School of Human, Health and Social Sciences

lthough considerable research has focused on defining the precise nature of the behavioural symptomatology of ADHD: inattention, impulsivity, and hyperactivity (Swanson et al., 1998), the results are equivocal. This is partly because few studies have investigated different types of attention in the same ADHD population. Thus, the major goal of the current study was to further investigate the precise nature of ADHD 'inattention' symptomatology using two accepted measures of sustained and transient attention in the same population. Sustained attention performance was measured in 12 ADHD children and 25 nonnal control children using a traditional Continuous Performance Task (CPT) and a novel CPT. Transient attention performance was measured using a Visual Search Task (VST). The results revealed that the same group of ADHD children performed significantly worse on both the CPTs compared to the normals but that their performance on the VST was similar to that of normals. These results indicate that combined-type ADHD participants show impaired sustained attention, but not impaired transient attention performance, supporting the view that different ADHD subtypes have different deficit of attention. E OG correction techniques attempt to account for ocular artefact in the electroencephalogram (EEG). In order to compare the efficacy of four different EOG correction algorithms (Verleger et al., 1982 [VGM]; Gratton et al., 1983 [GCD]; Semlitsch et al., 1986 [SPSA]; Croft & Barry, 2000 [CB]), blinks, vertical (VEM) and horizontal (HEM) eye movements from 26 subjects were corrected using each of the above four algorithms, and eye movement event-related potentials (ERPs) computed. Validation was assessed by (a) combining ERPs in such a way as to make them orthogonal to the EOG channels and determining the degree to which the orthogonality assumption was violated by the correction procedure, and (b) using a form of split-half reliability to assess how similar the ERPs were after correction. The two validation methods yielded equivalent results. In order of correction outcomes, HEMs were corrected better by CB, VGM/GCD then SPSA, VEMs by CB, VGM/GCD and then SPSA, and blinks by CB, SPSA, GCD and then VGM. The advantage of CB over the other procedures was appreciable for all eye movement types (blinks, eta2 > 0.72; VEMs, eta2 > 0.60; HEMs, eta2 > 0.27).

A challenging task in psychophysiology is the extraction of event-related potentials (ERPs) from the background electro-encephalogram. The task is made more difficult by the properties of ERPs, which typically consist of multiple features of variable latency, Iocalised in time and frequency. A novel technique is described for analysis of ERPs, adaptive wavelet filtering (AWF), which is proposed as an alternative to trial averaging. Band-limited detail representations of each trial are obtained using wavelet analysis. The Woody adaptive filter is then used to align trials with respect to the evoked response. In a simulation study, the AWF extracts 39% of higher-frequency signal variance from background noise, compared with less than 1% for standard averaging and the Woody filter. The AWF is applied to a dataset of 448 ERPs, comprising right-finger button presses from eight subjects. Average split-half reliability of the AWF on scales up to 12Hz was 0.51.

Convolutional neural networks (CNNs) represent an interesting method for adaptive image processing, and form a link between general feedforward neural networks and adaptive filters. Two dimensional CNNs are formed by one or more layers of two dimensional filters, with possible non-linear activation functions and/or down-sampling. CNNs possess key properties of translation invariance and spatially local connections (receptive fields). The present paper presents a description of the convolutional network architecture, and an application to a practical image processing application on a mobile robot. A CNN is used to detect and characterize cracks on an autonomous sewer inspection robot. The filter sizes used in all layers of the CNN used here was 5x5, and the common activation function used was a log-sigmoid. The network was trained using a dataset of 39 still image 320x240 pixel frames sampled from a pre-recorded sewer pipe inspection video. The results support the notion that data-based adaptive image processing methods such as CNNs are useful for image processing, or other applications where the input arrays are large, and spatially / temporally distributed. Further refinements of the CNN architecture, such as the implementation of separable filters, or extensions to three dimensional (ie video) processing, are suggested.

A novel non-parametric density estimator is developed based on geometric principles. A penalised centroidal Voronoi tessellation forms the basis of the estimator, which allows the data to self-organise in order to minimise estimate bias and variance. This approach is a marked departure from usual methods based on local averaging, and has the advantage of being naturally adaptive to local sample density (scale-invariance). The estimator does not require the introduction of a plug-in kernel, thus avoiding assumptions of symmetricity and morphology. A numerical experiment is conducted to illustrate the behaviour of the estimator, and it's characteristics are discussed.

Prediction occurs in many biological nervous systems e.g. in the cortex . We introduce a method of adapting the recurrent layer dynamics of an echo-state network (ESN) without attempting to train the weights directly. Initially a network is generated that fulfils the echo state -liquid state condition. A second network is then trained to predict the next internal state of the system. In simulation, the prediction of this module is then mixed with the actual activation of the internal state neurons, to produce dynamics that a partially driven by the network model, rather than the input data. The mixture is determined by a parameter α. The target function be produced by the network was sin 3 (0.24t), given an input function sin(0.24t). White noise was added to the input signal at 15% of the amplitude of the signal. Preliminary results indicate that self prediction may improve performance of an ESN when performing signal mappings in the presence of additive noise.

Locally adaptive density estimation presents challenges for parametric or non-parametric estimators. Several useful properties of tessellation density estimators (TDEs), such as low bias, scale invariance and sensitivity to local data morphology, make them an attractive alternative to standard kernel techniques. However, simple TDEs are discontinuous and produce highly unstable estimates due to their susceptibility to sampling noise. With the motivation of addressing these concerns, we propose applying TDEs within a bootstrap aggregation algorithm, and incorporating model selection with complexity penalization. We implement complexity reduction of the TDE via sub-sampling, and use information-theoretic criteria for model selection, which leads to an automatic and approximately ideal bias/variance compromise. The procedure yields a stabilized estimator that automatically adapts to the complexity of the generating distribution and the quantity of information at hand, and retains the highly desirable properties of the TDE. Simulation studies presented suggest a high degree of stability and sensitivity can be obtained using this approach.

In this paper the standard Echo State approach is combined with a topography, i.e. it is assigned with a position which implies certain constraints of the mutual connectivity between these neurons. The overall design of the network allows certain neurons to process new information earlier than others. As a consequence the connectivity of the trained output layer can be analyzed; conclusions can be drawn regarding which reservoir depth is sufficient to process the given task. In particular we look at connection strengths of different locations of the reservoir as a function of the test error which can be influenced by using ridge regression.

The effectiveness of Savitzky-Golay type symmetric polynomial smoothers is known to be strongly dependent on the window size. Many authors note that selection of the appropriate window size is essential for achieving the correct trade-off between noise reduction and avoiding the introduction of bias. However, it is often overlooked that, in the case of non-stationary signals, the optimal window size will vary with the dynamics of the signal. A multiresolution approach is outlined, along with criteria for varying window size with respect to translation, based on evaluation of the residuals of the smoothed data in the local region. Adaptive window polynomial smoothing is shown to be superior to fixed window smoothing for a test signal at various signal-tonoise ratios.

Gamblers who participate in skill-oriented games (such as poker and sportsbetting) are motivated to win over the long-term, and some monitor their betting outcomes to evaluate their performance and proficiency. In this study of Australian off-track horserace betting, we investigated which levels of sustained returns would be required to establish evidence of skill/expertise. We modelled a random strategy to simulate 'naïve' play, in which equal bets were placed on randomly selected horses using a representative sample of 211 weekend races. Results from a Monte Carlo simulation yielded a distribution of return-on-investments for varying number of bets (N), showing surprising volatility, even after a large number of repeated bets. After adjusting for the house advantage, a gambler would have to place over 10,000 bets in individual races with net returns exceeding 9 % to be reasonably considered an expert punter (a = .05). Moreover, a record of fewer bets would require even greater returns for demonstrating expertise. As such, validated expertise is likely to be rare among race bettors. We argue that the counterintuitively high threshold for demonstrating expertise by tracking historical performance is likely to exacerbate known cognitive biases in self-evaluation of expertise.

Although electronic gaming machine (EGM) jackpots are widespread, little research has yet considered the impact of this feature on gamblers’ behaviour. We present the results of an in-venue shadowing study, which provided measures of player investment and persistence (e.g. number of spins, time-on-machine) from participants undertaking one or more EGM sessions on their choice of machines. 234 participants (162 female) were recruited in-venue, with half (stratified by age and gender) primed by answering questions encouraging ’big-win’ oriented ideation. Primed participants were more likely to select jackpot-oriented EGMs, and primed at-risk [Problem Gambling Severity Index (PGSI)[4] gamblers tended to select machines with a higher median jackpot prize amount than others (W 1⁄4 18,423; p 1⁄4 0:003). Neither PGSI nor priming was associated with the rate at which participants switched machines. EGM jackpots were associated with great spend overall, and PGSI score was associated with a greater spend per play. Positive interactions were found between jackpots and PGSI, and PGSI and priming in terms of predicting greater persistence. Finally a structural model of session level variables is presented, that incorporates positive feedback between money won and number of plays in an EGM session.