Vidros de spins são sistemas extremamente complexos caracterizados por um número enorme de estados estáveis e meta estáveis. Se identificarmos cada um desses estados com uma informação memorizada, esses sistemas podem ser utilizados como memórias associativas ou endereçáveis por conteúdo. O modelo de vidro de spins passa então a ser chamado de rede de neurônios. Neste trabalho estudamos a termodinâmica e alguns aspectos dinâmicos de uma rede de neurônios com processamento paralelo ou síncrono - o Modelo de Little de memória associativa - no regime em que o número de informações memorizadas p cresce como p = αN, onde N é o número de neurônios. Usando a teoria simétrica em relação às réplicas obtemos o diagrama de fases no espaço de parâmetros do modelo no qual incluímos um termo de autointeração dos neurônios.A riqueza do diagrama de fases que possui uma superfície de pontos tricríticos é devida à competição entre os dois regimes assintóticos da dinâmica síncrona: pontos fixos e ciclos de período dois. / Spin glasses are very complex systems characterized by a huge number of stable and metastable states. If we identify each state with a memorized information then spin glasses may be used as associative or content addressable memories. This spin glass model is then called a neural network. In this work we study the thermodynamics and some dynamical aspects of a neural network with parallel or synchronous processing - Little\'s model of associative memory -in the regime where the number of memorized informations p grows as p = αN, where N is the number of neurons. Using the replica symmetric theory we determine the phase diagram in the space of the model\'s parameters, in which we include a neural self interaction term. The richness of the phase diagram which possesses a surface of tricritical points is due to the competition between the two asymptotic dynamical behaviours of the synchronous dynamics: fixed points and cycles of lenght two.

Previous research has consistently shown associative deficits in older adults learning and memory (Chalfonte & Johnson 1996; Naveh-Benjamin, 2000; Naveh-Benjamin, Hussain, & Bar-On 2003) that are related to decreases in hippocampal function (Driscoll et al., 2003; Mitchell, Johnson, Raye, & D’Esposito, 2000). However, older adults learn certain simple predictive relationships between events (Mutter & Williams, 2004) that involve basal ganglia dependent error-driven learning. The goal of the current study was to determine whether error-driven learning could reduce the age-related associative deficits that are associated with hippocampal decline. The results did not support the idea that error-driven learning enhanced older adults’ associative memory, although our study supported normal error-driven processing in older adults. Our study confirms prior findings showing that age differences in associative memory are greater following an error-driven learning task than following an observation learning task (Schmitt-Eliassen, Ferstl, Wiesner, Deuschl, & Witt, 2007; Shohamy et al., 2004). Therefore, the results of the study did not support enhanced associative memory for older adults due to errordriven processing.

With the rapid growth of load demand, the distribution system is becoming more and more complicated, and the operational efficiency and service quality deteriorated. Power system protection is important for service reliability and quality assurance. Various faults may occur due to natural and artificial calamity. To reduce the outage duration and promptly restore power services, fault section estimate has to be done effectively with appeared fault alarms. The distribution system containing numerous protective facilities and switch equipment ranges over wide boundary. It becomes very complicated for dispatchers to obtain restoration plan for out-of-service areas. To cope with the problem, an effective tool is helpful for the restoration. This thesis proposes the use of Bi-directional associative memory networks (BAMN) to develop alarm processing. And use of Probabilistic Neural Network (PNN) to develop fault section detection, fault isolation, and restoration system. A distribution system is selected for computer simulation to demonstrate the effectiveness of the proposed system. The thesis proposes to use Bi-directional Associative Memory Network¡]BAMN¡^ to pre-process the signal gained from SCADA Interface, and transmit correct signal to Probabilistic Neural Network (PNN) for restoration plan . Computer simulation shows a simplified model to shorten the processing time in this study.

We investigated the concurrent effects of arousal and encoding specificity as related to background music on associative memory accuracy. Extant literature suggested these factors affect memory, but their combined effect in musical stimuli was not clear and may affect memory differentially for young and older adults. Specifically, we sought to determine if music can be used as a mnemonic device to overcome the associative memory deficits typically experienced by healthy older adults. We used a paired-associates memory task in which young and older adults listened to either highly or lowly arousing music or to silence while simultaneously studying same gender face-name pairs. Participants' memory was then tested for these pairs while listening to either the same or different music selections. We found that young adults' memory performance was not affected by any of the music listening conditions. Music listening, however, was detrimental for older adults. Specifically, their memory performance was worse for all music conditions, particularly if the music was highly arousing. Young adults' pattern of results was not reflected in their subjective ratings of helpfulness; they felt that all music was helpful to their performance yet there was no indication of this in the results. Older adults were more aware of the detriment of music on their performance, rating some highly arousing music as less helpful than silence. We discuss possible reasons for this pattern and conclude that these results are most consistent with the theory that older adults' failure to inhibit processing of distracting task-irrelevant information, in this case background music, contributes to their elevated memory failures.

In this thesis, we build upon the work of Plate by advancing the theory and utility of Holographic Reduced Representations (HRRs). HRRs are a type of linear, associative memory developed by Plate and are an implementation of Hinton’s reduced representations. HRRs and HRR-like representations have been used to model human memory, to model understanding analogies, and to model the semantics of natural language. However, in previous research, HRRs are restricted to storing and retrieving vectors of random numbers, limiting both the ability of HRRs to model human performance in detail, and the potential applications of HRRs. We delve into the theory of HRRs and develop techniques to store and retrieve images, or other kinds of structured data, in an HRR. We also investigate square matrix representations as an alternative to HRRs, and use iterative training algorithms to improve HRR performance. This work provides a foundation for cognitive modellers and computer scientists to explore new applications of HRRs. / Thesis (Master, Computing) -- Queen's University, 2010-08-10 12:50:04.004

Neurocomputers offer a massively parallel computing paradigm by mimicking the human brain. Their efficient use in statistical information processing has been proposed to overcome critical bottlenecks with traditional computing schemes for applications such as image and speech processing, and associative memory. In neural networks information is generally represented by phase (e.g., oscillatory neural networks) or amplitude (e.g., cellular neural networks). Phase-based neurocomputing is constructed as a network of coupled oscillatory neurons that are connected via programmable phase elements. Representing each neuron circuit with one oscillatory device and implementing programmable phases among neighboring neurons, however, are not clearly feasible from circuits perspective if not impossible. In contrast to nascent oscillatory neurocomputing circuits, mature amplitude-based neural networks offer more efficient circuit solutions using simpler resistive networks where information is carried via voltage- and current-mode signals. Yet, such circuits have not been efficiently realized by CMOS alone due to the needs for an efficient summing mechanism for weighted neural signals and a digitally-controlled weighting element for representing couplings among artificial neurons. Large power consumption and high circuit complexity of such CMOS-based implementations have precluded adoption of amplitude-based neurocomputing circuits as well, and have led researchers to explore the use of emerging technologies for such circuits. Although they provide intriguing properties, previously proposed neurocomputing components based on emerging technologies have not offered a complete and practical solution to efficiently construct an entire system. In this thesis we explore the generalized problem of co-optimization of technology and architecture for such systems, and develop a recipe for device requirements and target capabilities. We describe four plausible technologies, each of which could potentially enable the implementation of an efficient and fully-functional neurocomputing system. We first investigate fully-digital neural network architectures that have been tried before using CMOS technology in which many large-size logic gates such as D flip-flops and look-up tables are required. Using a newly-proposed all-magnetic non-volatile logic family, mLogic, we demonstrate the efficacy of digitizing the oscillators and phase relationships for an oscillatory neural network by exploiting the inherent storage as well as enabling an all-digital cellular neural network hardware with simplified programmability. We perform system-level comparisons of mLogic and 32nm CMOS for both networks consisting of 60 neurons. Although digital implementations based on mLogic offer improvements over CMOS in terms of power and area, analog neurocomputing architectures seem to be more compatible with the greatest portion of emerging technologies and devices. For this purpose in this dissertation we explore several emerging technologies with unique device configurations and features such as mCell devices, ovenized aluminum nitride resonators, and tunable multi-gate graphene devices to efficiently enable two key components required for such analog networks – that is, summing function and weighting with compact D/A (digital-to-analog) conversion capability. We demonstrate novel ways to implement these functions and elaborate on our building blocks for artificial neurons and synapses using each technology. We verify the functionality of each proposed implementation using various image processing applications based on compact circuit simulation models for such post-CMOS devices. Finally, we design a proof-of-concept neurocomputing circuitry containing 20 neurons using 65nm CMOS technology that is based on the primitives that we define for our analog neurocomputing scheme. This allows us to fully recognize the inefficiencies of an all-CMOS implementation for such specific applications. We share our experimental results that are in agreement with circuit simulations for the same image processing applications based on proposed architectures using emerging technologies. Power and area comparisons demonstrate significant improvements for analog neurocomputing circuits when implemented using beyond- CMOS technologies, thereby promising huge opportunities for future energy-efficient computing.

The understanding and treatment of brain disorders as well as the development of intelligent machines is hampered by the lack of knowledge of how the brain fundamentally functions. Over the past century, we have learned much about how individual neurons and neural networks behave, however new tools are critically needed to interrogate how neural networks give rise to complex brain processes and disease conditions. Recent innovations in molecular techniques, such as optogenetics, have enabled neuroscientists unprecedented precision to excite, inhibit and record defined neurons. The impressive sensitivity of currently available optogenetic sensors and actuators has now enabled the possibility of analyzing a large number of individual neurons in the brains of behaving animals. To promote the use of these optogenetic tools, this thesis integrates cutting edge optogenetic molecular sensors which is ultrasensitive for imaging neuronal activity with custom wide field optical microscope to analyze a large number of individual neurons in living brains. Wide-field microscopy provides a large field of view and better spatial resolution approaching the Abbe diffraction limit of fluorescent microscope. To demonstrate the advantages of this optical platform, we imaged a deep brain structure, the Hippocampus, and tracked hundreds of neurons over time while mouse was performing a memory task to investigate how those individual neurons related to behavior. In addition, we tested our optical platform in investigating transient neural network changes upon mechanical perturbation related to blast injuries. In this experiment, all blasted mice show a consistent change in neural network. A small portion of neurons showed a sustained calcium increase for an extended period of time, whereas the majority lost their activities. Finally, using optogenetic silencer to control selective motor cortex neurons, we examined their contributions to the network pathology of basal ganglia related to Parkinson’s disease. We found that inhibition of motor cortex does not alter exaggerated beta oscillations in the striatum that are associated with parkinsonianism. Together, these results demonstrate the potential of developing integrated optogenetic system to advance our understanding of the principles underlying neural network computation, which would have broad applications from advancing artificial intelligence to disease diagnosis and treatment.

Vidros de spins são sistemas extremamente complexos caracterizados por um número enorme de estados estáveis e meta estáveis. Se identificarmos cada um desses estados com uma informação memorizada, esses sistemas podem ser utilizados como memórias associativas ou endereçáveis por conteúdo. O modelo de vidro de spins passa então a ser chamado de rede de neurônios. Neste trabalho estudamos a termodinâmica e alguns aspectos dinâmicos de uma rede de neurônios com processamento paralelo ou síncrono - o Modelo de Little de memória associativa - no regime em que o número de informações memorizadas p cresce como p = αN, onde N é o número de neurônios. Usando a teoria simétrica em relação às réplicas obtemos o diagrama de fases no espaço de parâmetros do modelo no qual incluímos um termo de autointeração dos neurônios.A riqueza do diagrama de fases que possui uma superfície de pontos tricríticos é devida à competição entre os dois regimes assintóticos da dinâmica síncrona: pontos fixos e ciclos de período dois. / Spin glasses are very complex systems characterized by a huge number of stable and metastable states. If we identify each state with a memorized information then spin glasses may be used as associative or content addressable memories. This spin glass model is then called a neural network. In this work we study the thermodynamics and some dynamical aspects of a neural network with parallel or synchronous processing - Little\'s model of associative memory -in the regime where the number of memorized informations p grows as p = αN, where N is the number of neurons. Using the replica symmetric theory we determine the phase diagram in the space of the model\'s parameters, in which we include a neural self interaction term. The richness of the phase diagram which possesses a surface of tricritical points is due to the competition between the two asymptotic dynamical behaviours of the synchronous dynamics: fixed points and cycles of lenght two.

Orientador: Peter Sussner / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Matematica, Estatistica e Computação Cientifica / Made available in DSpace on 2018-08-21T14:39:08Z (GMT). No. of bitstreams: 1 Medeiros_CarlosRenato_M.pdf: 3281824 bytes, checksum: 90da4e6d96fe7557a92fa34f461172e7 (MD5) Previous issue date: 2012 / Resumo: Em meados dos anos 90, a memória associativa morfológica (MAM) foi apresentada como um modelo de memória associativa distributiva que realiza determinadas operações morfológicas definidas na teoria matemática de álgebra mini-max. Os modelos de MAMs vêm em duas versões diferentes que são tolerantes a diferentes tipos de ruído nos padrões de entrada. Para superar esta desvantagem, recorremos à teoria mais recente da morfologia matemática em inf-semirreticulado cujos operadores elementares são autoduais e definimos um modelo de memória associativa neste quadro / Abstract: In the mid 1990's, the morphological associative memory (MAM) was introduced as a distributive associative memory model that performs certain morphological operations defined in the mathematical theory of mini-max algebra. MAM models come in two different versions that are tolerant to different types of noise in the input patterns. To overcome this drawback, we resort to the more recent theory of mathematical morphology (MM) on inf-semilattices whose elementary operators are self-dual and we define an associative memory (AM) model in this framework / Mestrado / Matematica Aplicada / Mestre em Matemática Aplicada

No description available.