Home About us Contact
|
Home About us Contact | ||
|
|
||
Learning Rule (learning + rule)
Selected AbstractsExpedient and Monotone Learning RulesECONOMETRICA, Issue 2 2004Tilman Börgers This paper considers learning rules for environments in which little prior and feedback information is available to the decision maker. Two properties of such learning rules are studied: absolute expediency and monotonicity. Both require that some aspect of the decision maker's performance improves from the current period to the next. The paper provides some necessary, and some sufficient conditions for these properties. It turns out that there is a large variety of learning rules that have the properties. However, all learning rules that have these properties are related to the replicator dynamics of evolutionary game theory. For the case in which there are only two actions, it is shown that one of the absolutely expedient learning rules dominates all others. [source] The application of NN technique to automatic generation control for the power system with three areas including smes unitsEUROPEAN TRANSACTIONS ON ELECTRICAL POWER, Issue 4 2003A. Demirören The study includes an application of layered neural network controller to study automatic generation control (AGC) problem of the power system, which contains superconducting magnetic energy storage (SMES) units. The effectiveness of SMES unit over frequency oscillations improvement against load perturbations in power system is well known. In addition, the proposed control scheme provides the steady state error of frequency and inadvertent interchange of tie-lines to be maintained in steady state values. The power system considered has three areas two of which including steam turbines while the other containing a hydro turbine, and all of them contain SMES units, in addition. In the power system each area with a steam turbine contains the non-linearity due to reheat effect of the steam turbine and all of the areas contain upper and lower constraints for generation rate. Only one neural network (NN) controller, which controls all the inputs of each area in the power system, is considered. In the NN controller, back propagation-through-time algorithm is used as neural network learning rule. The performance of the power system is simulated by using conventional integral controller and NN controller for the cases with or without SMES units in all areas, separately. By comparing the results for both cases, it can be seen that the performance of NN controller is better than conventional controllers. [source] An oscillatory interference model of grid cell firingHIPPOCAMPUS, Issue 9 2007Neil Burgess Abstract We expand upon our proposal that the oscillatory interference mechanism proposed for the phase precession effect in place cells underlies the grid-like firing pattern of dorsomedial entorhinal grid cells (O'Keefe and Burgess (2005) Hippocampus 15:853,866). The original one-dimensional interference model is generalized to an appropriate two-dimensional mechanism. Specifically, dendritic subunits of layer II medial entorhinal stellate cells provide multiple linear interference patterns along different directions, with their product determining the firing of the cell. Connection of appropriate speed- and direction- dependent inputs onto dendritic subunits could result from an unsupervised learning rule which maximizes postsynaptic firing (e.g. competitive learning). These inputs cause the intrinsic oscillation of subunit membrane potential to increase above theta frequency by an amount proportional to the animal's speed of running in the "preferred" direction. The phase difference between this oscillation and a somatic input at theta-frequency essentially integrates velocity so that the interference of the two oscillations reflects distance traveled in the preferred direction. The overall grid pattern is maintained in environmental location by phase reset of the grid cell by place cells receiving sensory input from the environment, and environmental boundaries in particular. We also outline possible variations on the basic model, including the generation of grid-like firing via the interaction of multiple cells rather than via multiple dendritic subunits. Predictions of the interference model are given for the frequency composition of EEG power spectra and temporal autocorrelograms of grid cell firing as functions of the speed and direction of running and the novelty of the environment. © 2007 Wiley-Liss, Inc. [source] A learning rule for place fields in a cortical model: Theta phase precession as a network effectHIPPOCAMPUS, Issue 7 2005Silvia Scarpetta Abstract We show that a model of the hippocampus introduced recently by Scarpetta et al. (2002, Neural Computation 14(10):2371,2396) explains the theta phase precession phenomena. In our model, the theta phase precession comes out as a consequence of the associative-memory-like network dynamics, i.e., the network's ability to imprint and recall oscillatory patterns, coded both by phases and amplitudes of oscillation. The learning rule used to imprint the oscillatory states is a natural generalization of that used for static patterns in the Hopfield model, and is based on the spike-time-dependent synaptic plasticity, experimentally observed. In agreement with experimental findings, the place cells' activity appears at consistently earlier phases of subsequent cycles of the ongoing theta rhythm during a pass through the place field, while the oscillation amplitude of the place cells' firing rate increases as the animal approaches the center of the place field and decreases as the animal leaves the center. The total phase precession of the place cell is lower than 360°, in agreement with experiments. As the animal enters a receptive field, the place cells' activity comes slightly less than 180° after the phase of maximal pyramidal cell population activity, in agreement with the findings of Skaggs et al. (1996, Hippocampus 6:149,172). Our model predicts that the theta phase is much better correlated with location than with time spent in the receptive field. Finally, in agreement with the recent experimental findings of Zugaro et al. (2005, Nature Neuroscience 9(1):67,71), our model predicts that theta phase precession persists after transient intrahippocampal perturbation. © 2005 Wiley-Liss, Inc. [source] Competitive Hebbian learning and the hippocampal place cell system: Modeling the interaction of visual and path integration cuesHIPPOCAMPUS, Issue 3 2001Alex Guazzelli Abstract The hippocampus has long been thought essential for implementing a cognitive map of the environment. However, almost 30 years since place cells were found in rodent hippocampal field CA1, it is still unclear how such an allocentric representation arises from an egocentrically perceived world. By means of a competitive Hebbian learning rule responsible for coding visual and path integration cues, our model is able to explain the diversity of place cell responses observed in a large set of electrophysiological experiments with a single fixed set of parameters. Experiments included changes observed in place fields due to exploration of a new environment, darkness, retrosplenial cortex inactivation, and removal, rotation, and permutation of landmarks. To code for visual cues for each landmark, we defined two perceptual schemas representing landmark bearing and distance information over a linear array of cells. The information conveyed by the perceptual schemas is further processed through a network of adaptive layers which ultimately modulate the resulting activity of our simulated place cells. In path integration terms, our system is able to dynamically remap a bump of activity coding for the displacement of the animal in relation to an environmental anchor. We hypothesize that path integration information is computed in the rodent posterior parietal cortex and conveyed to the hippocampus where, together with visual information, it modulates place cell activity. The resulting network yields a more direct treatment of partial remapping of place fields than other models. In so doing, it makes new predictions regarding the nature of the interaction between visual and path integration cues during new learning and when the system is challenged with environmental changes. Hippocampus 2001;11:216,239. © 2001 Wiley-Liss, Inc. [source] Monetary Policy, Price Stability and Output Gap StabilizationINTERNATIONAL FINANCE, Issue 2 2002Vitor Gaspar Using a standard New,Keynesian model, this paper examines three reasons why monetary policy should primarily focus on price stability rather than the stabilization of output around potential, even if there appears to be an exploitable trade,off between the volatility of inflation and that of the output gap. First, we discuss the well,known time,inconsistency problem associated with active output gap stabilization. Increasing the relative weight on inflation stabilization improves the equilibrium outcome. Second, we analyse some of the problems associated with the substantial uncertainty that surrounds estimates of potential output. We argue that focusing on price stability is a robust monetary policy strategy in the face of such uncertainty. Finally, we consider the case where private agents are trying to estimate the inflation generating process using an ,ad hoc', but reasonable learning rule. By emphasizing a single goal the central bank facilitates the process of learning, thereby stablizing both inflation and the output gap. [source] Stochastic stability of a neural-net robot controller subject to signal-dependent noise in the learning ruleINTERNATIONAL JOURNAL OF ADAPTIVE CONTROL AND SIGNAL PROCESSING, Issue 6 2010Abraham K. Ishihara Abstract We consider a neural network-based controller for a rigid serial link manipulator with uncertain plant parameters. We assume that the training signal to the network is corrupted by signal-dependent noise. A radial basis function network is utilized in the feedforward control to approximate the unknown inverse dynamics. The weights are adaptively adjusted according to a gradient descent plus a regulation term (Narendra's e -modification). We prove a theorem that extends the Yoshizawa D-boundedness results to the stochastic setting. As in the deterministic setting, this result is particularly useful for neural network robot control when there exists bounded torque disturbances and neural net approximation errors over a known compact set. Using this result, we establish bounds on the feedback gains and learning rate parameters that guarantee the origin of the closed-loop system is semi-globally, uniformly bounded in expected value. Copyright © 2009 John Wiley & Sons, Ltd. [source] NEURAL NETWORK MODELING OF END-OVER-END THERMAL PROCESSING OF PARTICULATES IN VISCOUS FLUIDSJOURNAL OF FOOD PROCESS ENGINEERING, Issue 2010YANG MENG ABSTRACT Modeling of the heat transfer process in thermal processing is important for the process design and control. Artificial neural networks (ANNs) have been used in recent years in heat transfer modeling as a potential alternative to conventional dimensionless correlation approach and shown to be even better performers. In this study, ANN models were developed for apparent heat transfer coefficients associated with canned particulates in high viscous Newtonian and non-Newtonian fluids during end-over-end thermal processing in a pilot-scale rotary retort. A portion of experimental data obtained for the associated heat transfer coefficients were used for training while the rest were used for testing. The principal configuration parameters were the combination of learning rules and transfer functions, number of hidden layers, number of neurons in each hidden layer and number of learning runs. For the Newtonian fluids, the optimal conditions were two hidden layers, five neurons in each hidden layer, the delta learning rule, a sine transfer function and 40,000 learning runs, while for the non-Newtonian fluids, the optimal conditions were one hidden layer, six neurons in each hidden layer, the delta learning rule, a hyperbolic tangent transfer function and 50,000 learning runs. The prediction accuracies for the ANN models were much better compared with those from the dimensionless correlations. The trained network was found to predict responses with a mean relative error of 2.9,3.9% for the Newtonian fluids and 4.7,5.9% for the non-Newtonian fluids, which were 27,62% lower than those associated with the dimensionless correlations. Algebraic solutions were included, which could be used to predict the heat transfer coefficients without requiring an ANN. PRACTICAL APPLICATIONS The artificial neural network (ANN) model is a network of computational elements that was originally developed to mimic the function of the human brain. ANN models do not require the prior knowledge of the relationship between the input and output variables because they can discover the relationship through successive training. Moreover, ANN models can predict several output variables at the same time, which is difficult in general regression methods. ANN concepts have been successfully used in food processing for prediction, quality control and pattern recognition. ANN models have been used in recent years for heat transfer modeling as a potential alternative to conventional dimensionless correlation approach and shown to be even better performers. In this study, ANN models were successfully developed for the heat transfer parameters associated with canned particulate high viscous Newtonian and non-Newtonian fluids during an end-over-end rotation thermal processing. Optimized configuration parameters were obtained by choosing appropriate combinations of learning rule, transfer function, learning runs, hidden layers and number of neurons. The trained network was found to predict parameter responses with mean relative errors considerably lower than from dimensionless correlations. [source] Scaling Up Learning Models in Public Good GamesJOURNAL OF PUBLIC ECONOMIC THEORY, Issue 2 2004Jasmina Arifovic We study three learning rules (reinforcement learning (RL), experience weighted attraction learning (EWA), and individual evolutionary learning (IEL)) and how they perform in three different Groves,Ledyard mechanisms. We are interested in how well these learning rules duplicate human behavior in repeated games with a continuum of strategies. We find that RL does not do well, IEL does significantly better, as does EWA, but only if given a small discretized strategy space. We identify four main features a learning rule should have in order to stack up against humans in a minimal competency test: (1) the use of hypotheticals to create history, (2) the ability to focus only on what is important, (3) the ability to forget history when it is no longer important, and (4) the ability to try new things. [source] ARTIFICIAL NEURAL NETWORK MODELING FOR REFORESTATION DESIGN THROUGH THE DOMINANT TREES BOLE-VOLUME ESTIMATIONNATURAL RESOURCE MODELING, Issue 4 2009MARIA J. DIAMANTOPOULOU Abstract In the management of restoration reforestations or recreational reforestations of trees, the density of the planted trees and the site conditions can influence the growth and bole volume of the dominant tree. The ability to influence growth of these trees in a reforestation contributes greatly to the formation of large dimension trees and thereby to the production of commercially valuable wood. The potential of two artificial neural network (ANN) architectures in modeling the dominant,Pinus brutia,tree bole volume in reforestation configuration at 12 years of age was investigated: (1) the multilayer perceptron architecture using a back-propagation algorithm and (2) the cascade-correlation architecture, utilizing (a) either the nonlinear Kalman's filter theory or (b) the adaptive gradient descent learning rule. The incentive for developing bole-volume equations using ANN techniques was to demonstrate an alternative new methodology in the field of reforestation design, which would enable estimation and optimization of the bole volume of dominant trees in reforestations using easily measurable site and competition factors. The usage of the ANNs for the estimation of dominant tree bole volume through site and competition factors can be a very useful tool in forest management practice. [source] Identification based adaptive iterative learning controllerASIAN JOURNAL OF CONTROL, Issue 5 2010Suhail Ashraf Abstract In recent years, more research in the control field has been in the area of self-learning and adaptable systems, such as a robot that can teach itself to improve its performance. One of the more promising algorithms for self-learning control systems is Iterative Learning Control (ILC), which is an algorithm capable of tracking a desired trajectory within a specified error limit. Conventional ILC algorithms have the problem of relatively slow convergence rate and adaptability. This paper suggests a novel approach by combining system identification techniques with the proposed ILC approach to overcome the aforementioned problems. The ensuing design procedure is explained and results are accrued from a number of simulation examples. A key point in the proposed scheme is the computation of gain matrices using the steepest descent approach. It has been found that the learning rule can be guaranteed to converge if certain conditions are satisfied. Copyright © 2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society [source] Autoassociator networks: insights into infant cognitionDEVELOPMENTAL SCIENCE, Issue 2 2004Sylvain Sirois This paper presents autoassociator neural networks. A first section reviews the architecture of these models, common learning rules, and presents sample simulations to illustrate their abilities. In a second section, the ability of these models to account for learning phenomena such as habituation is reviewed. The contribution of these networks to discussions about infant cognition is highlighted. A new, modular approach is presented in a third section. In the discussion, a role for these learning models in a broader developmental framework is proposed. [source] Expedient and Monotone Learning RulesECONOMETRICA, Issue 2 2004Tilman Börgers This paper considers learning rules for environments in which little prior and feedback information is available to the decision maker. Two properties of such learning rules are studied: absolute expediency and monotonicity. Both require that some aspect of the decision maker's performance improves from the current period to the next. The paper provides some necessary, and some sufficient conditions for these properties. It turns out that there is a large variety of learning rules that have the properties. However, all learning rules that have these properties are related to the replicator dynamics of evolutionary game theory. For the case in which there are only two actions, it is shown that one of the absolutely expedient learning rules dominates all others. [source] The E-Correspondence PrincipleECONOMICA, Issue 293 2007GEORGE W. EVANS We present a new application of Samuelson's Correspondence Principle to the analysis of comparative dynamics in stochastic rational expectations models. Our version, which we call the E-correspondence principle, applies to rational expectations equilibria that are stable under least squares and closely related learning rules. With this technique it is sometimes possible to study, without explicitly solving for the equilibrium, how qualitative properties of the equilibrium are affected by changes in the model parameters. Applications to overlapping generations and New Keynesian models illustrate the potential of the technique. [source] Scaling Up Learning Models in Public Good GamesJOURNAL OF PUBLIC ECONOMIC THEORY, Issue 2 2004Jasmina Arifovic We study three learning rules (reinforcement learning (RL), experience weighted attraction learning (EWA), and individual evolutionary learning (IEL)) and how they perform in three different Groves,Ledyard mechanisms. We are interested in how well these learning rules duplicate human behavior in repeated games with a continuum of strategies. We find that RL does not do well, IEL does significantly better, as does EWA, but only if given a small discretized strategy space. We identify four main features a learning rule should have in order to stack up against humans in a minimal competency test: (1) the use of hypotheticals to create history, (2) the ability to focus only on what is important, (3) the ability to forget history when it is no longer important, and (4) the ability to try new things. [source] SELF-ORGANIZING PEER-TO-PEER SOCIAL NETWORKSCOMPUTATIONAL INTELLIGENCE, Issue 3 2008Fang Wang Peer-to-peer (P2P) systems provide a new solution to distributed information and resource sharing because of its outstanding properties in decentralization, dynamics, flexibility, autonomy, and cooperation, summarized as DDFAC in this paper. After a detailed analysis of the current P2P literature, this paper suggests to better exploit peer social relationships and peer autonomy to achieve efficient P2P structure design. Accordingly, this paper proposes Self-organizing peer-to-peer social networks (SoPPSoNs) to self-organize distributed peers in a decentralized way, in which neuron-like agents following extended Hebbian rules found in the brain activity represent peers to discover useful peer connections. The self-organized networks capture social associations of peers in resource sharing, and hence are called P2P social networks. SoPPSoNs have improved search speed and success rate as peer social networks are correctly formed. This has been verified through tests on real data collected from the Gnutella system. Analysis on the Gnutella data has verified that social associations of peers in reality are directed, asymmetric and weighted, validating the design of SoPPSoN. The tests presented in this paper have also evaluated the scalability of SoPPSoN, its performance under varied initial network connectivity and the effects of different learning rules. [source] | ||||||||||||||||