Visualization of learning in multilayer perceptron networks using principal component analysis

This paper is concerned with the use of scientific visualization methods for the analysis of feedforward neural networks (NNs). Inevitably, the kinds of data associated with the design and implementation of neural networks are of very high dimensionality, presenting a major challenge for visualization. A method is described using the well-known statistical technique of principal component analysis (PCA). This is found to be an effective and useful method of visualizing the learning trajectories of many learning algorithms such as back-propagation and can also be used to provide insight into the learning process and the nature of the error surface.

An equivalent algorithm for simulating thermal effects of magma intrusion problems in porous rocks

An equivalent algorithm is proposed to simulate thermal effects of the magma intrusion in geological systems, which are composed of porous rocks. Based on the physical and mathematical equivalence, the original magma solidification problem with a moving boundary between the rock and intruded magma is transformed into a new problem without the moving boundary but with a physically equivalent heat source. From the analysis of an ideal solidification model, the physically equivalent heat source has been determined in this paper. The major advantage in using the proposed equivalent algorithm is that the fixed finite element mesh with a variable integration time step can be employed to simulate the thermal effect of the intruded magma solidification using the conventional finite element method. The related numerical results have demonstrated the correctness and usefulness of the proposed equivalent algorithm for simulating the thermal effect of the intruded magma solidification in geological systems. (C) 2003 Elsevier B.V. All rights reserved.

Is affective learning really resistant to extinction? Stimulus ratings and affective priming give different answers

Affective learning, the learning of likes and dislikes, is proposed to differ from signal learning, the learning of relationships between events. However, affective learning research varies in the methodology used, and in addition, researchers concerned primarily with affective learning tend to use different paradigms from those concerned with signal learning. The current research used an affective priming task in addition to verbal ratings to assess changes in the valence of neutral geometric shapes in an aversive differential conditioning procedure. After acquisition, affective learning was present as indexed by ratings and affective priming, whereas after extinction, affective learning remained significant only in the ratings. This study suggests that different measures of affective learning may be differentially sensitive to valence, which has implications for studies that employ verbal ratings as the sole measure of affective learning. Moreover, there is no evidence from the current study that affective learning differs from signal learning.

Extinction of affective learning: No evidence for dual process accounts of human learning

The acquisition and extinction of affective valence to neutral geometrical shape conditional stimuli was investigated in three experiments. Experiment 1 employed a differential conditioning procedure with aversive shock USs. Differential electrodermal responding was evident during acquisition and lost during extinction. As indexed by verbal ratings, the CS1 acquired negative valence during acquisition,which was reduced after extinction. Affective priming, a reaction time based demand free measure of stimulus valence, failed to provide evidence for affective learning. Experiment 2 employed pictures of happy and angry faces as USs.Valence ratings after acquisitionweremore positive for theCS paired with happy faces (CS-H) and less positive for the CS paired with angry faces (CS-A) than during baseline. Extinction training reduced the extent of acquired valence significantly for both CSs, however, ratings of the CS-A remained different from baseline. Affective priming confirmed these results yielding differences between CS-A and CS-H after acquisition for pleasant and unpleasant targets, but for pleasant targets only after extinction. Experiment 3 replicated the design of Experiment 2, but presented the US pictures backwardly masked. Neither rating nor affective priming measures yielded any evidence for affective learning. The present results confirm across two different experimental procedures that, contrary to predictions from dual process accounts of human learning, affective learning is subject to extinction.

Why fear snakes and spiders? Evidence for a learning-based account of fear-relevant stimulus-processing

Fear-relevant stimuli, such as snakes, spiders and heights, preferentially capture attention as compared to nonfear-relevant stimuli. This is said to reflect an encapsulated mechanism whereby attention is captured by the simple perceptual features of stimuli that have evolutionary significance. Research, using pictures of snakes and spiders, has found some support for this account; however, participants may have had prior fear of snakes and spiders that influenced results. The current research compared responses of snake and spider experts who had little fear of snakes and spiders, and control participants across a series of affective priming and visual search tasks. Experts discriminated between dangerous and nondangerous snakes and spiders, and expert responses to pictures of nondangerous snakes and spiders differed from those of control participants. The current results dispute that stimulus fear relevance is based purely on perceptual features, and provides support for the role of learning and experience.

The Australian Universities Teaching Committee project in 'Learning Outcomes and Curriculum Development in Psychology'

The Australian Universities Teaching Committee (AUTC) funds projects intended to improve the quality of teaching and learning in specific disciplinary areas. The project brief for 'Learning Outcomes and Curriculum Development in Psychology' for 2004/2005 was to 'produce an evaluative overview of courses ... with a focus on the specification and assessment of learning outcomes and ... identify strategic directions for universities to enhance teaching and learning'. This project was awarded to a consortium from The University of Queensland, University of Tasmania, and Southern Cross University. The starting point for this project is an analysis of the scientist-practitioner model and its role in curriculum design, a review of current challenges at a conceptual level, and consideration of the implications of recent changes to universities relating to such things as intemationalisation of programs and technological advances. The project will seek to bring together stakeholders from around the country in order to survey the widest possible range of perspectives on the project brief requirements. It is hoped also to establish mechanisms for fiiture scholarly discussion of these issues, including the establishment of an Australian Society for the Teaching of Psychology and an annual conference.

Special issue on methods and learning in functional MRI: Introductory remarks

This special issue represents a further exploration of some issues raised at a symposium entitled “Functional magnetic resonance imaging: From methods to madness” presented during the 15th annual Theoretical and Experimental Neuropsychology (TENNET XV) meeting in Montreal, Canada in June, 2004. The special issue’s theme is methods and learning in functional magnetic resonance imaging (fMRI), and it comprises 6 articles (3 reviews and 3 empirical studies). The first (Amaro and Barker) provides a beginners guide to fMRI and the BOLD effect (perhaps an alternative title might have been “fMRI for dummies”). While fMRI is now commonplace, there are still researchers who have yet to employ it as an experimental method and need some basic questions answered before they venture into new territory. This article should serve them well. A key issue of interest at the symposium was how fMRI could be used to elucidate cerebral mechanisms responsible for new learning. The next 4 articles address this directly, with the first (Little and Thulborn) an overview of data from fMRI studies of category-learning, and the second from the same laboratory (Little, Shin, Siscol, and Thulborn) an empirical investigation of changes in brain activity occurring across different stages of learning. While a role for medial temporal lobe (MTL) structures in episodic memory encoding has been acknowledged for some time, the different experimental tasks and stimuli employed across neuroimaging studies have not surprisingly produced conflicting data in terms of the precise subregion(s) involved. The next paper (Parsons, Haut, Lemieux, Moran, and Leach) addresses this by examining effects of stimulus modality during verbal memory encoding. Typically, BOLD fMRI studies of learning are conducted over short time scales, however, the fourth paper in this series (Olson, Rao, Moore, Wang, Detre, and Aguirre) describes an empirical investigation of learning occurring over a longer than usual period, achieving this by employing a relatively novel technique called perfusion fMRI. This technique shows considerable promise for future studies. The final article in this special issue (de Zubicaray) represents a departure from the more familiar cognitive neuroscience applications of fMRI, instead describing how neuroimaging studies might be conducted to both inform and constrain information processing models of cognition.

A new algorithm and refined bounds for extended gcd computation

Extended gcd computation is interesting itself. It also plays a fundamental role in other calculations. We present a new algorithm for solving the extended gcd problem. This algorithm has a particularly simple description and is practical. It also provides refined bounds on the size of the multipliers obtained.

The Qu-Prolog unification algorithm: Formalisation and correctness

Qu-Prolog is an extension of Prolog which performs meta-level computations over object languages, such as predicate calculi and lambda-calculi, which have object-level variables, and quantifier or binding symbols creating local scopes for those variables. As in Prolog, the instantiable (meta-level) variables of Qu-Prolog range over object-level terms, and in addition other Qu-Prolog syntax denotes the various components of the object-level syntax, including object-level variables. Further, the meta-level operation of substitution into object-level terms is directly represented by appropriate Qu-Prolog syntax. Again as in Prolog, the driving mechanism in Qu-Prolog computation is a form of unification, but this is substantially more complex than for Prolog because of Qu-Prolog's greater generality, and especially because substitution operations are evaluated during unification. In this paper, the Qu-Prolog unification algorithm is specified, formalised and proved correct. Further, the analysis of the algorithm is carried out in a frame-work which straightforwardly allows the 'completeness' of the algorithm to be proved: though fully explicit answers to unification problems are not always provided, no information is lost in the unification process.

Analysis and implementation of a new constant acceleration subcycling algorithm

An algorithm for explicit integration of structural dynamics problems with multiple time steps is proposed that averages accelerations to obtain subcycle states at a nodal interface between regions integrated with different time steps. With integer time step ratios, the resulting subcycle updates at the interface sum to give the same effect as a central difference update over a major cycle. The algorithm is shown to have good accuracy, and stability properties in linear elastic analysis similar to those of constant velocity subcycling algorithms. The implementation of a generalised form of the algorithm with non-integer time step ratios is presented. (C) 1997 by John Wiley & Sons, Ltd.

The subcycled Newmark algorithm

The popular Newmark algorithm, used for implicit direct integration of structural dynamics, is extended by means of a nodal partition to permit use of different timesteps in different regions of a structural model. The algorithm developed has as a special case an explicit-explicit subcycling algorithm previously reported by Belytschko, Yen and Mullen. That algorithm has been shown, in the absence of damping or other energy dissipation, to exhibit instability over narrow timestep ranges that become narrower as the number of degrees of freedom increases, making them unlikely to be encountered in practice. The present algorithm avoids such instabilities in the case of a one to two timestep ratio (two subcycles), achieving unconditional stability in an exponential sense for a linear problem. However, with three or more subcycles, the trapezoidal rule exhibits stability that becomes conditional, falling towards that of the central difference method as the number of subcycles increases. Instabilities over narrow timestep ranges, that become narrower as the model size increases, also appear with three or more subcycles. However by moving the partition between timesteps one row of elements into the region suitable for integration with the larger timestep these the unstable timestep ranges become extremely narrow, even in simple systems with a few degrees of freedom. As well, accuracy is improved. Use of a version of the Newmark algorithm that dissipates high frequencies minimises or eliminates these narrow bands of instability. Viscous damping is also shown to remove these instabilities, at the expense of having more effect on the low frequency response.

Simulated-annealing-based genetic algorithm for modeling the optical constants of solids

We propose a simulated-annealing-based genetic algorithm for solving model parameter estimation problems. The algorithm incorporates advantages of both genetic algorithms and simulated annealing. Tests on computer-generated synthetic data that closely resemble optical constants of a metal were performed to compare the efficiency of plain genetic algorithms against the simulated-annealing-based genetic algorithms. These tests assess the ability of the algorithms to and the global minimum and the accuracy of values obtained for model parameters. Finally, the algorithm with the best performance is used to fit the model dielectric function to data for platinum and aluminum. (C) 1997 Optical Society of America.