La production de mots intéresse un vaste réseau cérébral principalement localisé dans l'hémisphère gauche, synthétisé dans une revue de littérature traitant des études de production de mots en enregistrements intracérébraux. Nous nous sommes principalement focalisés sur l'implication du réseau cérébral lors de l'accès lexical en utilisant deux protocoles de dénomination d'images manipulant l'interférence sémantique, connue pour moduler les liens entre les étapes sémantiques et lexicales. Nous avons comparé la dynamique spatiotemporelle de ces protocoles utilisés de façon interchangeable dans la littérature, mais qui diffèrent dans l'implication de paramètres méthodologiques pouvant faire intervenir des mécanismes mnésiques tels que la familiarisation et la répétition. Notre hypothèse étant que ces paramètres étant si divergents, le réseau neuronal sous-jacent devrait au moins être modulé selon le protocole étudié. Notre étude EEG révèle deux patterns d'activité électrophysiologique distincts entre les protocoles attribuables à l'effet de familiarisation. Nous avons étudié l'implication de la structure hippocampique dans la production de mot via des enregistrements intracérébraux. Les analyses en temps-fréquence et en champs locaux montrent que l'hippocampe est impliqué dans l'apprentissage progressif des liens sémantico-lexicaux, mais aussi dans leur maintien en mémoire durant une courte période et dans leur récupération. Ce travail de thèse a permis de révéler que la dynamique spatiotemporelle de la dénomination d'images est modulée par différents facteurs, ce qui va à l'encontre de l'implication d'un réseau partagé par ces deux protocoles de dénomination d'images. / Words production involves a vast brain network mainly localized in the left hemisphere, summarized in a review of literature based on studies of word production in intracranial recordings.We focused principally on the involvement of the cerebral network during lexical access by using two picture naming protocols manipulating the semantic interference effect, known to modulate the links between semantic and lexical processes. We compare the spatiotemporal dynamics of these protocols interchangeably used in the literature, but which differ in the involvement of methodological parameters that may involve mnemonic mechanisms such as familiarity and repetition. Our hypothesis was that these parameters are so divergent that the underlying neural network should at least be modulated by the protocol investigated. Our EEG study reveals two distinct patterns of electrophysiological activity between the protocols due to the familiarization effect. We studied the involvement of the hippocampal structure in the production of word through intracranial recordings. The time-frequency and the local field analyses show that the hippocampus is involved in the progressive learning links between semantic and lexicon, but also in keeping them in memory for a short period and in their recovery. This work revealed that the spatiotemporal dynamics of picture naming is modulated by various factors, which goes against the involvement of a common network shared by these two picture naming protocols.

Working memory (WM) is a process that allows for the temporary and limited storage of information for an immediate goal or to be stored into a more permanent system. A large number of studies have led to the widely accepted view that WM is mediated by the frontoparietal network (FPN), consisting of areas in the prefrontal cortex (PFC) and posterior parietal cortex (PPC). Current evidence suggests that task specific patterns of neuronal oscillatory activity within the FPN play a fundamental role in WM, and yet specific spatio-temporal properties of this activity are not well characterized. This study utilized multisite local field potential (LFP) data recorded from PFC and PPC sites in two macaque monkeys trained to perform a rule-based, Oculomotor Delayed Match-to-Sample task. The animals were required to learn which of two rules determined the correct match (Location matching or Identity matching). Following a 500 ms fixation period, a sample stimulus was presented for 500 ms, followed by a randomized delay lasting 800-1200 ms in which no stimulus was present. At the end of the delay period, a match stimulus was presented, consisting of two of three possible objects presented at two of three possible locations. When the match stimulus appeared, the monkey made a saccadic eye movement to the target. The rule in effect determined which object served as the target. Time-frequency plots of three spectral measures (power, coherence, and Wiener Granger Causality (WGC) were computed from MultiVariate AutoRegressive LFP time-series models estimated in a 100-ms window that was slid across each of three analysis epochs (fixation, sample, and delay). Low (25- 55 Hz) and high gamma (65- 100 Hz) activity were investigated separately due to evidence that they may be functionally distinct. Within each epoch, recording sites in the PPC and PFC were classified into groups according to the similarity of their power t-f plots derived by a K-means clustering algorithm. From the power-based site groups, the corresponding coherence and WGC were analyzed. This classification procedure uncovered spatial, temporal, and frequency dynamics of FPN involvement in WM and other co-occurring processes, such as sensory and target related processes. These processes were distinguishable by rule and performance accuracy across all three spectral measures- power, coherence, and WGC. Location and Identity rule were distinguishable by the low and high-gamma range. / Includes bibliography. / Dissertation (Ph.D.)--Florida Atlantic University, 2014. / FAU Electronic Theses and Dissertations Collection

Visuospatial working memory (VSWM) comprises specialised subsystems devoted to storage of visual features and spatial locations. Recently, research has been focused on understanding feature binding in memory and how bound objects are temporarily held in working memory. In the current thesis we have addressed two broad questions: What is the nature of bound visual representations in working memory? Is there a specific deficit in binding in individuals with learning difficulties? In Study 1, young adults were required to recall locations, objects and object-location bindings under visual or spatial concurrent task conditions. A clear double dissociation pattern was observed: movement discrimination mainly disrupted location memory, whereas colour discrimination mainly disrupted object memory. Such selective interference was also observed for object-location memory, suggesting that bound object representations depend on the updating of specific feature information. In Study 2, two groups of children with specific learning disabilities dyslexia and nonverbal learning disability (NLD) were compared to typically developing children in three tasks that required forward and backward recall of locations, colours, and colour-location bindings. Only children with NLD have impairments in memory for locations and colours, especially in backward recall of locations, and there were no group differences for the colour-location binding task. The patterns seen in recall of locations and colours separately were no longer present when these features had to be recalled together, suggesting the specificity of binding processes. Finally, in Study 3, two groups of children at-risk of learning disabilities (verbal and nonverbal) were compared to typically developing children in VSWM for colours, shapes, and shape-colour bindings. It was observed that memory for shape-colour binding is impaired in both groups at risk of learning disabilities, whereas memory for either shapes or colours are spared. This provides further support that problems in memory binding may be widespread across different populations with learning difficulties and atypical development. In summary, taken together, our results are in line with an associative view of binding, i.e., bound object representation results from associative links between different types of features. VSWM seems to operate on both feature- and object-level information. / A memória de trabalho visuoespacial possui subsistemas especializados na retenção temporária de características visuais e localizações espaciais. Recentemente, diversas pesquisas procuram elucidar os mecanismos de integração (ou binding) de características na memória e como objetos integrados são temporariamente armazenados. Nesta tese, abordamos duas questões amplas: Qual a natureza de representações integradas na memória de trabalho? Há um déficit específico na integração de informações em indivíduos com dificuldades de aprendizagem? No Estudo 1, adultos jovens (estudantes universitários) realizaram tarefas de recordar localizações, objetos e conjunções objeto-localização em diferentes condições experimentais de interferência, que poderia ser uma tarefa concorrente visual ou espacial. Uma clara dissociação dupla foi observada: a discriminação de movimento dificultou a recordação de localizações e a discriminação de cores interferiu na recordação dos objetos. Tal interferência seletiva também foi observada na memória para conjunção objeto-localização, indicando que representações integradas dependem da atualização de traços de memória específicos. No Estudo 2, crianças com transtornos específicos de aprendizagem dislexia e transtorno de aprendizagem não-verbal (TANV) foram comparadas a crianças com desenvolvimento típico em três tarefas que exigiam a recordação em ordem direta e inversa de sequências de localizações, cores e conjunções cores-localizações. Crianças com TANV apresentaram déficits de memória para localizações e cores, especialmente quando as localizações deveriam ser recordadas em ordem inversa, e não houve diferenças entre grupos na tarefa de cor-localização. Os padrões observados na recordação de cores e localizações em separado não foram observados quando essas informações deveriam ser recordadas de maneira integrada, sugerindo a especificidade de processos de integração de características. Por fim, no Estudo 3, dois grupos de crianças em risco de transtornos de aprendizagem (verbal e não-verbal) foram comparadas a crianças em desenvolvimento típico em tarefas que exigiam memória para cores, formas e conjunções forma-cor. Foi observado que ambos os grupos com dificuldades de aprendizagem apresentaram um déficit de memória para a conjunção forma-cor, com memória preservada para cores e formas separadamente. Isso traz evidências adicionais que problemas de memória para conjunções podem ser generalizados para diversas populações com dificuldades de aprendizagem e desenvolvimento atípico. Resumindo, nosso conjunto de resultados estão de acordo com uma perspectiva associativa da conjunção ou binding, isto é, representações integradas resultam de ligações associativas entre diferentes tipos de traços ativados. A memória de trabalho visuoespacial parece funcionar com informações de ambos os níveis características básicas e objetos integrados.

Negation is a common property of languages, in that there are few languages, if any, that lack means to revert the truth-value of a statement. A challenge to cross-lingual studies of negation lies in the fact that languages encode and use it in different ways. Although this variation has been extensively researched in linguistics, little has been done in automated language processing. In particular, we lack computational models of processing negation that can be generalized across language. We even lack knowledge of what the development of such models would require. These models however exist and can be built by means of existing cross-lingual resources, even when annotated data for a language other than English is not available. This thesis shows this in the context of detecting string-level negation scope, i.e. the set of tokens in a sentence whose meaning is affected by a negation marker (e.g. 'not'). Our contribution has two parts. First, we investigate the scenario where annotated training data is available. We show that Bi-directional Long Short Term Memory (BiLSTM) networks are state-of-the-art models whose features can be generalized across language. We also show that these models suffer from genre effects and that for most of the corpora we have experimented with, high performance is simply an artifact of the annotation styles, where negation scope is often a span of text delimited by punctuation. Second, we investigate the scenario where annotated data is available in only one language, experimenting with model transfer. To test our approach, we first build NEGPAR, a parallel corpus annotated for negation, where pre-existing annotations on English sentences have been edited and extended to Chinese translations. We then show that transferring a model for negation scope detection across languages is possible by means of structured neural models where negation scope is detected on top of a cross-linguistically consistent representation, Universal Dependencies. On the other hand, we found cross-lingual lexical information only to help very little with performance. Finally, error analysis shows that performance is better when a negation marker is in the same dependency substructure as its scope and that some of the phenomena related to negation scope requiring lexical knowledge are still not captured correctly. In the conclusions, we tie up the contributions of this thesis and we point future work towards representing negation scope across languages at the level of logical form as well.

No description available.

Sequences are ubiquitous in the domain of biology. One of the current best machine learning techniques for analysing sequences is the long short-term memory (LSTM) network. Owing to significant barriers to adoption in biology, focussed efforts are required to realize the use of LSTMs in practice. Thus, the aim of this work is to improve the state of LSTMs for biology, and we focus on biological tasks pertaining to physiological signals, peripheral neural signals, and molecules. This goal drives the three subplots in this thesis: biological applications, visualizations, and extensions. We start by demonstrating the utility of LSTMs for biological applications. On two new physiological-signal datasets, LSTMs were found to outperform hidden Markov models. LSTM-based models, implemented by other researchers, also constituted the majority of the best performing approaches on publicly available medical datasets. However, even if these models achieve the best performance on such datasets, their adoption will be limited if they fail to indicate when they are likely mistaken. Thus, we demonstrate on medical data that it is straightforward to use LSTMs in a Bayesian framework via dropout, providing model predictions with corresponding uncertainty estimates. Another dataset used to show the utility of LSTMs is a novel collection of peripheral neural signals. Manual labelling of this dataset is prohibitively expensive, and as a remedy, we propose a sequence-to-sequence model regularized by Wasserstein adversarial networks. The results indicate that the proposed model is able to infer which actions a subject performed based on its peripheral neural signals with reasonable accuracy. As these LSTMs achieve state-of-the-art performance on many biological datasets, one of the main concerns for their practical adoption is their interpretability. We explore various visualization techniques for LSTMs applied to continuous-valued medical time series and find that learning a mask to optimally delete information in the input provides useful interpretations. Furthermore, we find that the input features looked for by the LSTM align well with medical theory. For many applications, extensions of the LSTM can provide enhanced suitability. One such application is drug discovery -- another important aspect of biology. Deep learning can aid drug discovery by means of generative models, but they often produce invalid molecules due to their complex discrete structures. As a solution, we propose a version of active learning that leverages the sequential nature of the LSTM along with its Bayesian capabilities. This approach enables efficient learning of the grammar that governs the generation of discrete-valued sequences such as molecules. Efficiency is achieved by reducing the search space from one over sequences to one over the set of possible elements at each time step -- a much smaller space. Having demonstrated the suitability of LSTMs for biological applications, we seek a hardware efficient implementation. Given the success of the gated recurrent unit (GRU), which has two gates, a natural question is whether any of the LSTM gates are redundant. Research has shown that the forget gate is one of the most important gates in the LSTM. Hence, we propose a forget-gate-only version of the LSTM -- the JANET -- which outperforms both the LSTM and some of the best contemporary models on benchmark datasets, while also reducing computational cost.

Orientador: Anna Diva Plasencia Lotufo / Abstract: Many environmental, economic and societal factors are leading fire brigades to be increasingly solicited, and, as a result, they face an ever-increasing number of interventions, most of the time on constant resource. On the other hand, these interventions are directly related to human activity, which itself is predictable: swimming pool drownings occur in summer while road accidents due to ice storms occur in winter. One solution to improve the response of firefighters on constant resource is therefore to predict their workload, i.e., their number of interventions per hour, based on explanatory variables conditioning human activity. The present work aims to develop three models that are compared to determine if they can predict the firefighters' response load in a reasonable way. The tools chosen are the most representative from their respective categories in Machine Learning, such as XGBoost having as core a decision tree, a classic method such as Multi-Layer Perceptron and a more advanced algorithm like Long Short-Term Memory both with neurons as a base. The entire process is detailed, from data collection to obtaining the predictions. The results obtained prove a reasonable quality prediction that can be improved by data science techniques such as feature selection and adjustment of hyperparameters. / Resumo: Muitos fatores ambientais, econômicos e sociais estão levando as brigadas de incêndio a serem cada vez mais solicitadas e, como consequência, enfrentam um número cada vez maior de intervenções, na maioria das vezes com recursos constantes. Por outro lado, essas intervenções estão diretamente relacionadas à atividade humana, o que é previsível: os afogamentos em piscina ocorrem no verão, enquanto os acidentes de tráfego, devido a tempestades de gelo, ocorrem no inverno. Uma solução para melhorar a resposta dos bombeiros com recursos constantes é prever sua carga de trabalho, isto é, seu número de intervenções por hora, com base em variáveis explicativas que condicionam a atividade humana. O presente trabalho visa desenvolver três modelos que são comparados para determinar se eles podem prever a carga de respostas dos bombeiros de uma maneira razoável. As ferramentas escolhidas são as mais representativas de suas respectivas categorias em Machine Learning, como o XGBoost que tem como núcleo uma árvore de decisão, um método clássico como o Multi-Layer Perceptron e um algoritmo mais avançado como Long Short-Term Memory ambos com neurônios como base. Todo o processo é detalhado, desde a coleta de dados até a obtenção de previsões. Os resultados obtidos demonstram uma previsão de qualidade razoável que pode ser melhorada por técnicas de ciência de dados, como seleção de características e ajuste de hiperparâmetros. / Mestre

Recognizing human activities using deep learning methods has significance in many fields such as sports, motion tracking, surveillance, healthcare and robotics. Inertial sensors comprising of accelerometers and gyroscopes are commonly used for sensor based HAR. In this study, a Bidirectional Long Short-Term Memory (BLSTM) approach is explored for human activity recognition and classification for closely related activities on a body worn inertial sensor data that is provided by the UTD-MHAD dataset. The BLSTM model of this study could achieve an overall accuracy of 98.05% for 15 different activities and 90.87% for 27 different activities performed by 8 persons with 4 trials per activity per person. A comparison of this BLSTM model is made with the Unidirectional LSTM model. It is observed that there is a significant improvement in the accuracy for recognition of all 27 activities in the case of BLSTM than LSTM.

Despite years of research supporting a link between sleep and cognitive functioning, little research has focused on the effect of sleep on cognitive factors in the college population. According to the trace reactivation hypothesis, sleep plays a crucial role in memory consolidation and learning. Cognitivism further indicates that the processing of information is limited when sleep is restricted. The purpose of this quantitative study was to examine the effect of sleep on vigilance, short-term memory, and learning; an additional goal was to test whether these variables influence grade point average (GPA). Sleep and GPA data were collected using self-report questionnaires. Vigilance was measured using the Digit Vigilance Test (DVT). Short-term memory and learning were measured using subtests of the Wide Range Assessment of Memory and Learning, 2nd Edition (WRAML 2). Data were collected from 20 undergraduate students to detect an effect size of .90 with a power of at least .80, one tailed p < .05. Test assumptions were satisfied prior to analysis using SPSS. Independent samples t tests were used to compare the difference between adequate and inadequate sleep on measures of vigilance, short-term memory, and learning. Results indicated that inadequate sleep had a significant effect on short-term memory and learning but not on vigilance. A multiple linear regression analysis revealed that sleep, vigilance, short-term memory, and learning were not significant predictors of GPA. The small sample size may have limited the study conclusion. This study may help students, educators, and school administrators to improve academic performance as well as develop effective learning strategies.

While a growing body of research indicates that older adults typically perform more poorly on many types of memory tasks than do younger adults, relatively little research has addressed the question of whether this trend continues unchanged into the late ninth and tenth decades of life. Such decrements in memory have been reported as linear declines from early adulthood up until about 80 years of age. Questions arise as to whether such memory declines slow or accelerate in very advanced aging, and to what extent differences are due to aging, per se, or variables that intervene between age and memory.To address these two questions, six memory types - verbal recall, nonverbal recall, short-term memory, working memory, face recognition, and prospective memory - were examined using both cross-sectional and longitudinal methodologies. The six types of memory and the influence of verbal processing speed, nonverbal processing speed, and intelligence were examined in mixed-gender groups of 20 - 40 (n = 40, M = 30.7, SD = 5.52), 50 - 70 (n = 44, M = 59.2, SD = 4.94), and 85+ year olds (n = 42, M = 87.8, SD = 2.43), at two points, the second occurring two years after the first. Each participant completed tests of word recall, geometric shapes recall, short-term memory (digit span), working memory (letter-number sequencing), face recognition, and prospective memory. Additionally, there were two processing speed tasks (Identical Pictures and Finding As), and the National Adult Reading Test of verbal fluency was used to estimate intelligence. The Mini-Mental State Examination and the Beck Depression Inventory (BDI-II) were used to screen for dementia and depression, respectively.At Time 1 testing the 85+ participants showed declines in all memory types (compared to the 20 - 40 year olds). Nonverbal recall (66.2% lower than the young group), working memory (46.2%), verbal recall (45%), and prospective memory (38.2%) produced the largest differences, short-term memory (12.3%) and face recognition (14.7%) the least. Two years later, the 85+ years old participants had shown further declines, relative to the 20 - 40 years group. Nonverbal recall (72.3% lower than the young group), prospective memory (63.2%), working memory (55.3%), and verbal recall (54.7%) continued to produce the largest decrements, with short-term memory (18.9%) and face recognition (19.8%) the least. The results for the young and middle participants did not change appreciably between Time 1 and Time 2. The difference between unadjusted scores and scores adjusted for intelligence, verbal processing speed, and nonverbal processing speed, increased markedly between Time 1 and Time 2 testing for the oldest-old participants.These findings support the view that while memory declines may be approximately linear from age 20 to 80 years, there is a sharp decline in most types of memory after the age of 85 years, recall and working memory suffering the most. Intelligence and processing speed have an effect on some types of memory, but age is by far the largest contributor to memory decline. Furthermore, as expected, all memory types declined over the two-year period, with prospective memory, verbal recall, nonverbal recall, and working memory showing the greatest declines. Short-term memory and face recognition declined at a noticeably slower rate.