Dissertations / Theses on the topic 'Perturbation training'

Thesis: S.M., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2015.
Cataloged from PDF version of thesis.
Includes bibliographical references (pages 63-67).
Today, robots are often deployed to work separately from people. Combining the strengths of humans and robots, however, can potentially lead to a stronger joint team. To have fluid human-robot collaboration, these teams must train to achieve high team performance and flexibility on new tasks. This requires a computational model that supports the human in learning and adapting to new situations. In this work, we design and evaluate a computational learning model that enables a human-robot team to co-develop joint strategies for performing novel tasks requiring coordination. The joint strategies are learned through "perturbation training," a human team-training strategy that requires practicing variations of a given task to help the team generalize to new variants of that task. Our Adaptive Perturbation Training (AdaPT) algorithm is a hybrid of transfer learning and reinforcement learning techniques and extends the Policy Reuse in Q-Learning (PRQL) algorithm to learn more quickly in new task variants. We empirically validate this advantage of AdaPT over PRQL through computational simulations. We then augment our algorithm AdaPT with a co-learning framework and a computational bi-directional communication protocol so that the robot can work with a person in live interactions. These three features constitute our human-robot perturbation training model. We conducted human subject experiments to show proof-of-concept that our model enables a robot to draw from its library of prior experiences in a way that leads to high team performance. We compare our algorithm with a standard reinforcement learning algorithm Q-learning and find that AdaPT-trained teams achieved significantly higher reward on novel test tasks than Q-learning teams. This indicates that the robot's algorithm, rather than just the human's experience of perturbations, is key to achieving high team performance. We also show that our algorithm does not sacrifice performance on the base task after training on perturbations. Finally, we demonstrate that human-robot training in a simulation environment using AdaPT produced effective team performance with an embodied robot partner.
by Ramya Ramakrishnan.
S.M.

This thesis explores stability and perturbation in learning situated in a counselling training culture that is shaped by humanistic-integrative core theory values. The case study on which the thesis is based focuses on a group of six students and two tutors on a Diploma in Counselling and Human Relations course. As an integral part of the training programme, the participants were asked to use a conceptual model imported from a psychotherapy context. This model was initially chosen for the purposes of evaluating its effectiveness as a learning tool for use in counselling training. During the study, the emphasis changed to an interest in understanding how use of this model affected the process of learning. The thesis is based on an in-depth analysis of the impact of this model on the learning process in the group. The case study is shaped by an `action research' (McCutcheon et al., 1987) model that both reflects the inquirer's position as a tutor and researcher in the counselling community investigated and the professional development dimension of the research. Data was gathered over a period of seven months, using individual and group interviews, and observations of participants' training practices. The research methodology stems from an interpretivist, or hermeneutic tradition of story telling. Within the qualitative parameters of this philosophical orientation, the data is constructed on the basis of a critical understanding of Wenger's (1998) situated perspective on learning. From this perspective, learning assumes a social, or relational location. The research suggests that counselling training can be usefully understood as a participatory process that includes both stability and perturbation, and this has implications for counselling training practices that are shaped by a core theoretical model.

Approximately one in three adults 65 years and older fall each year and these falls lead to a substantial number of serious injuries and deaths. Numerous interventions have been proposed for fall-prevention but the efficacy can vary, and may be due to the general nature of the interventions. Older adults may be able to improve their ability to recover from a postural perturbation through perturbation-based balance training (PBBT), similar to the way other motor skills can be improved through training. The purpose of the first study was to investigate the effects of age and fall risk on the efficacy of PBBT. Participants (young adults, older adults at low-risk of falling, older adults at high-risk of falling) completed PBBT on a moving platform three times a week for one month. Balance was quantified using the time to stabilization of the COP and normalized to platform displacement (nTTS), where a decrease in nTTS can be interpreted as an improvement in balance. A significant main effect of group revealed high-risk fallers had a significantly higher nTTS than young adults and a significant main effect of session revealed nTTS was significantly lower one week and one month post-training than before training. The purpose of the second study was to investigate the effect of training amount on the efficacy of PBBT in older adults. Ten healthy older adults completed PBBT either three times a week or five times a week for four weeks. Both training amounts were sufficient for significant improvements in nTTS one week after training. However, training five times a week was necessary for older adults to maintain improvements in nTTS one month post training. The purpose of the third study was to investigate the need for PBBT after strength training in order to improve balance in older adults. A torque-driven, three-segment, musculoskeletal model and forward dynamic simulations were used to address the hypothesis. Increasing joint strength was beneficial in recovering balance from a postural perturbation only after re-optimization of the torque activation. These results provide support for supplementing strength training fall prevention interventions in older adults with task-related practice.
Ph. D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Mechanical Engineering, 2011.
Cataloged from PDF version of thesis.
Includes bibliographical references (p. 149-156).
Rehabilitation of human motor function is an issue of the utmost significance, and the demand for the effective rehabilitation service is even growing with the graying of the population. Robotic technology has provided promising ways to assist recovery of the motor function of upper extremities. In contrast, current robotic therapy for lower extremities has shown inferior efficacy. In this thesis, the source of the limited efficacy of current robotic walking therapy is addressed. Essential mechanical components for robustly stable walking are identified as energy dissipation and proper compensation. Based on these essential components, design criteria of effective robotic walking therapy are suggested as foot-ground interaction and ankle actuation. A novel strategy of robot aided walking therapy reflecting the design criteria is proposed; dynamically entraining human gait with periodic ankle torque from a robot. Experiments with normal subjects and neurologically impaired subjects support the feasibility of the proposed rehabilitation strategy. The gait period of subjects entrain to the periodic mechanical perturbation with a measurable basin of entrainment, and the entrainment always accompanies phase-locking so that the mechanical perturbation assists propulsion. These observations are affected neither by auditory feedback nor by a distractor task for normal subjects, and consistently observed in impaired subjects. A highly simplified one degree of freedom walking model without supra-spinal control or an intrinsic self-sustaining neural oscillator (a rhythmic pattern generator) encapsulated the essence of these observations. This suggests that several prominent limit-cycle features of human walking may stem from peripheral mechanics mediated by simple afferent feedback without significant involvement of supra-spinal control or central pattern generator. The competence of the highly simplified model supports that the proposed entrainment therapy may be effective for a wide range of neurological impairments.
by Jooeun Ahn.
Ph.D.

Slip-induced fall related injuries are a serious public health issue among older adults leading to considerable mortality, morbidity, and immobility. Existing proactive exercise interventions have produced mixed results on the success of reducing fall accidents. A training intervention may be effective in reducing slip-induced falls, if it can help older adults to practice movements related to recovery responses. The purpose of this study was to evaluate two different training interventions using a moveable platform and virtual reality in order to improve reactive recovery in older adults. Thirty-six older adults were recruited and randomly assigned to three groups (moveable platform training, virtual reality training, and control). The training groups underwent three sessions including baseline slip, training, and transfer of training on a slippery surface. The control group underwent three similar sessions as the training groups, with the training session replaced with a normal walking session. Kinematic, kinetic, and EMG data were collected during all the sessions. The moveable platform training group was repeatedly exposed to simulated slips induced by anterior-posterior movement of a platform. The virtual reality training group was repeatedly exposed to perturbation induced by visual tilts in the virtual environment while walking on the treadmill. Various biomechanical and neuromuscular characteristics were identified to quantify the effects of training. The results indicated a beneficial effect of both training methods in improving recovery reactions in older adults via proactive and reactive adjustments. The reactive adjustments involved faster response to a slip perturbation mediated by reduced time for onset and peak muscle activation (specifically knee flexor), reduced knee and ankle coactivity, reduced time for peak knee, hip, and trunk angles, and angular velocity. The proactive adjustments involved an increased center-of-mass velocity and transitional acceleration of center-of-mass. The overall fall frequency was reduced in the training groups as compared to the control group through improvements in proactive and reactive responses.
Ph. D.

BACKGROUND: il morbo di Parkinson è una patologia neurodegenerativa caratterizzata da un ampio spettro di sintomi clinici; tra questi i deficit del cammino e del controllo posturale rappresentano la componente preminente dell’alterato quadro motorio tipico di questa patologia. Tra gli interventi non farmacologici il treadmill rappresenta un’opzione terapeutica dalla quale si stanno sviluppando nuovi approcci, quali l’aggiunta di destabilizzazioni posturali. OBIETTIVO: confrontare l’efficacia del training tramite treadmill perturbato, con il training convenzionale tramite treadmill, in soggetti affetti da morbo di Parkinson, per quanto concerne gli outcome del cammino e del controllo posturale. DISEGNO DI STUDIO: revisione sistematica di Randomized Controlled Trials, redatta secondo la checklist del PRISMA statement. FONTI DEI DATI: banche dati elettroniche indagate PubMed, CENTRAL, PEDro, Trip Medical Database, CINAHL. Ricerche aggiuntive sono state condotte tramite l’analisi delle citazioni bibliografiche di articoli trattanti i medesimi argomenti. La ricerca è stata condotta da Marzo 2020 al 15 Ottobre 2020. CRITERI DI ELEGGIBILITA’: sono stati consultati RCT che indagassero la comparazione tra il PTT e il CTT, in soggetti affetti da morbo di Parkinson, per quanto concerne gli outcome riguardanti il miglioramento del cammino e del controllo posturale. RISULTATI: 5 studi inclusi. L’analisi complessiva dei risultati non indica una superiore efficacia PTT, rispetto al CTT, nella riabilitazione di pazienti affetti da morbo di Parkinson. Entrambe le tipologie di intervento mostrano miglioramenti significativi relativamente al cammino, e miglioramenti non rilevanti riguardo al controllo posturale. Risultano però delle “tendenze positive” a favore dell’intervento con PTT per alcuni parametri sia del cammino, sia del controllo posturale. CONCLUSIONI: il riscontro di “tendenze positive” nel PTT fornisce uno spunto rilavante per la pratica clinica e per ricerche future.

O objetivo deste estudo foi avaliar o efeito do treinamento de força muscular dos membros inferiores na estabilidade postural de mulheres idosas. As participantes (n =21) eram saudáveis, com idade entre 60 e 75 anos (M = 64,4 anos), e foram designadas a um grupo de treinamento (TF) ou a um grupo controle (CO). Esses grupos foram comparados em estabilidade postural e força antes e após um programa de treinamento de força para o grupo TF. A estabilidade postural foi avaliada em posturas eretas estáticas com apoio unipodal e bipodal, e em situações em que a estabilidade postural bipodal foi perturbada de forma previsível ou imprevisível. A perturbação previsível foi produzida por um movimento voluntário, por meio da elevação rápida com as mãos, de cargas conhecidas: 1 Kg, 3 Kg ou 5 Kg. A perturbação imprevisível foi produzida pela alteração inesperada da carga de 3 Kg para a carga de 1 Kg ou de 5 Kg antes de sua elevação. Os resultados indicaram aumento da força muscular após o treinamento de força. Quanto à estabilidade postural, nãob foram observadas diferenças significativas entre os grupos após o treinamento para as tarefas de apoio bipodal e apoio unipodal. A estabilidade postural foi afetada principalmente, quando uma carga mais leve foi elevada na situação imprevisível, gerando maior deslocamento anterior e posterior do CP. Entretanto, não houve efeito do treinamento de força no deslocamento do CP. Após o treinamento de força, o grupo TF apresentou uma redução do tempo de deslocamento posterior do CP e uma redução na variabilidade de amplitude do CP após ajustes compensatórios. Estes resultados sugerem que o treinamento de força não afeta o equilíbrio de idosos em situações estáticas ou componentes de ajustes iniciais com a perturbação da postura, mas houve um efeito positivo nos componentes tardios de ajustes após a perturbação
The goal of this study was to investigate the effect of strength training of the lower limbs on postural stability of older women. Participants (n =21) were healthy, with ages ranging between 60 and 75 years (M=64.4 years); they were assigned to two groups: strength training (ST) and control (CO). These groups were compared for postural stability and muscular strength before and after a strength training program for the ST group. Postural stability was evaluated in uni and bipodal static stance, and in situations of predictable and unpredictable perturbation of balance. The predictable disturbance was produced through the action of lifting a known load (1 kg, 3 kg or 5 kg) with the hands. The unpredictable disturbance was produced by an unexpected alteration of the load of 3 kg to 1 kg or 5 kg right before lifting the load up. The results indicated increased muscle strength after strength training. Regarding postural stability, no significant difference was detected between groups for uni/bipodal static postures. Postural stability was most affected when a lighter weight was lifted in the unpredictable situation, leading to greater anterior and posterior center of pressure (CP) displacement. Strength training, however, had no effect on the initial CP displacement. After training, the ST group showed shorter time of posterior CP displacement and lower variability of CP range after compensatory adjustments. These results suggest that strength training for elderly individuals does not affect stability in static postures or the early component of adjustments to balance perturbation, but it has a positive effect on the late component of adjustments to balance perturbation

Die posturale Instabilität zählt zu den am meisten beeinträchtigenden Symptomen der Parkinson-Krankheit. Die Störung der motorischen Gleichgewichtskontrolle ist progressiv im Verlauf und weder durch medikamentöse noch durch operative Methoden zufriedenstellend einzudämmen. In der Bewegungstherapie werden häufig Gleichgewichtsübungen empfohlen, um ein Fortschreiten der körperlichen Einschränkungen zu verringern. Der aktuelle wissenschaftliche Stand lässt allerdings eine Einschätzung zur Effektivität von Gleichgewichtstraining bei Parkinsonpatienten kaum zu. Dies ist u. a. auf einen Mangel an geeigneten Testverfahren zur Beurteilung der posturalen Instabilität zurückzuführen. In der vorliegenden Untersuchung wurden die Auswirkungen eines 12-wöchigen Gleichgewichtstrainings bei Parkinsonpatienten auf die posturale Stabilität in einem umfassenden Testdesign, bestehend aus alltagsmotorischen, biomechanischen und subjektiven Testverfahren evaluiert. In nahezu allen eingesetzten Testverfahren zeigte sich ein mehr oder weniger deutlicher Trend zu einer Verbesserung der posturalen Stabilität. Durch den Einsatz einer Testbatterie konnte somit insgesamt von einem positiven Einfluss eines Gleichgewichtstrainings auf die posturale Stabilität von Parkinsonpatienten ausgegangen werden. Weiterhin werden zahlreiche positive Nebeneffekte bzgl. der allgemeinen Fitness, der psychischen und sozialen Situation durch ein zielgerichtetes Gruppentraining unter geschulter Anleitung vermutet.

L’évolution des populations occidentales s’accompagne d’une augmentation de la sédentarité et des maladies métaboliques qui accroissent les problèmes de santé. Ces évolutions ont des répercussions sur le muscle squelettique qui voit sa capacité à produire de l’énergie aérobie diminuer. Néanmoins, le muscle squelettique est très plastique et les capacités oxydatives musculaires s’améliorent rapidement par l'activité physique. Les mitochondries sont des éléments majeurs des capacités oxydatives musculaires et la compréhension des mécanismes moléculaires qui régissent la biogenèse et la fonction mitochondriale est nécessaire pour prescrire au mieux l’activité physique.L’exercice intermittent semble être de plus en plus utilisé dans la pratique. Plusieurs arguments sont mis en avant pour préconiser cette modalité : 1) le temps passé à haute consommation d’oxygène, 2) la haute intensité et 3) les perturbations métaboliques induites par les variations d’intensité au cours de l’exercice. Cependant, l’influence des perturbations métaboliques sur les capacités oxydatives musculaires n’a pas encore été clairement démontrée. L’objet des mes travaux de thèse s’est donc focalisé sur ces perturbations métaboliques et leurs effets sur les voies de signalisation impliquées dans la biogenèse mitochondriale. Afin de caractériser l’implication des perturbations métaboliques dans la stimulation des voies de signalisation de la biogenèse mitochondriale, nous avons comparé l’influence d’exercices aigus sur ces voies de signalisation. Deux protocoles nous ont permis d’investiguer l’influence des variations métaboliques. Le premier a consisté, lors d’un exercice de intermittent, à identifier la durée du cycle induisant les plus grandes perturbations métaboliques et à caractériser les effets de la modalité d’exercice sur un exercice de 30 minutes de pédalage à 70%WRpic. Le second protocole visait à déterminer l’influence de la répétition des perturbations métaboliques sur les voies de signalisation régulant la biogenèse mitochondriale.Afin d’identifier la durée de cycle produisant le plus de variations métaboliques, nous avons analysé l’évolution de la consommation d’oxygène et quantifié les variations métaboliques. Pour cela nous avons utilisé trois paramètres : 1) un paramètre quantitatif, 2) un paramètre qualitatif et 3) un index associant les paramètres quantitatif et qualitatif. La comparaison de trois durées de cycle différentes (30s d’effort:30s de récupération passive ; 60s:60s et 120s:120s) nous a permis de mettre en évidence que la modalité 60s:60s est celle qui induit le plus de variations métaboliques et cela pour une dépense énergétique identique pour les trois modalités.Notre seconde étude a consisté à comparer 30 minutes de pédalage à 70%WRpic sous deux modalités différentes : continue (1 bloc de 30min) et intermittente (30 bloc de 1min entrecoupés de 1min). La répétition de phase d’exercice et de repos lors de l’exercice intermittent créée plus de perturbation du métabolisme et entraîne une phosphorylation supérieure de l'AMPK, CaMKII et p38 MAPK. Ces kinases sont situées en amont de PGC-1α, un important régulateur de la biogenèse mitochondriale dans le muscle squelettique. Ces résultats mettent donc en évidence un effet spécifique des perturbations métaboliques sur les voies de signalisation contrôlant la biogenèse mitochondriale.Ces travaux ouvrent de nouvelles perspectives sur les méthodes de réentraînement de personnes sédentaires ou atteintes de pathologie chronique. Les futurs travaux viseront à confirmer nos résultats lors d’interventions chroniques et d’explorer ces effets chez différentes populations
Western life evolution is associated with an increase in sedentary behaviours and metabolic diseases leading to health alteration. This evolution affects the skeletal muscle, which is characterized by a decrease in its ability to produce aerobic energy. However, skeletal muscle is a highly malleable tissue, capable of considerable metabolic adaptations in response to physical activity. Mitochondria produce the aerobic energy within the skeletal muscle. Understanding the molecular mechanisms that regulate mitochondrial biogenesis and its function is necessary to improve physical activity prescription.The intermittent exercise is currently used in rehabilitation programs. Several arguments are put forward to utilizing this method: 1) the time spent at high oxygen consumption, 2) the high intensity of exercise and 3) the metabolic disturbances induced by variations of intensity during exercise. However, the influence of metabolic disturbances on muscle oxidative capacity has not been clearly demonstrated. The purpose of my thesis work has therefore focused on these metabolic perturbations and their effects on signalling pathways involved in mitochondrial biogenesis. In order to characterize the influence of metabolic disturbances on the signalling pathways involved in mitochondrial biogenesis, we compared the influence of acute exercises. We realized two protocols to investigate the influence of metabolic disturbances. The first study compared three intermittent exercises in order to identify the optimal duty-cycle duration to induce the biggest metabolic disturbances and to compare metabolic responses of intermittent and continuous exercise performed at 70%WRpic. The second protocol evaluated the influence of the repetition of metabolic disturbances on signalling pathways involved in mitochondrial biogenesis.In order to identify the duty-cycle duration producing more metabolic fluctuations, we analysed the changes of oxygen consumption and quantified metabolic variations. We used three parameters: 1) a quantitative parameter, 2) a qualitative parameter, and 3) an index combining quantitative and qualitative parameters. Comparison of three different duty-cycle durations (30s work:30s passive recovery; 60s:60s, and 120s:120s) revealed that the 60s:60s modality induces more metabolic fluctuations for a same energy expenditure.Our second study compared 30 minutes of pedalling at 70%WRpic realized by two different modalities: continuous (30min 1 block) and intermittent (30 1min block interspersed by 1min of passive recovery). Repetition of transitions from rest to exercise during the intermittent exercise creates higher metabolic disturbances and leads to a higher phosphorylation of AMPK, p38 MAPK and CaMKII. These kinases are upstream of PGC-1α, an important regulator of mitochondrial biogenesis in skeletal muscle. All together, these results demonstrate that metabolic disturbances are involved in mitochondrial signalling pathways activation.This work opens up new perspectives on exercise training prescription for sedentary or chronic pathology people. Future work will aim to confirm our results in chronic interventions and explore these effects in different populations

abstract: Injuries and death associated with fall incidences pose a significant burden to society, both in terms of human suffering and economic losses. The main aim of this dissertation is to study approaches that can reduce the risk of falls. One major subset of falls is falls due to neurodegenerative disorders such as Parkinson’s disease (PD). Freezing of gait (FOG) is a major cause of falls in this population. Therefore, a new FOG detection method using wavelet transform technique employing optimal sampling window size, update time, and sensor placements for identification of FOG events is created and validated in this dissertation. Another approach to reduce the risk of falls in PD patients is to correctly diagnose PD motor subtypes. PD can be further divided into two subtypes based on clinical features: tremor dominant (TD), and postural instability and gait difficulty (PIGD). PIGD subtype can place PD patients at a higher risk for falls compared to TD patients and, they have worse postural control in comparison to TD patients. Accordingly, correctly diagnosing subtypes can help caregivers to initiate early amenable interventions to reduce the risk of falls in PIGD patients. As such, a method using the standing center-of-pressure time series data has been developed to identify PD motor subtypes in this dissertation. Finally, an intervention method to improve dynamic stability was tested and validated. Unexpected perturbation-based training (PBT) is an intervention method which has shown promising results in regard to improving balance and reducing falls. Although PBT has shown promising results, the efficacy of such interventions is not well understood and evaluated. In other words, there is paucity of data revealing the effects of PBT on improving dynamic stability of walking and flexible gait adaptability. Therefore, the effects of three types of perturbation methods on improving dynamics stability was assessed. Treadmill delivered translational perturbations training improved dynamic stability, and adaptability of locomotor system in resisting perturbations while walking.
Dissertation/Thesis
Doctoral Dissertation Biomedical Engineering 2019

碩士
國立臺灣大學
物理治療學研究所
107
Background: Patients with degenerative cervical myelopathy (DCM) may still present with balance and functional impairments after surgery. Perturbation-based balance training had positive effects on balance and functional performance in elderly population and patients with neurological disorders. However, no study so far had conducted perturbation-based balance training in postoperative patients with DCM. Purpose: To evaluate the effects of perturbation-based balance training on balance and functional performance in postoperative patients with DCM. Design: A two-arm prospective controlled clinical trial. Methods: Fifteen postoperative patients with DCM (DCM group) and 14 age-matched healthy adults (healthy control group) were recruited in the study. The DCM group received a 4-week perturbation-based balance training using a perturbation treadmill. The static postural control was assessed by mean velocity of center of pressure (COP) during quiet standing with eyes open and eyes closed. The dynamic postural control was assessed by center of mass (COM) variance and reaction time during standing with perturbation. The functional performance was assessed by gait speed and COM-COP inclination angle during level-ground walking, as well as the Timed Up and Go Test (TUG) performance and the muscle strength. The disability questionnaires included Visual Analog Scale, Neck Disability Index, and Lower Extremity Function of Japanese Orthopaedic Association Cervical Myelopathy Evaluation Questionnaire. The assessments were conducted pre-test and post-test for DCM group but only once for the healthy control group. Results: In the DCM group, significant improvements after training were observed in the mean velocity of COP in eyes closed condition, COM variance, reaction time, gait speed, and TUG performance. The mean velocity of COP in eyes open condition, COM-COP inclination angles, muscle strength, and the disability questionnaires did not change significantly. For between-group comparisons, significant differences that were observed pre-test were not observed post-test, except for the mean velocity of COP in eyes open condition, COM-COP inclination angle in the frontal plane, and muscle strength. Conclusion: The 4-week perturbation-based balance training is a potential rehabilitation strategy for addressing balance and functional impairments in postoperative patients with DCM. In addition, the training regimens offer a practical reference for future studies on populations with balance disorders. However, further intervention for the wide-based gait is needed to enhance the ambulation efficiency in postoperative patients with DCM

Privacy preservation is important for machine learning and datamining, but measures designed to protect private information sometimes result in a trade off: reduced utility of the training samples. This thesis introduces a privacy preserving approach that can be applied to decision-tree learning, without concomitant loss of accuracy. It describes an approach to the preservation of privacy of collected data samples in cases when information of the sample database has been partially lost. This approach converts the original sample datasets into a group of unreal datasets, where an original sample cannot be reconstructed without the entire group of unreal datasets. This approach does not perform well for sample datasets with low frequency, or when there is low variance in the distribution of all samples. However, this problem can be solved through a modified implementation of the approach introduced later in this thesis, by using some extra storage.

In this study an ecosystemic approach to self-hypnosis was utilised as a tool to explore and describe the healing of sport injuries. Four injured Subjects, from four different sporting activities participated in the study. Self-hypnosis/hypnosis was used as a linguistic means to perturb the problem-defining ideas within which the sport injury was embedded. Problem dis-solution involved a process of reframing each Subject's current reality through dialogue, and a new reality for each respective Subject was co-constructed through consequent linguistic differentiation. The hypnotist, participating in the linguistic domain as an equal participant, looked for intended meanings in each respective conversational exchange with the athletes, and synthesised information creatively. This process and the thinking behind each case study is described in detail in this dissertation.
Psychology
M.A. (Psychology)

Neural networks was shown to be very susceptible to small and imperceptible perturbations on its input. In this thesis, we study perturbations on two-layer piecewise linear networks. Such studies are essential in training neural networks that are robust to noisy input. One type of perturbations we consider is `1 norm bounded perturbations. Training Deep Neural Networks (DNNs) that are robust to norm bounded perturbations, or adversarial attacks, remains an elusive problem. While verification based methods are generally too expensive to robustly train large networks, it was demonstrated in [1] that bounded input intervals can be inexpensively propagated per layer through large networks. This interval bound propagation (IBP) approach lead to high robustness and was the first to be employed on large networks. However, due to the very loose nature of the IBP bounds, particularly for large networks, the required training procedure is complex and involved. In this work, we closely examine the bounds of a block of layers composed of an affine layer followed by a ReLU nonlinearity followed by another affine layer. In doing so, we propose probabilistic bounds, true bounds with overwhelming probability, that are provably tighter than IBP bounds in expectation. We then extend this result to deeper networks through blockwise propagation and show that we can achieve orders of magnitudes tighter bounds compared to IBP. With such tight bounds, we demonstrate that a simple standard training procedure can achieve the best robustness-accuracy tradeoff across several architectures on both MNIST and CIFAR10. We, also, consider Gaussian perturbations, where we build on a previous work that derives the first and second output moments of a two-layer piecewise linear network [2]. In this work, we derive an exact expression for the second moment, by dropping the zero mean assumption in [2].