Generic model Research Papers - Academia.edu (original) (raw)

In this paper, we introduce a general modeling technique, called vector-quantized temporal associative memory (VQTAM), which uses Kohonen's self-organizing map (SOM) as an alternative to multilayer perceptron (MLP) and radial basis... more

In this paper, we introduce a general modeling technique, called vector-quantized temporal associative memory (VQTAM), which uses Kohonen's self-organizing map (SOM) as an alternative to multilayer perceptron (MLP) and radial basis function (RBF) neural models for dynamical system identification and control. We demonstrate that the estimation errors decrease as the SOM training proceeds, allowing the VQTAM scheme to be understood as a self-supervised gradient-based error reduction method. The performance of the proposed approach is evaluated on a variety of complex tasks, namely: i) time series prediction; ii) identification of SISO/MIMO systems; and iii) nonlinear predictive control. For all tasks, the simulation results produced by the SOM are as accurate as those produced by the MLP network, and better than those produced by the RBF network. The SOM has also shown to be less sensitive to weight initialization than MLP networks. We conclude the paper by discussing the main properties of the VQTAM and their relationships to other well established methods for dynamical system identification. We also suggest directions for further work.

Complexity theory of circuits strongly suggests that deep architectures can be much more ef cient (sometimes exponentially) than shallow architectures, in terms of computational elements required to represent some functions. Deep... more

Complexity theory of circuits strongly suggests that deep architectures can be much more ef cient (sometimes exponentially) than shallow architectures, in terms of computational elements required to represent some functions. Deep multi-layer neural networks have many levels of non-linearities allowing them to compactly represent highly non-linear and highly-varying functions. However, until recently it was not clear how to train such deep networks, since gradient-based optimization starting from random initialization ...

This paper is an attempt to integrate in a general model the major findings reported earlier in this series on: lability and history dependence of circadian period, τ (Pittendrigh and Daan, 1976 a); dependence of τ and α on light... more

This paper is an attempt to integrate in a general model the major findings reported earlier in this series on: lability and history dependence of circadian period, τ (Pittendrigh and Daan, 1976 a); dependence of τ and α on light intensity as described in Aschoff's Rule (Daan and Pittendrigh, 1976b); the interrelationships between τ and phase response curves (Daan and Pittendrigh, 1976a); and those inconsistencies between experimental facts on entrainment and theoretical predictions based on a single oscillator with fixed parameters τ and PRC, which pointed to a more complex system (Pittendrigh and Daan, 1976b). The qualitative model developed consists of two oscillators. The evidence that two separate oscillators are involved in circadian activity rhythms rests largely on the “splitting” phenomenon, known to occur in several species of mammals and birds. The empirical regularities of “splitting” in hamsters exposed to constant illumination (LL) are described: Splitting, i.e. the dissociation of a single activity band into two components which become stably coupled in circa 180° antiphase, occurs in about 50% of the animals in 100–200 lux, and has not been observed in lower light intensities. Splitting never occurred before 40 days after the transition to LL, unless the pretreatment had been LL of low intensity. In some animals the unsplit condition returned spontaneously. The attainment of antiphase is usually accompanied by a decrease in τ, and refusion of the two components by an increase in τ. These data show that both the split and the unsplit condition are metastable states, characterized by different phase relationships (ψ EM ) of two constituent oscillators. ψEM is history-dependent and determines τ of the coupled system. Observations in Peromyscus leucopus transferred from LL to DD to LD 12∶12 show that the two components of the bimodal activity peak (in LD) can for some time run at different frequencies (in DD), suggesting that bimodality of activity peaks and splitting are based on the same two-oscillator system. The model developed assumes the existence of two oscillators or principal groups of oscillators E and M, with opposite dependence of spontaneous frequency on light intensity. The dependence of the phase relationship (ψ EM ) between the two on light intensity and the dependence of τ onψ EM account for all the history-dependent characteristics of circadian pacemakers, and for the interdependence of τ, PRC, and τ-lability. The model qualitatively accommodates the interdependence of τ and α summarized in Aschoffs Rule. It is noted that the major intuitive elements in the model have been found to characterize an explicit version of it in computer simulations. The relevance of the model to seasonal change in photoperiod is discussed. A pacemaker comprising two oscillators mutually interacting but coupled separately to sunrise and sunset enhances its competence to accommodate to seasonal change in the daily pattern of external conditions; and it could well be involved in the pacemaker's known ability to discriminate between daylengths in the phenomena of photoperiodic induction.

This paper describes a probabilistic approach to the problem of recognizing places based on their appearance. The system we present is not limited to localization, but can determine that a new observation comes from a previously unseen... more

This paper describes a probabilistic approach to the problem of recognizing places based on their appearance. The system we present is not limited to localization, but can determine that a new observation comes from a previously unseen place, and so augment its map. Effectively this is a SLAM system in the space of appearance. Our probabilistic approach allows us to explicitly account for perceptual aliasing in the environment—identical but indistinctive observations receive a low probability of having come from the same place. We achieve this by learning a generative model of place appearance. By partitioning the learning problem into two parts, new place models can be learned online from only a single observation of a place. The algorithm complexity is linear in the number of places in the map,
and is particularly suitable for online loop closure detection in mobile robotics.

Deep multi-layer neural networks have many levels of non-linearities allowing them to compactly represent highly non-linear and highly-varying functions. However, until recently it was not clear how to train such deep networks, since... more

Deep multi-layer neural networks have many levels of non-linearities allowing them to compactly represent highly non-linear and highly-varying functions. However, until recently it was not clear how to train such deep networks, since gradient-based optimization starting from random initialization often appears to get stuck in poor solutions. Hinton et al. recently proposed a greedy layer-wise unsupervised learning procedure relying on the training algorithm of restricted Boltzmann machines (RBM) to initialize the parameters of a ...

Early intervention at the onset of psychotic disorders is a highly attractive theoretical notion that is receiving increasing international interest. In practical terms, it amounts to first deciding when a psychotic disorder can be said... more

Early intervention at the onset of psychotic disorders is a highly attractive theoretical notion that is receiving increasing international interest. In practical terms, it amounts to first deciding when a psychotic disorder can be said to have commenced and then offering potentially effective treatment at the earliest possible point. A second element involves ensuring that this intervention constitutes best practice for this phase of illness and is not merely the translation of standard treatments developed for later stages and the more persistently ill subgroups of the disorder. Furthermore, it means ensuring that this best practice model is actually delivered to patients and families. The relative importance of these elements in relation to outcome has not yet been established. This article outlines a framework for preventive intervention in early psychosis, based on more than a decade of experience initially gained within a first-generation model. This experience has been follow...

This paper reviews the cohesive process zone model, a general model which can deal with the nonlinear zone ahead of the crack tip––due to plasticity or microcracking––present in many materials. Furthermore, the cohesive zone model is able... more

This paper reviews the cohesive process zone model, a general model which can deal with the nonlinear zone ahead of the crack tip––due to plasticity or microcracking––present in many materials. Furthermore, the cohesive zone model is able to adequately predict the behaviour of uncracked structures, including those with blunt notches, and not only the response of bodies with cracks––a usual drawback of most fracture models. The cohesive zone model, originally applied to concrete and cementitious composites, can be used with success for other materials. More powerful computer programs and better knowledge of material properties may widen its potential field of application. In this paper, the cohesive zone model is shown to provide good predictions for concrete and for different notched samples of a glassy polymer (PMMA) and some steels. The paper is structured in two main sections: First, the cohesive model is reviewed and emphasis is on determination of the softening function, an essential ingredient of the cohesive model, by inverse analysis procedures. The second section is devoted to some examples of the predictive capability of the cohesive zone model when applied to different materials; concrete, PMMA and steel.

In this paper, we consider how rich sources of information on consumer choice can help to identify demand parameters in a widely used class of differentiated products demand models. Most important, we show how to use “secondâ€choiceâ€... more

In this paper, we consider how rich sources of information on consumer choice can help to identify demand parameters in a widely used class of differentiated products demand models. Most important, we show how to use “secondâ€choice†data on automotive purchases to obtain good estimates of substitution patterns in the automobile industry. We use our estimates to make outâ€ofâ€sample predictions