This thesis explores the use of artificial neural networks for modelling cognitive processes. It presents an exposition of the neural network paradigm, and evaluates its viability in relation to the classical, symbolic approach in cognitive science. Classical researchers have approached the description of cognition by concentrating mainly on an abstract, algorithmic level of description in which the information processing properties of cognitive processes are emphasised. The approach is founded on seminal ideas about computation, and about algorithmic description emanating, amongst others, from the work of Alan Turing in mathematical logic. In contrast to the classical conception of cognition, neural network approaches are based on a form of neurocomputation in which the parallel distributed processing mechanisms of the brain are highlighted. Although neural networks are generally accepted to be more neurally plausible than their classical counterparts, some classical researchers have argued that these networks are best viewed as implementation models, and that they are therefore not of much relevance to cognitive researchers because information processing models of cognition can be developed independently of considerations about implementation in physical systems. In the thesis I argue that the descriptions of cognitive phenomena deriving from neural network modelling cannot simply be reduced to classical, symbolic theories. The distributed representational mechanisms underlying some neural network models have interesting properties such as similarity-based representation, content-based retrieval, and coarse coding which do not have straightforward equivalents in classical systems. Moreover, by placing emphasis on how cognitive processes are carried out by brain-like mechanisms, neural network research has not only yielded a new metaphor for conceptualising cognition, but also a new methodology for studying cognitive phenomena. Neural network simulations can be lesioned to study the effect of such damage on the behaviour of the system, and these systems can be used to study the adaptive mechanisms underlying learning processes. For these reasons, neural network modelling is best viewed as a significant theoretical orientation in the cognitive sciences, instead of just an implementational endeavour. / Psychology / D. Litt. et Phil. (Psychology)

Indiana University-Purdue University Indianapolis (IUPUI) / Alzheimer's disease (AD) is a neurodegenerative disorder characterized by gradual loss of brain functions, usually preceded by memory impairments. It has been widely affecting aging Americans over 65 old and listed as 6th leading cause of death. More importantly, unlike other diseases, loss of brain function in AD progression usually leads to the significant decline in self-care abilities. And this will undoubtedly exert a lot of pressure on family members, friends, communities and the whole society due to the time-consuming daily care and high health care expenditures. In the past decade, while deaths attributed to the number one cause, heart disease, has decreased 16 percent, deaths attributed to AD has increased 68 percent. And all of these situations will continue to deteriorate as the population ages during the next several decades. To prevent such health care crisis, substantial efforts have been made to help cure, slow or stop the progression of the disease. The massive data generated through these efforts, like multimodal neuroimaging scans as well as next generation sequences, provides unprecedented opportunities for researchers to look into the deep side of the disease, with more confidence and precision. While plenty of efforts have been made to pull in those existing machine learning and statistical models, the correlated structure and high dimensionality of imaging and genetics data are generally ignored or avoided through targeted analysis. Therefore their performances on imaging genetics study are quite limited and still have plenty to be improved. The primary contribution of this work lies in the development of novel prior knowledge-guided regression and association models, and their applications in various neurobiological problems, such as identification of cognitive performance related imaging biomarkers and imaging genetics associations. In summary, this work has achieved the following research goals: (1) Explore the multimodal imaging biomarkers toward various cognitive functions using group-guided learning algorithms, (2) Development and application of novel network structure guided sparse regression model, (3) Development and application of novel network structure guided sparse multivariate association model, and (4) Promotion of the computation efficiency through parallelization strategies.

Networks of interneurons within the spinal cord coordinate the rhythmic activation of muscles during locomotion. These networks are subject to extensive neuromodulation, ensuring appropriate behavioural output. Astrocytes are proposed to detect neuronal activity via Gαq-linked G-protein coupled receptors and to secrete neuromodulators in response. However, there is currently a paucity of evidence that astrocytic information processing of this kind is important in behaviour. Here, it is shown that protease-activated receptor-1 (PAR1), a Gαq-linked receptor, is preferentially expressed by glia in the spinal cords of postnatal mice. During ongoing locomotor-related network activity in isolated spinal cords, PAR1 activation stimulates release of adenosine triphosphate (ATP), which is hydrolysed to adenosine extracellularly. Adenosine then activates A1 receptors to reduce the frequency of locomotor-related bursting recorded from ventral roots. This entails inhibition of D1 dopamine receptors, activation of which enhances burst frequency. The effect of A1 blockade scales with network activity, consistent with activity-dependent production of adenosine by glia. Astrocytes also regulate activity by controlling the availability of D-serine or glycine, both of which act as co-agonists of glutamate at N-methyl-D-aspartate receptors (NMDARs). The importance of NMDAR regulation for locomotor-related activity is demonstrated by blockade of NMDARs, which reduces burst frequency and amplitude. Bath-applied D-serine increases the frequency of locomotor-related bursting but not intense synchronous bursting produced by blockade of inhibitory transmission, implying activity-dependent regulation of co-agonist availability. Depletion of endogenous D-serine increases the frequency of locomotor-related but not synchronous bursting, indicating that D-serine is required at a subset of NMDARs expressed by inhibitory interneurons. Blockade of the astrocytic glycine transporter GlyT1 increases the frequency of locomotor-related activity, but application of glycine has no effect, indicating that GlyT1 regulates glycine at excitatory synapses. These results indicate that glia play an important role in regulating the output of spinal locomotor networks.

Establishing and maintaining successful communication stands as a critical prerequisite for achieving the goals of inducing and studying advanced computation in small-scale living neuronal networks. The following work establishes a novel and effective method for communicating arbitrary "sensory" input information to cultures of living neurons, living neuronal networks (LNNs), consisting of approximately 20 000 rat cortical neurons plated on microelectrode arrays (MEAs) containing 60 electrodes. The sensory coding algorithm determines a set of effective codes (symbols), comprised of different spatio-temporal patterns of electrical stimulation, to which the LNN consistently produces unique responses to each individual symbol. The algorithm evaluates random sequences of candidate electrical stimulation patterns for evoked-response separability and reliability via a support vector machine (SVM)-based method, and employing the separability results as a fitness metric, a genetic algorithm subsequently constructs subsets of highly separable symbols (input patterns). Sustainable input/output (I/O) bit rates of 16-20 bits per second with a 10% symbol error rate resulted for time periods of approximately ten minutes to over ten hours. To further evaluate the resulting code sets' performance, I used the system to encode approximately ten hours of sinusoidal input into stimulation patterns that the algorithm selected and was able to recover the original signal with a normalized root-mean-square error of 20-30% using only the recorded LNN responses and trained SVM classifiers. Response variations over the course of several hours observed in the results of the sine wave I/O experiment suggest that the LNNs may retain some short-term memory of the previous input sample and undergo neuroplastic changes in the context of repeated stimulation with sensory coding patterns identified by the algorithm.