Please read the abstract in the section 00front of this document / Dissertation (MSc (Computer Science))--University of Pretoria, 2007. / Computer Science / unrestricted

Robotic simulators for use in Evolutionary Robotics (ER) have certain challenges associated with the complexity of their construction and the accuracy of predictions made by these simulators. Such robotic simulators are often based on physics models, which have been shown to produce accurate results. However, the construction of physics-based simulators can be complex and time-consuming. Alternative simulation schemes construct robotic simulators from empirically-collected data. Such empirical simulators, however, also have associated challenges, such as that some of these simulators do not generalize well on the data from which they are constructed, as these models employ simple interpolation on said data. As a result of the identified challenges in existing robotic simulators for use in ER, this project investigates the potential use of Artificial Neural Networks, henceforth simply referred to as Neural Networks (NNs), as alternative robotic simulators. In contrast to physics models, NN-based simulators can be constructed without needing an explicit mathematical model of the system being modeled, which can simplify simulator development. Furthermore, the generalization capabilities of NNs suggest that NNs could generalize well on data from which these simulators are constructed. These generalization abilities of NNs, along with NNs’ noise tolerance, suggest that NNs could be well-suited to application in robotics simulation. Investigating whether NNs can be effectively used as robotic simulators in ER is thus the endeavour of this work. Since not much research has been done in employing NNs as robotic simulators, many aspects of the experimental framework on which this dissertation reports needed to be carefully decided upon. Two robot morphologies were selected on which the NN simulators created in this work were based, namely a differentially steered robot and an inverted pendulum robot. Motion tracking and robotic sensor logging were used to acquire data from which the NN simulators were constructed. Furthermore, custom code was written for almost all aspects of the study, namely data acquisition for NN training, the actual NN training process, the evolution of robotic controllers using the created NN simulators, as well as the onboard robotic implementations of evolved controllers. Experimental tests performed in order to determine ideal topologies for each of the NN simulators developed in this study indicated that different NN topologies can lead to large differences in training accuracy. After performing these tests, the training accuracy of the created simulators was analyzed. This analysis showed that the NN simulators generally trained well and could generalize well on data not presented during simulator construction. In order to validate the feasibility of the created NN simulators in the ER process, these simulators were subsequently used to evolve controllers in simulation, similar to controllers developed in related studies. Encouraging results were obtained, with the newly-evolved controllers allowing real-world experimental robots to exhibit obstacle avoidance and light-approaching behaviour with a reasonable degree of success. The created NN simulators furthermore allowed for the successful evolution of a complex inverted pendulum stabilization controller in simulation. It was thus clearly established that NN-based robotic simulators can be successfully employed as alternative simulation schemes in the ER process.

The drive for autonomy in manufacturing is making increasing demands on control systems, both for improved performance and for extra flexibility. This is reflected in the research and development of autonomously guided vehicles which must operate safely in ill-defined, complex and time-varying environments. Traditional control systems generally make infeasible assumptions which limit their application within this domain, and therefore current research has concentrated on Intelligent Control techniques in order to make the control systems flexible and robust. An integral part of intelligence is the ability to learn from a systems interaction with its environment, and this thesis provides a unified description of several adaptive neural and fuzzy networks. The recent resurgence of interest in these two anthropomorphic techniques has seen these algorithms widely applied within learning control systems, although a firm theoretical framework which can compare different networks and establish convergence and stability conditions has not evolved. Such results are essential if these adaptive algorithms are to be used in real-world applications where safety and correctness are prime concerns. The work described in this thesis addresses these questions by introducing a class of systems called associative memory networks, which is used to describe the similarities and differences which exist between certain fuzzy and neural algorithms. All of the networks can be implemented within a 3-layer structure, where the output is linearly dependent on a set of adjustable parameters. This allows parameter convergence to be established when a gradient descent training rule is used, and the rate of convergence can be directly related to the condition of the network's basis functions. The size, shape and position of these basis functions gives each network its own specific modelling attributes, since the learning rules are identical. Therefore it is important to study the network's internal representation as this provides information about how each network generalises (both interpolation and extrapolation), the rate of parameter convergence and the type of nonlinear functions which can be successfully modelled. Three networks are described in detail: the Albus CMAC, the is given of the Albus CMAC which illustrates its desirable features for on-line, nonlinear adaptive modelling and control: local learning and a computational cost which depends linearly on the input space dimension. The modelling capabilities of the algorithm are rigorously analysed and it is shown that they are strongly dependent on the generalisation parameter, and a set of consistency equations is derived which specify how the network generalises. The adaptive B-spline network, which embodies a piecewise polynomial representation, is also described and used for nonlinear modelling and constructing a static rule base which guides and autonomous vehicle into a parking slot. B-splines are also used for on-line, constrained trajectory generation where they approximate a set of velocity or positional subgoals. Fuzzy systems are typically ill-defined, although the approach taken in this thesis is to use algebraic rather than truncation operators and smooth fuzzy sets which means that the modelling capabilities of the fuzzy network can be determined exactly, and convergence and stability results can be derived for these algorithms. These results focus research on the learning, modelling and representational abilities of the networks by providing a common framework for their analysis. The desirable features of the networks (local learning, linearly dependent on the parameter set, fuzzy interpretation) are emphasised, and the algorithms are all evaluated on a common time series prediciton problem.

Analyzing large volumes of video data is a challenging and time-consuming task. Automating this process would very valuable, especially in ecological research where massive amounts of video can be used to unlock new avenues of ecological research into the behaviour of animals in their environments. Deep Neural Networks, particularly Deep Convolutional Neural Networks, are a powerful class of models for computer vision. When combined with Recurrent Neural Networks, Deep Convolutional models can be applied to video for frame level video classification. This research studies two datasets: penguins and seals. The purpose of the research is to compare the performance of image-only CNNs, which treat each frame of a video independently, against a combined CNN-RNN approach; and to assess whether incorporating the motion information in the temporal aspect of video improves the accuracy of classifications in these two datasets. Video and image-only models offer similar out-of-sample performance on the simpler seals dataset but the video model led to moderate performance improvements on the more complex penguin action recognition dataset.

No description available.

No / This paper presents realistic avatar movements using a limited number of sensors. An inverse kinematics algorithm, SHAKF, is used to configure an articulated skeletal model, and a neural network is employed to predict the movement of joints not bearing sensors. The results show that the neural network is able to give a very close approximation to the actual rotation of the joints. This allows a substantial reduction in the number of sensors to configure an articulated human skeletal model.

The hippocampus has been the focus of memory research for decades. While the functional role of this structure is not fully understood, it is widely recognized as being vital for rapid yet accurate encoding and retrieval of associative memories. Since the discovery of adult hippocampal neurogenesis (AHN) in the dentate gyrus (DG) by Altman and Das in the 1960s, many theories and models have been formulated to explain the functional role it plays in learning and memory. These models postulate different ways in which new neurons are introduced into the DG and their functional importance for learning and memory. Few, if any, previous models have incorporated the unique properties of young adult-born dentate granule cells (DGCs) and their developmental trajectory. In this thesis, we propose a novel computational model of the DG that incorporates the developmental trajectory of these DGCs, including changes in synaptic plasticity, connectivity, excitability and lateral inhibition, using a modified version of the restricted boltzmann machine (RBM). Our results show superior performance on memory reconstruction tasks for both recent and distally learned items, when the unique characteristics of young DGCs are taken into account. The unique properties of the young neurons contribute to reducing retroactive and proactive interference, at both short and long time scales, despite the reduction in pattern separation due to their hyperexcitability. Our replacement model is subsequently extended to support learning dependent regulation of neurogenesis and apoptosis, using a convergence based approach to network growing and pruning. This hybrid additive and replacement model provides a more realistic and flexible approach to investigating the role of neurogenesis regulation in learning and memory. Finally, we incorporate the dentate gyrus model into a full hippocampal circuit to assess cued recall performance. Once again, our neurogenesis model shows decreased proactive and retroactive interference. / Thesis / Master of Science (MSc)

An analysis of the fire occurrences parameters is essential to save human lives, property, timber resources and conservation of biodiversity. Data conversion formats such as raster to ASCII facilitate the integration of various GIS software’s in the context of RS and GIS modeling. This research explores fire occurrences in relation to human interaction, fuel density interaction, euclidean distance from the perennial streams and slope using artificial neural networks. The human interaction (ignition source) and density of fuels is assessed by Newton’s Gravitational theory. Euclidean distance to perennial streams and slope that do posses a significant role were derived using GIS tools. All the four non linear predictor variables were modeled using the inductive nature of neural networks. The Self organizing feature map (SOM) utilized for fire size risk classification produced an overall classification accuracy of 62% and an overall kappa coefficient of 0.52 that is moderate (fair) for annual fires.

No description available.

Affective computing has gained significant attention from researchers in the last decade due to the wide variety of applications that can benefit from this technology. Often, researchers describe affect using emotional dimensions such as arousal and valence. Valence refers to the spectrum of negative to positive emotions while arousal determines the level of excitement. Describing emotions through continuous dimensions (e.g. valence and arousal) allows us to encode subtle and complex affects as opposed to discrete emotions, such as the basic six emotions: happy, anger, fear, disgust, sad and neutral. Recognizing spontaneous and subtle emotions remains a challenging problem for computers. In our work, we employ two modalities of information: video and audio. Hence, we extract visual and audio features using deep neural network models. Given that emotions are time-dependent, we apply the Temporal Convolutional Neural Network (TCN) to model the variations in emotions. Additionally, we investigate an alternative model that combines a Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN). Given our inability to fit the latter deep model into the main memory, we divide the RNN into smaller segments and propose a scheme to back-propagate gradients across all segments. We configure the hyperparameters of all models using Gaussian processes to obtain a fair comparison between the proposed models. Our results show that TCN outperforms RNN for the recognition of the arousal and valence emotional dimensions. Therefore, we propose the adoption of TCN for emotion detection problems as a baseline method for future work. Our experimental results show that TCN outperforms all RNN based models yielding a concordance correlation coefficient of 0.7895 (vs. 0.7544) on valence and 0.8207 (vs. 0.7357) on arousal on the validation dataset of SEWA dataset for emotion prediction.