The mammalian neocortex is integral to all aspects of cognition, in particular perception across all sensory modalities. Whether computational principles can be identified that would explain why the cortex is so versatile and capable of adapting to various inputs is not clear. One well-known hypothesis is that the cortex implements a generative model, actively synthesising internal explanations of the sensory input. This ‘analysis by synthesis’ could be instantiated in the top-down connections in the hierarchy of cortical regions, and allow the cortex to evaluate its internal model and thus learn good representations of sensory input over time. Few computational models however exist that implement these principles. In this thesis, we investigate the deep Boltzmann machine (DBM) as a model of analysis by synthesis in the cortex, and demonstrate how three distinct perceptual phenomena can be interpreted in this light: visual hallucinations, bistable perception, and object-based attention. A common thread is that in all cases, the internally synthesised explanations go beyond, or deviate from, what is in the visual input. The DBM was recently introduced in machine learning, but combines several properties of interest for biological application. It constitutes a hierarchical generative model and carries both the semantics of a connectionist neural network and a probabilistic model. Thus, we can consider neuronal mechanisms but also (approximate) probabilistic inference, which has been proposed to underlie cortical processing, and contribute to the ongoing discussion concerning probabilistic or Bayesian models of cognition. Concretely, making use of the model’s capability to synthesise internal representations of sensory input, we model complex visual hallucinations resulting from loss of vision in Charles Bonnet syndrome.We demonstrate that homeostatic regulation of neuronal firing could be the underlying cause, reproduce various aspects of the syndrome, and examine a role for the neuromodulator acetylcholine. Next, we relate bistable perception to approximate, sampling-based probabilistic inference, and show how neuronal adaptation can be incorporated by providing a biological interpretation for a recently developed sampling algorithm. Finally, we explore how analysis by synthesis could be related to attentional feedback processing, employing the generative aspect of the DBM to implement a form of object-based attention. We thus present a model that uniquely combines several computational principles (sampling, neural processing, unsupervised learning) and is general enough to uniquely address a range of distinct perceptual phenomena. The connection to machine learning ensures theoretical grounding and practical evaluation of the underlying principles. Our results lend further credence to the hypothesis of a generative model in the brain, and promise fruitful interaction between neuroscience and Deep Learning approaches.

While railroad trespassing is a dangerous activity with significant security and safety risks, regular patrolling of potential trespassing sites is infeasible due to exceedingly high resource demands and personnel costs. There is thus a need to design an automated trespass detection and early warning prediction tool leveraging state-of-the-art machine learning techniques. Leveraging video surveillance through security cameras, this thesis designs a novel approach called ARTS (Automated Railway Trespassing detection System) that tackles the problem of detecting trespassing activity. In particular, we adopt a CNN-based deep learning architecture (Faster-RCNN) as the core component of our solution. However, these deep learning-based methods, while effective, are known to be computationally expensive and time consuming, especially when applied to a large amount of surveillance data. Given the sparsity of railroad trespassing activity, we design a dual-stage deep learning architecture composed of an inexpensive prefiltering stage for activity detection followed by a high fidelity trespass detection stage for robust classification. The former is responsible for filtering out frames that show little to no activity, this way reducing the amount of data to be processed by the later more compute-intensive stage which adopts state-of-the-art Faster-RCNN to ensure effective classification of trespassing activity. The resulting dual-stage architecture ARTS represents a flexible solution capable of trading-off performance and computational time. We demonstrate the efficacy of our approach on a public domain surveillance dataset.

In this work, we develop an end-to-end neural network-based computer vision system to automatically identify where each person within a 2-D image of a school classroom is looking (“gaze following�), as well as who she/he is looking at. Automatic gaze following could help facilitate data-mining of large datasets of classroom observation videos that are collected routinely in schools around the world in order to understand social interactions between teachers and students. Our network is based on the architecture by Recasens, et al. (2015) but is extended to (1) predict not only where, but who the person is looking at; and (2) predict whether each person is looking at a target inside or outside the image. Since our focus is on classroom observation videos, we collect gaze dataset (48,907 gaze annotations over 2,263 classroom images) for students and teachers in classrooms. Results of our experiments indicate that the proposed neural network can estimate the gaze target - either the spatial location or the face of a person - with substantially higher accuracy compared to several baselines.

Submitted by Maicon Juliano Schmidt (maicons) on 2015-06-12T19:13:14Z No. of bitstreams: 1 Ingo Jost.pdf: 1217467 bytes, checksum: bf67cd6724b1cd182a12a3cd7b5af1eb (MD5) / Made available in DSpace on 2015-06-12T19:13:14Z (GMT). No. of bitstreams: 1 Ingo Jost.pdf: 1217467 bytes, checksum: bf67cd6724b1cd182a12a3cd7b5af1eb (MD5) Previous issue date: 2015-02-26 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Deep Learning é uma sub-área de Aprendizado de Máquina que tem obtido resultados sa- tisfatórios em várias áreas de aplicação, implementada por diferentes algoritmos, como Stacked Auto-encoders ou Deep Belief Networks. Este trabalho propõe uma modelagem que aplica uma implementação de um classificador que aborda técnicas de Deep Learning em Mineração de Opiniões, área que tem sido alvo de constantes estudos, dada a necessidade das corporações buscarem a compreensão que clientes possuem de seus produtos ou serviços. O favorecimento do crescimento de Mineração de Opiniões também se dá pelo ambiente colaborativo da Web 2.0, em que várias ferramentas propiciam a emissão de opiniões. Os dados utilizados passaram por um refinamento na etapa de pré-processamento com o intuito de aplicar Deep Learning, da qual uma das principais atribuições é a seleção de características, em dados refinados em vez de dados mais brutos. A promissora tecnologia de Deep Learning combinada com a estratégia de refinamento demonstrou nos experimentos a obtenção de resultados competitivos com outros estudos relacionados e abrem perspectiva de extensão deste trabalho. / Deep Learning is a Machine Learning’s sub-area that have achieved satisfactory results in different application areas, implemented by different algorithms, such as Stacked Auto- encoders or Deep Belief Networks. This work proposes a research that applies a classifier that implements Deep Learning concepts in Opinion Mining, area has been approached by con- stant researches, due the need of corporations seeking the understanding that customers have of your products or services. The Opinion Mining’s growth is favored also by the collaborative Web 2.0 environment, where multiple tools provide issuing opinions. The data used for exper- iments were refined in preprocessing step in order to apply Deep Learning, which it one of the main tasks the feature selection, in refined data, instead of applying Deep Learning in more raw data. The refinement strategy combined with the promising technology of Deep Learning has demonstrated in preliminary experiments the achievement of competitive results with other studies and opens the perspective for extension of this work.

Background: This research was designed to explore the experience and understanding of counsellors' and psychotherapists' engagement with metaphors in the therapeutic process. The aim is to reflect on the experience of therapists involved in therapeutic metaphors from differing perspectives. Methodology: In a heuristic study a group of seven therapists (counsellors and psychotherapists) shared their use of metaphors in their therapy practice. Data were collected through an informal conversational interview that supported the participants to share their experiences in a natural dialogue. Findings: The experience of using metaphor in therapy appears to involve a multi-faceted web of generation, construction and development between the therapist and client. Various levels of depth of metaphor in therapy were identified along with links to transferential and cultural issues. Metaphors of hope also appear to be potentially important. Discussion: The findings suggest that the use of metaphors in therapy is pervasive. Metaphors that reflect an empathic connection and encounter between therapist and client were identified. Dualistic thinking around the origination of metaphors in therapy is challenged and the concept of co-creation and the mutual development of moving metaphors is discussed. Environmental and cultural influences are considered alongside transferential aspects. Conclusion: It appears that the use of metaphor in therapy is pervasive and offers an opportunity for therapeutic change. The consideration of the construction of metaphors and their mutual development may be useful for therapists to consider. This research highlights the need for more investigation with regard to client perspectives, the environmental impacts on metaphors in therapy and who the therapist and client stand for metaphorically for each other.

Dans le but d'étudier certains aspects dynamiques de la convection profonde atmosphérique, nous avons développé un modèle bidimensionnel axé sur le développement d'une cellule orageuse simple. Ce modèle considère des éléments de volume, où nous faisons l'hypothèse que les différents champs thermodynamiques sont homogènes. Ces volumes sont fixes dans l'espace et le temps et sont traversés par les flots d'air humide de sorte que leurs contenus varient au cours du temps. Durant ces évolutions l'eau subit des changements de phases. Ces phénomènes, simultanés dans la nature, sont représentés par un mécanisme à étapes successives dans le modèle. Une première étape se déroule en système ouvert: l'air circule pendant un pas de temps d'intégration entre les éléments de volume; l'air conserve ses propriétés pendant le déplacement. Les deux étapes suivantes se produisent dans chaque élément de volume considéré alors en système fermé et isolé: une première étape d'homogénéisation de l'air en pression et ensuite en température et enfin une étape de restauration de l'équilibre des phases de l'eau compte tenu de la nouvelle répartition des constituants et de leur état thermodynamique. Cette discrétisation du mécanisme d'évolution du contenu des éléments de volume nous permet d'utiliser les lois de la thermodynamique classique dans des systèmes ouverts. Ce mécanisme mène à une équation thermodynamique originale. Les autres équations du modèle sont les équations de l'hydrodynamique classique, les équations de la quantité de mouvement et de continuité. Pour l'intégration des équations, nous avons utilisé une méthode de filtrage numérique basée sur les transformations de Laplace, due à P. Lynch (1984) et adaptée à l'intégration par J. Van Isacker (1985). Au niveau du calcul, les champs de masse, de pression et des quantités de mouvement sont adaptés aux échanges de matières entre éléments de volume voisins à l'aide du processus d'intégration. Les équilibrages de phases interviennent comme ajustement du résultat de l'intégration. Ils modifient le défaut de balance hydrostatique qui sera minimisé au cours du pas d'intégration suivant grâce au filtrage de la méthode numérique. Les simulations réalisées à l'aide du modèle restituent de manière raisonnable les caractéristiques essentielles de la convection profonde atmosphérique. Nous avons utilisé le modèle pour étudier le développement d'un orage de masse d'air de manière plus approfondie. Ainsi, le développement initial, la croissance de la cellule convective, la formation de vortex ont été corrélés avec la structure de la flottabilité dans l'étude des mécanismes mis en oeuvre. Nous avons examiné les déplacements horizontaux et les accélérations verticales en termes de mélanges de masses d'air et des changements de phases qu'ils induisent. Dans l'étude de l'évolution des différentes formes d'énergie, cinétique, potentielle et interne et de leurs conversions, nous avons recherché les contributions dominantes à leurs variations et montré les rôles prépondérants joués par les processus de changement de phase et d'homogénéisation locale de la pression dans la variation de l'énergie interne. Dans l'examen de l'effet dynamique de la convection profonde sur le courant moyen, nous avons montré que, dans certains cas, nous avons non seulement transfert vertical d'énergie cinétique mais également création d'énergie cinétique du courant moyen. Le cumulonimbus peut dans certains cas agir comme moteur pour les mouvements atmosphériques à plus grande échelle.

Background: Patients with Parkinson’s disease (PD) and Essential tremor (ET) who havebeen treated with Deep Brain Stimulation (DBS) generally experience a positive effect,particularly regarding the motor symptoms. However, the patients’ communication skillsare often negatively affected and the assessment instrument currently used withinneurological clinical care is not sufficiently sensitive to assess these patients’ speechclarity, voice and intelligibility satisfactorily.Aim: This study’s purpose was to develop a prototype assessment tool for speech clarity,intelligibility and voice, with speech and language pathology (SLP) validity, that isadaptable to a neurological clinical care setting.Method: The assessment tool was designed using general design methodology. Aprototype was constructed and tested on speech samples of read text for reliability. ThreeSLP’s, three DBS nurses and three naive listeners (NL) were represented in the test group.Levels of agreement were calculated using Percent Close Agreement, PCA.Results: The results indicate a relatively high level of agreement between the groups, inparticular the SLP group and the DBS group (μ: 0.82, 0.79, and 0.74).Conclusion: The results demonstrate the need for an assessment tool with SLPcompetence to assess speech clarity, intelligibility and voice within neurological clinicalcare. The assessment tool was shown to be a useful and adequate prototype that can easilyevolve into a truly useful and versatile perceptual speech assessment tool. The results ofthis study should be treated cautiously, considering the test groups’ modest size. / Bakgrund: Patienter med Parkinsons sjukdom (PD) och patienter med Essentiell tremor(ET) som behandlats med Deep Brain Stimulation (DBS) upplever i allmänhet en positiveffekt, framför allt gällande de motoriska symtomen. Emellertid påverkas oftapatienternas kommunikativa färdigheter negativt. De bedömingsmaterial som användsinom den kliniska nerurologiska vården; UPDRS/ETRS är alltför trubbiga för att kunna geen tillfredsställande bild av patientens tal, röst och förståelighet.Mål: Skapa ett bedömningsverktyg för tal, förståelighet och röst med logopedisk validitet,och som kan användas inom den kliniska neurologiska verksamheten i samband medDBS-behandling.Metod: Bedömningsverktyget designades enligt generell designmetodik. En prototypskapades och testades för reliabilitet på röstexempel av en läst text. Tre logopeder, treDBS-sköterskor och tre naiva lyssnare deltog i testningen. Grad av samstämmighetberäknades med Percent Close Agreement, PCA.Resultat: Resultaten indikerar en relativt hög grad av samstämmighet mellan grupperna(μ: 0.82, 0.79, respektive 0.74). Logopederna bedömde nästan alla röstexempel sompatienter i behov av logopedhjälp. DBS-gruppen och gruppen med naiva lyssnarebedömde ett mindre antal ha behov av logoped.Slutsats: Resultaten belyser behovet av ett bedömningsverktyg med logopedisk validitetför bedöming av tal, förståelighet och röst inom den kliniska neurologiska verksamheten.Bedömingsverktyget som framtagits i denna studie är en användbar och adekvat prototypsom enkelt skulle kunna utvecklas till ett verkligt användbart och mångsidigt perceptuelltbedömningsmaterial. Dock ska resultaten i denna studie tolkas en smula försiktigt medtanke på de låga antalet deltagare.

Thesis (M.Sc.)--Memorial University of Newfoundland, 2002. / Includes bibliographical references.

Bridge bents (deep beams) in the State of Texas have experienced diagonal cracking problems with increasing frequency. These field related issues, taken in combination with discrepancies that exist between design provisions for strut and tie modeling (STM), were the impetus for the funding of the current project. The overall objective of the project was to develop safe and consistent design guidelines in regard to both the strength and serviceability of deep beams. In order to accomplish this research objective and related tasks, a database of 868 deep beam tests was assembled from previous research. Inadvertently, many of the beams in this database were considerably smaller, did not contain sufficient information, or contained very little shear reinforcement. As a result, filtering criteria were used to remove 724 tests from the database. The criteria were chosen to consider only beams that represent bent caps designed in the field. In addition to the 144 tests that remained in the database, 34 tests were conducted as part of the current experimental program resulting in 178 total tests available for evaluation purposes. Two additional tests were conducted on beams without shear reinforcement, thus they did not meet the filtering criteria. However, the results from these tests provided valuable information regarding deep beam behavior. Beams that were fabricated and tested as part of the current experimental program ranged in size from, 36"x48", 21"x75", 21"x42", and 21"x23". These tests represent some of the largest deep beam shear tests ever conducted. STM details that were investigated included: (i) the influence that triaxial confinement of the load or support plate has on strength and serviceability performance; and (ii) the influence that multiple stirrup legs distributed across the web has on strength and serviceability performance. Based on the findings of the experimental and analytical program, a new strut-and-tie modeling procedure was proposed for the design of deep beam regions. The procedure is based on an explicitly defined single-panel truss model with non-hydrostatic nodes. An important aspect of the new STM design methodology is that it was comprehensively derived based on all the stress checks that constitute an STM design. Thus, the new method considers every facet of a STM design. The newly proposed STM procedure is simple, more accurate, and more conservative in comparison with the ACI 318-08 and AASHTO LRFD (2008) STM design provisions. As such, the implementation of the new design provisions into ACI 318 and AASHTO LRFD is recommended. / text

Silicon carbide (SiC) is currently under development for high power bipolar devices such as insulated gate bipolar transistors (IGBTs). A major issue for these devices is the charge carrier lifetime, which, in the absence of structural defects such as dislocations, is influenced by point defects and their associated deep levels. These defects provide energy levels within the bandgap and may act as either recombination or trapping centers, depending on whether they interact with both conduction and valence band or only one of the two bands. Of all deep levels know in 4H-SiC, the intrinsic carbon vacancy related Z1/2 is the most problematic since it is a very effective recombination center which is unavoidably formed during growth. Its concentration in the epilayer can be decreased for the production of high voltage devices by injecting interstitial carbon, for example by oxidation, which, however, results in the formation of other new deep levels. Apart from intrinsic crystal flaws, extrinsic defects such as transition metals may also produce deep levels within the bandgap, which in literature have so far only been shown to produce trapping effects. The focus of the thesis is the transient electrical and optical characterization of deep levels in SiC and their influence on the carrier lifetime. For this purpose, deep level transient spectroscopy (DLTS) and minority carrier transient spectroscopy (MCTS) variations were used in combination with time-resolved photoluminescence (TRPL). Paper 1 deals with a lifetime limiting deep level related to Fe-incorporation in n-type 4H-SiC during growth and papers 2 and 3 focus on identifying the main intrinsic recombination center in p-type 4H-SiC. In paper 4, the details of the charge carrier capture behavior of the deeper donor levels of the carbon vacancy, EH6/7, are investigated. Paper 5 deals with trapping effects created by unwanted incorporation of high amounts of boron during growth of n-type 4H-SiC which hinders the measurement of the carrier lifetime by room temperature TRPL. Finally, paper 6 is concerned with the characterization of oxidation-induced deep levels created in n- and p-type 4H- and 6H-SiC as a side-product of lifetime improvement by oxidation. In paper 1, the appearance of a new recombination center in n-type 4H-SiC, the RB1 level is discussed and the material is analyzed using room temperature TRPL, DLTS and pnjunction DLTS. The level appears to originate from a reactor contamination with Fe, a transition metal that generally leads to the formation of several trapping centers in the bandgap. Here it is found that under specific circumstances beneficial to the growth of high-quality material with a low Z1/2 concentration, the Fe incorporation also creates an additional recombination center capable of limiting the carrier lifetime. In paper 2, all deep levels found in p-type 4H-SiC grown at Linköping University which are accessible by DLTS and MCTS are investigated with regard to their efficiency as recombination centers. We find that none of the detectable levels is able to reduce carrier lifetime in p-type significantly, which points to the lifetime killer being located in the top half of the bandgap and having a large hole to electron capture cross section ratio (such as Z1/2, which is found in n-type material), making it undetectable by DLTS and MCTS. Paper 3 compares carrier lifetimes measured by temperature-dependent TRPL measurements in n- and p-type 4H-SiC and it is shown that the lifetime development over a large temperature range (77 - 1000 K) is similar in both types. This is interpreted as a further indication that the carbon vacancy related Z1/2 level is the main lifetime killer in p-type. In paper 4, the hole and electron capture cross sections of the near midgap deep levels EH6/7 are characterized. Both levels are capable of rapid electron capture but have only small hole capture rates, making them insignificant as recombination centers, despite their advantageous position near midgap. Minority carrier trapping by boron, which is both a p-type dopant and an unavoidable contaminant in 4H-SiC grown by CVD, is investigated in paper 5. Since even the shallow boron acceptor levels are relatively deep in the bandgap, minority trap and-release effects are detectable in room-temperature TRPL measurements. In case a high density of boron exists in n-type 4H-SiC, for example leached out from damaged graphite reactor parts during growth, we demonstrate that these trapping effects may be misinterpreted in room temperature TRPL measurements as a long free carrier lifetime. Paper 6 uses MCTS, DLTS, and room temperature TRPL to characterize the oxidation induced deep levels ON1 and ON2 in n- and p-type 4H- and their counterparts OS1-OS3 in 6H-SiC. The levels are found to all be positive-U, coupled two-levels defects which trap electrons efficiently but exhibit very inefficient hole capture once the defect is fully occupied by electrons. It is shown that these levels are incapable of significantly influencing carrier lifetime in epilayers which underwent high temperature lifetime enhancement oxidations. Due to their high density after oxidation and their high thermal stability they may, however, act to compensate n-type doping in low-doped material.