual fuzzy system and the nonlinear mapping between variables was implemented by neural network. The algorithm has the adaptive learning ability of neural network and the indetermi- nacy of a cloud model in processing knowledge, which makes the fuzzy system have more persuasion in the process of knowledge inference, realizing the online adaptive regulation of PID parameters and avoiding the defects of the traditional PID controller. Simulation results show that the algorithm is simple, fast and robust with good control performance and application value.     

Nonequilibrium response spectroscopy of voltage-sensitive ion channel gating

We describe a new electrophysiological technique called nonequilibrium response spectroscopy, which involves application of rapidly fluctuating (as high as 14 kHz) large-amplitude voltage clamp waveforms to ion channels. As a consequence of the irreversible (in the sense of Carnot) exchange of energy between the fluctuating field and the channel protein, the gating response is exquisitely sensitive to features of the kinetics that are difficult or impossible to adequately resolve by means of traditional stepped potential protocols. Here we focus on the application of dichotomous (telegraph) noise voltage fluctuations, a broadband Markovian colored noise that fluctuates between two values. Because Markov kinetic models of channel gating can be embedded within higher-dimensional Markov models that take into account the effects of the voltage fluctuations, many features of the response of the channels can be calculated algebraically. This makes dichotomous noise and its generalizations uniquely suitable for model selection and kinetic analysis. Although we describe its application to macroscopic ionic current measurements, the nonequilibrium response method can also be applied to gating and single channel current recording techniques. We show how data from the human cardiac isoform (hH1a) of the Na+ channel expressed in mammalian cells can be acquired and analyzed, and how these data reveal hidden aspects of the molecular kinetics that are not revealed by conventional methods.