37 research outputs found
Literary 'bone collectors': recovering the voices of an older generation of Chinese women in British Columbia (1923-1967) in works by Wayson Choy, SKY Lee, and Denise Chong
No full textDrawing from theories of Chinese indigenous psychology, subjective well-being, and cultural identity, this study examines the representations of Chinese women immigrants in four texts by Wayson Choy, SKY Lee, and Denise Chong. These women lived through the Canadian Chinese-exclusion period of 1923-1947, and many witnessed the introduction of the points-based system in 1967. Situating my examination in the social, historical, and cultural contexts of the multiple worlds where they lived, I suggest that through these characters, who use their own language/dialect to tell their own stories and demonstrate their own values, these texts literally and figuratively give this generation of women a voice and present them as complex humans. In this way, the authors are actually ‘bone collectors’; they excavate neglected aspects of Canadian history to find the ‘bones’ of their elders and attempt to piece these together, while being aware that they are not fully capable of doing so
Reliability analysis of sea-dikes in Shanghai city, China
No full textThis research focuses on the estimation the failure probability of Shanghai sea-dikes system, taking into account several failure mechanisms. This method follows fully probabilistic approach, in which all relevant parameters for the resistances and the hydraulic loads vary according to specific distributions.Civil Engineering | Hydraulic Engineerin
Water shell: A trip for water
No full textWater shell is a Utopian vision based on the thinking of water in the Tashkent context. It is the reflection over the political and chronological aspect of water: The water distribution is largely decided by the political importance of places, and so intervened by city development. Besides, the water body is shrinking in Tashkent in a long-time span. From the research, an assumed ironic story comes out: Water in Tashkent is irreversibly shrinking, so the water shell is built to take powerful people to find a new land full of water. The water shell floats or walks along the city main canal secretly and silently. It can be seen as an amphibious adaptation to the dry context of Tashkent. The water shell takes in water from outside, purifies, and produces hydro-energy. The inner space operates as the mechanism of water flows. It explores the intrinsic properties of water.Architecture, Urbanism and Building Science
Research on the Similarities between the Plot of Ji Chun Tai and Content of Sichuan Opera
Full text linkJi Chun Tai is the masterpiece of Sichuan dialect on late Qing Dynasty, composed of 40 vernacular short stories. It is divided into four parts, namely, Yuan Ji, Heng Ji, Li Ji, and Zhen Ji. Each part contains ten short stories. The author of Ji Chun Tai is a literator from Zhong Jiang who failed in imperial examination System in late Qing Dynasty. There are a large number of Sichuan Opera elements in those forty vernacular short stories. Generally speaking, the plot of Ji Chun Tai is full of ups and downs, together with relatively concentrated conflicts, which reflects the characteristics of Sichuan opera. Besides, the thought of persuasion and punishment, strong superstitious color, and detective story in Ji Chui Tai are combined together to reflect the characteristics of Sichuan Opera
Cloud-assisted rendering support for mobile online gaming
No full textThe main purpose of this project is to provide cloud-assisted rendering support to an existing cloud gaming system. They current system is a cross-platform cloud gaming system. The games run on cloud server while players interact with the games through clients. The author’s task consists of two components namely recording of videos and playing of the recorded videos. The project is developed using Python 2.6.6, PHP 5.3.3, HTML5, JavaScript and MySQL 5.1.73. Development platform is on Centos 6.5 Virtual Machine. The system is ran and tested on the VM using Google Chrome 39.0.2171.95 and Mozilla Firefox 24.7.0. At the end of the project, the author should provide the service to allow the game player to start and stop the recording of game during gameplay as well as providing the portal for the player to replay the videos. The completed tasks should be integrated into the current running system. For further improvement, the mobile version of the added functionalities can be explored and experimented. In addition more features can be added such as screen capture function and linkage to social network accounts. As storage of video can be limited on the server, the system can also consider integration with cloud storage system such as Rackspace. It can reduce both the load of server and the latency for users overseas.Bachelor of Engineering (Computer Science
Cyber-Physical System Security of Distribution Systems
No full textThe Information and Communications Technology (ICT) for control and monitoring of power systems is a layer on top of the physical power system infrastructure. The cyber system and physical power system components form a tightly coupled Cyber-Physical System (CPS). Sources of vulnerabilities arise from the computing and communication systems of the cyber-power grid. Cyber intrusions targeting the power grid are serious threats to the reliability of electricity supply that is critical to society and the economy. In a typical Information Technology environment, numerous attack scenarios have shown how unauthorized users can access and manipulate protected information from a network domain. The need for cyber security has led to industry standards that power grids must meet to ensure that the monitoring, operation, and control functions are not disrupted by cyber intrusions. Cyber security technologies such as encryption and authentication have been deployed on the CPS. Intrusion or anomaly detection and mitigation tools developed for power grids are emerging. This survey paper provides the basic concepts of cyber vulnerabilities of distribution systems and CPS security. The important ICT subjects for distribution systems covered in this paper include Supervisory Control And Data Acquisition, Distributed Energy Resources, including renewable energy and smart meters.Green Open Access added to TU Delft Institutional Repository 'You share, we take care!' - Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.Intelligent Electrical Power Grid
Risk stratification of acute upper gastrointestinal bleeding patients by transferring knowledge learnt from a critical care database
No full textPh.D.Acute upper gastrointestinal bleeding (AUGIB) is a common medical emergency. Risk stratification of AUGIB patients can direct patients into the appropriate level of care without compromising patients’ safety. Nevertheless, current clinical scores have suboptimal performance in predicting high risk patients and limited clinical adoption. Limited data size from any single centre and the diversity of causes of mortality of AUGIB patients are significant challenges for adopting advanced machine learning algorithms, such as deep learning, to accurately predict mortality and other risks of patients. Meanwhile, high risk patients may require therapeutic endoscopy intervention to stop bleeding or prevent recurrent bleeding. Machine learning algorithms can be applied in this process for better patient management.This study aims to develop and apply machine learning algorithms to improve the management of AUGIB patients by 1) predicting the patients’ risks of mortality and their need for endoscopic intervention, and 2) modelling a magnetic navigation system (MNS) for more precise control of endoscopic instrument to treat AUGIB patients.In the first study, patients’ risk of mortality was predicted by developing a multi-task recurrent neural network with attention mechanisms. The multi-task learning model used the reconstruction of a patient’s physiological time series as an auxiliary task. The multi-task model was evaluated on a large public critical care database, i.e. MIMIC-III, and the performance was compared with a clinical Simplified Acute Physiology Score (SAPS-II). During the same initial 24-hour period into critical care, the multi-task learning model achieved better sensitivity (0.503±0.020 versus 0.365±0.021) than SAPS-II. Furthermore, our result showed that even when the observation period was shortened to 6 hours, the multi-task learning model could still predict with similar accuracy as SAPS-II.In the second study, a multi-layer perceptron (MLP) neural network was developed to predict multiple clinical outcomes of AUGIB patients, including hospital mortality, the need for therapeutic endoscopy intervention, and a composite clinical outcome of these two outcomes. During the training process, the multi-task learning model developed in the first study was used as a teacher model to provide additional information. The MLP student model was trained to minimize a knowledge distillation loss function, which is a conditional, weighted sum of the ground truth labels and teacher model’s predictions. Experiments were conducted on data collected from 300 AUGIB patients. With the knowledge from the teacher model, the MLP student model was able to correctly predict at least half of the death events in the testing dataset; improve the sensitivity in identifying patients who need therapeutic endoscopy intervention by 2.7%; and improve the specificity of predicting the composite endpoint by 10.4%. This study demonstrated, for the first time, the feasibility of improving risk stratification of AUGIB patients by transferring knowledge learnt from a different patient group.Lastly, two machine learning methods, random forests (RF) and MLP, were applied to model a MNS. The MNS has potential to be used to assist endoscopists in performing therapeutic endoscopy for AUGIB haemostasis. Nevertheless, existing modelling approaches of MNS assume a linear behaviour of multiple magnetic coils, leading to significant modelling errors and less precise control of endoscopic instruments at higher magnetic fields. Both machine learning algorithms produced lower mean error in predicting the resultant magnetic fields. The RF model and the MLP model reduced the error of a multipole baseline model by 40% and over 80%, respectively. This study demonstrated the feasibility of using machine learning algorithms to model a MNS. The MLP model can account for the complex, nonlinear behaviour of the resultant magnetic fields, especially at high currents, and is critical for precisely controlling endoscopic instruments or therapeutic drugs for AUGIB patients using magnetic navigation.To conclude, this work developed and applied machine learning algorithms for different procedures during the management of AUGIB patients, including to assess the risk of mortality and need of therapeutic endoscopic intervention of these patients, and to allow precise control of instruments and drugs in these patients during therapeutic endoscopy.急性上消化道出血（AUGIB）是一種常見的醫療急症。 对AUGIB病人的危險分層可以指導患者接受適當的護理，并且不會影響患者的安全。然而，目前的臨床評分系统在預測高風險患者方面表現欠佳。任何單個医疗中心的有限數據量和AUGIB患者死亡原因的多樣性对开发先進的機器學習算法（例如深度學習）來準確預測AUGIB患者的死亡和其他風險造成重大挑戰。同時，高風險患者可能需要治療性內鏡干預以止血或預防復發性出血，機器學習算法可應用於此過程以實現更好的患者管理。本研究旨在開發和應用機器學習算法以改善對AUGIB患者的管理， 具體分為：1）預測患者的死亡風險及其對內鏡干預的需求; 2）對磁導航系統（MNS）建模以更精確地控制治療AUGIB患者的內鏡儀器。第一階段研究開發了一個利用註意機制來預測患者住院死亡概率的多任務循環神經網絡。它可以同時重建患者的生理時間序列並預測其死亡風險。通過在大型公開重症監護數據集上進行的實驗，同樣通過利用首24小時重症監護数据來於預測其住院死亡概率，該方法與傳統的Simplified Acute Physiology Score（SAPS-II）評分相比，實現了更好的靈敏度（0.503±0.020和0.365±0.021）。我們的结果显示当减少至6小時的觀察数据时，多任務學習模型仍可达到同SAPS-II相似的水平。第二項研究開發了多層感知器（MLP）神經網絡來預測AUGIB患者的多種臨床結果，包括住院死亡概率，治療性內鏡干預的需要，以及一個包括前两种的綜合臨床結果。在訓練过程中，第一项研究中開發的多任務學習模型被視為教師模型，以提供額外的信息。 MLP學生模型經過訓練，旨在最大限度地減少一個知識蒸餾損失函數。该损失函数是对數據集中的標籤和教師模型預測的信息进行有條件地加權求和。在對300名AUGIB患者進行的實驗中，利用教師模型的知識，MLP學生模型能夠準確地預測測試集中至少一半的死亡事件；在識別需要治療性內鏡干預的患者上的敏感性提高了2.7％；并在預測需要治療性內窺鏡干預和/或死亡的患者時，特異性提高了10.4％。該研究首次證明了通過迁移從不同患者中学到的知識來改善对AUGIB患者的风险預測的可行性。在最后一阶段实验中，兩種机器学习方法，隨機森林（RF）和MLP，被用来模擬一套磁導航系統（MNS）。MNS具有很大的潜力被用于协助內鏡醫師進行針對AUGIB止血的內鏡治療。現有建模方法假定這些系統的線性行為，導致在較高磁場的非線性區域內會產生顯著的建模誤差。兩種機器學習算法在預測磁場強度時均產生相对較低的平均誤差;與最先進的線性模型相比，RF和MLP模型分别降低了40％和至少80％的平均誤差。該研究證明了使用機器學習算法對MNS進行建模的可行性。 特別是在高電流時，MLP模型可以更好地考虑磁場的複雜非線性行為。这對於使用磁導航系统来精確控制內窺器械或治療藥物以完成对AUGIB患者的治疗是至關重要的。綜上所述，這項工作將机器学习方法應用於AUGIB患者管理期間的不同阶段，包括評估患者的死亡以及需要治療性內鏡干預的風險，並协助進行治療性內鏡以达到精確控制內鏡治疗器械和藥物的目的。Yu, Ruoxi.Thesis Ph.D. Chinese University of Hong Kong 2019.Includes bibliographical references (leaves 95-108).Abstracts also in Chinese.Title from PDF title page (viewed on 20, May 2021)
On the spectral efficiency of MMSE-based MIMO OTFS systems
No full textWe investigate the spectral efficiency (SE) of an uplink multi-user multiple-input multiple-output (MIMO) system with orthogonal time frequency space (OTFS) modulation. Two multiple access schemes are considered hereafter, namely, delay division multiple access (DDMA) and Doppler division multiple access (DoDMA). To avoid multi-user interference (MUI), we separate the delay Doppler (DD) domain resource blocks assigned to different users by guard bands. We design a minimum-meansquare-error (MMSE) receiver to combine the received signals for the detection of different users. We analyze the SE for the considered MIMO-OTFS system and quantify the performance gains achieved over a MIMO system with orthogonal frequency-division multiple access (OFDMA). Our simulation results demonstrate a noticeable improvement in the performance of MIMO-OTFS over MIMO-OFDMA.<br/
