Dissertations / Theses on the topic 'Text Generation Using Neural Networks'

A related work report is a section in a research paper which integrates key information from a list of related scientific papers providing context to the work being presented. Related work reports can either be descriptive or integrative. Integrative related work reports provide a high-level overview and critique of the scientific papers by comparing them with each other, providing fewer details of individual studies. Descriptive related work reports, instead, provide more in-depth information about each mentioned study providing information such as methods and results of the cited works. In order to write a related work report, scientist have to identify, condense/summarize, and combine relevant information from different scientific papers. However, such task is complicated due to the available volume of scientific papers. In this context, the automatic generation of related work reports appears to be an important problem to tackle. The automatic generation of related work reports can be considered as an instance of the multi-document summarization problem where, given a list of scientific papers, the main objective is to automatically summarize those scientific papers and generate related work reports. In order to study the problem of related work generation, we have developed a manually annotated, machine readable data-set of related work sections, cited papers (e.g. references) and sentences, together with an additional layer of papers citing the references. We have also investigated the relation between a citation context in a citing paper and the scientific paper it is citing so as to properly model cross-document relations and inform our summarization approach. Moreover, we have also investigated the identification of explicit and implicit citations to a given scientific paper which is an important task in several scientific text mining activities such as citation purpose identification, scientific opinion mining, and scientific summarization. We present both extractive and abstractive methods to summarize a list of scientific papers by utilizing their citation network. The extractive approach follows three stages: scoring the sentences of the scientific papers based on their citation network, selecting sentences from each scientific paper to be mentioned in the related work report, and generating an organized related work report by grouping the sentences of the scientific papers that belong to the same topic together. On the other hand, the abstractive approach attempts to generate citation sentences to be included in a related work report, taking advantage of current sequence-to-sequence neural architectures and resources that we have created specifically for this task. The thesis also presents and discusses automatic and manual evaluation of the generated related work reports showing the viability of the proposed approaches.<br>La sección de trabajos relacionados de un artículo científico resume e integra información clave de una lista de documentos científicos relacionados con el trabajo que se presenta. Para redactar esta sección del artículo científico el autor debe identificar, condensar/resumir y combinar información relevante de diferentes artículos. Esta tarea es complicada debido al gran volumen disponible de artículos científicos. En este contexto, la generación automática de tales secciones es un problema importante a abordar. La generación automática de secciones de trabajo relacionados puede ser considerada como una instancia del problema de resumen de documentos múltiples donde, dada una lista de documentos científicos, el objetivo es resumir automáticamente esos documentos científicos y generar la sección de trabajos relacionados. Para estudiar este problema, hemos creado un corpus de secciones de trabajos relacionados anotado manualmente y procesado automáticamente. Asimismo, hemos investigado la relación entre las citaciones y el artículo científico que se cita para modelar adecuadamente las relaciones entre documentos y, así, informar nuestro método de resumen automático. Además, hemos investigado la identificación de citaciones implícitas a un artículo científico dado que es una tarea importante en varias actividades de minería de textos científicos. Presentamos métodos extractivos y abstractivos para resumir una lista de artículos científicos utilizando su red de citaciones. El enfoque extractivo sigue tres etapas: cálculo de la relevancia las oraciones de cada artículo en función de la red de citaciones, selección de oraciones de cada artículo científico para integrarlas en el resumen y generación de la sección de trabajos relacionados agrupando las oraciones por tema. Por otro lado, el enfoque abstractivo intenta generar citaciones para incluirlas en un resumen utilizando redes neuronales y recursos que hemos creado específicamente para esta tarea. La tesis también presenta y discute la evaluación automática y manual de los resúmenes generados automáticamente, demostrando la viabilidad de los enfoques propuestos.<br>Una secció d’antecedents o estat de l’art d’un articulo científic resumeix la informació clau d'una llista de documents científics relacionats amb el treball que es presenta. Per a redactar aquesta secció de l’article científic l’autor ha d’identificar, condensar / resumir i combinar informació rellevant de diferents articles. Aquesta activitat és complicada per causa del gran volum disponible d’articles científics. En aquest context, la generació automàtica d’aquestes seccions és un problema important a abordar. La generació automàtica d’antecedents o d’estat de l’art pot considerar-se com una instància del problema de resum de documents. Per estudiar aquest problema, es va crear un corpus de seccions d’estat de l’art d’articles científics manualment anotat i processat automàticament. Així mateix, es va investigar la relació entre citacions i l’article científic que es cita per modelar adequadament les relacions entre documents i, així, informar el nostre mètode de resum automàtic. A més, es va investigar la identificació de citacions implícites a un article científic que és un problema important en diverses activitats de mineria de textos científics. Presentem mètodes extractius i abstractius per resumir una llista d'articles científics utilitzant el conjunt de citacions de cada article. L’enfoc extractiu segueix tres etapes: càlcul de la rellevància de les oracions de cada article en funció de les seves citacions, selecció d’oracions de cada article científic per a integrar-les en el resum i generació de la secció de treballs relacionats agrupant les oracions per tema. Per un altre costat, l’enfoc abstractiu implementa la generació de citacions per a incloure-les en un resum que utilitza xarxes neuronals i recursos que hem creat específicament per a aquest tasca. La tesi també presenta i discuteix l'avaluació automàtica i el manual dels resums generats automàticament, demostrant la viabilitat dels mètodes proposats.

Submitted by JOSIANE SANTOS DE OLIVEIRA (josianeso) on 2019-03-07T14:41:05Z No. of bitstreams: 1 Roger Alan Stein_.pdf: 476239 bytes, checksum: a87a32ffe84d0e5d7a882e0db7b03847 (MD5)<br>Made available in DSpace on 2019-03-07T14:41:05Z (GMT). No. of bitstreams: 1 Roger Alan Stein_.pdf: 476239 bytes, checksum: a87a32ffe84d0e5d7a882e0db7b03847 (MD5) Previous issue date: 2018-03-28<br>CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior<br>Efficient distributed numerical word representation models (word embeddings) combined with modern machine learning algorithms have recently yielded considerable improvement on automatic document classification tasks. However, the effectiveness of such techniques has not been assessed for the hierarchical text classification (HTC) yet. This study investigates application of those models and algorithms on this specific problem by means of experimentation and analysis. Classification models were trained with prominent machine learning algorithm implementations—fastText, XGBoost, and Keras’ CNN—and noticeable word embeddings generation methods—GloVe, word2vec, and fastText—with publicly available data and evaluated them with measures specifically appropriate for the hierarchical context. FastText achieved an LCAF1 of 0.871 on a single-labeled version of the RCV1 dataset. The results analysis indicates that using word embeddings is a very promising approach for HTC.<br>Modelos eficientes de representação numérica textual (word embeddings) combinados com algoritmos modernos de aprendizado de máquina têm recentemente produzido uma melhoria considerável em tarefas de classificação automática de documentos. Contudo, a efetividade de tais técnicas ainda não foi avaliada com relação à classificação hierárquica de texto. Este estudo investiga a aplicação daqueles modelos e algoritmos neste problema em específico através de experimentação e análise. Modelos de classificação foram treinados usando implementações proeminentes de algoritmos de aprendizado de máquina—fastText, XGBoost e CNN (Keras)— e notórios métodos de geração de word embeddings—GloVe, word2vec e fastText—com dados disponíveis publicamente e avaliados usando métricas especificamente adequadas ao contexto hierárquico. Nesses experimentos, fastText alcançou um LCAF1 de 0,871 usando uma versão da base de dados RCV1 com apenas uma categoria por tupla. A análise dos resultados indica que a utilização de word embeddings é uma abordagem muito promissora para classificação hierárquica de texto.

Many software companies use issue tracking systems to organize their work. However, when working on large projects, across multiple teams, a problem of finding the correctteam to solve a certain issue arises. One team might detect a problem, which must be solved by another team. This can take time from employees tasked with finding the correct team and automating the dispatching of these issues can have large benefits for the company. In this thesis, the use of machine learning methods, mainly convolutional neural networks (CNN) for text classification, has been applied to this problem. For natural language processing both word- and character-level representations are commonly used. The results in this thesis suggests that the CNN learns different information based on whether word- or character-level representation is used. Furthermore, it was concluded that the CNN models performed on similar levels as the classical Support Vector Machine for this task. When compared to a human expert, working with dispatching issues, the best CNN model performed on a similar level when given the same information. The high throughput of a computer model, therefore, suggests automation of this task is very much possible.

Large datasets are a crucial requirement to achieve high performance, accuracy, and generalisation for any machine learning task, such as prediction or anomaly detection, However, it is not uncommon for datasets to be small or imbalanced since gathering data can be difficult, time-consuming, and expensive. In the task of collecting vehicle sensor time series data, in particular when the vehicle has an abnormal behaviour, these struggles are present and may hinder the automotive industry in its development. Synthetic data generation has become a growing interest among researchers in several fields to handle the struggles with data gathering. Among the methods explored for generating data, generative adversarial networks (GANs) have become a popular approach due to their wide application domain and successful performance. This thesis focuses on generating multivariate time series data that are similar to vehicle sensor readings from the air pressures in the brake system of vehicles with an abnormal behaviour, meaning there is a leakage somewhere in the system. A novel GAN architecture called TimeGAN was trained to generate such data and was then evaluated using both qualitative and quantitative evaluation metrics. Two versions of this model were tested and compared. The results obtained proved that both models learnt the distribution and the underlying information within the features of the real data. The goal of the thesis was achieved and can become a foundation for future work in this field.<br>När man applicerar en modell för att utföra en maskininlärningsuppgift, till exempel att förutsäga utfall eller upptäcka avvikelser, är det viktigt med stora dataset för att uppnå hög prestanda, noggrannhet och generalisering. Det är dock inte ovanligt att dataset är små eller obalanserade eftersom insamling av data kan vara svårt, tidskrävande och dyrt. När man vill samla tidsserier från sensorer på fordon är dessa problem närvarande och de kan hindra bilindustrin i dess utveckling. Generering av syntetisk data har blivit ett växande intresse bland forskare inom flera områden som ett sätt att hantera problemen med datainsamling. Bland de metoder som undersökts för att generera data har generative adversarial networks (GANs) blivit ett populärt tillvägagångssätt i forskningsvärlden på grund av dess breda applikationsdomän och dess framgångsrika resultat. Denna avhandling fokuserar på att generera flerdimensionell tidsseriedata som liknar fordonssensoravläsningar av lufttryck i bromssystemet av fordon med onormalt beteende, vilket innebär att det finns ett läckage i systemet. En ny GAN modell kallad TimeGAN tränades för att genera sådan data och utvärderades sedan både kvalitativt och kvantitativt. Två versioner av denna modell testades och jämfördes. De erhållna resultaten visade att båda modellerna lärde sig distributionen och den underliggande informationen inom de olika signalerna i den verkliga datan. Målet med denna avhandling uppnåddes och kan lägga grunden för framtida arbete inom detta område.

This thesis addresses the extraction of medical knowledge from clinical text using deep learning techniques. In particular, the proposed methods focus on cancer clinical trial protocols and chest x-rays reports. The main results are a proof of concept of the capability of machine learning methods to discern which are regarded as inclusion or exclusion criteria in short free-text clinical notes, and a large scale chest x-ray image dataset labeled with radiological findings, diagnoses and anatomic locations. Clinical trials provide the evidence needed to determine the safety and effectiveness of new medical treatments. These trials are the basis employed for clinical practice guidelines and greatly assist clinicians in their daily practice when making decisions regarding treatment. However, the eligibility criteria used in oncology trials are too restrictive. Patients are often excluded on the basis of comorbidity, past or concomitant treatments and the fact they are over a certain age, and those patients that are selected do not, therefore, mimic clinical practice. This signifies that the results obtained in clinical trials cannot be extrapolated to patients if their clinical profiles were excluded from the clinical trial protocols. The efficacy and safety of new treatments for patients with these characteristics are not, therefore, defined. Given the clinical characteristics of particular patients, their type of cancer and the intended treatment, discovering whether or not they are represented in the corpus of available clinical trials requires the manual review of numerous eligibility criteria, which is impracticable for clinicians on a daily basis. In this thesis, a large medical corpora comprising all cancer clinical trials protocols in the last 18 years published by competent authorities was used to extract medical knowledge in order to help automatically learn patient’s eligibility in these trials. For this, a model is built to automatically predict whether short clinical statements were considered inclusion or exclusion criteria. A method based on deep neural networks is trained on a dataset of 6 million short free-texts to classify them between elegible or not elegible. For this, pretrained word embeddings were used as inputs in order to predict whether or not short free-text statements describing clinical information were considered eligible. The semantic reasoning of the word-embedding representations obtained was also analyzed, being able to identify equivalent treatments for a type of tumor in an analogy with the drugs used to treat other tumors. Results show that representation learning using deep neural networks can be successfully leveraged to extract the medical knowledge from clinical trial protocols and potentially assist practitioners when prescribing treatments. The second main task addressed in this thesis is related to knowledge extraction from medical reports associated with radiographs. Conventional radiology remains the most performed technique in radiodiagnosis services, with a percentage close to 75% (Radiología Médica, 2010). In particular, chest x-ray is the most common medical imaging exam with over 35 million taken every year in the US alone (Kamel et al., 2017). They allow for inexpensive screening of several pathologies including masses, pulmonary nodules, effusions, cardiac abnormalities and pneumothorax. For this task, all the chest-x rays that had been interpreted and reported by radiologists at the Hospital Universitario de San Juan (Alicante) from Jan 2009 to Dec 2017 were used to build a novel large-scale dataset in which each high-resolution radiograph is labeled with its corresponding metadata, radiological findings and pathologies. This dataset, named PadChest, includes more than 160,000 images obtained from 67,000 patients, covering six different position views and additional information on image acquisition and patient demography. The free text reports written in Spanish by radiologists were labeled with 174 different radiographic findings, 19 differential diagnoses and 104 anatomic locations organized as a hierarchical taxonomy and mapped onto standard Unified Medical Language System (UMLS) terminology. For this, a subset of the reports (a 27%) were manually annotated by trained physicians, whereas the remaining set was automatically labeled with deep supervised learning methods using attention mechanisms and fed with the text reports. The labels generated were then validated in an independent test set achieving a 0.93 Micro-F1 score. To the best of our knowledge, this is one of the largest public chest x-ray databases suitable for training supervised models concerning radiographs, and also the first to contain radiographic reports in Spanish. The PadChest dataset can be downloaded on request from http://bimcv.cipf.es/bimcv-projects/padchest/. PadChest is intended for training image classifiers based on deep learning techniques to extract medical knowledge from chest x-rays. It is essential that automatic radiology reporting methods could be integrated in a clinically validated manner in radiologists’ workflow in order to help specialists to improve their efficiency and enable safer and actionable reporting. Computer vision methods capable of identifying both the large spectrum of thoracic abnormalities (and also the normality) need to be trained on large-scale comprehensively labeled large-scale x-ray datasets such as PadChest. The development of these computer vision tools, once clinically validated, could serve to fulfill a broad range of unmet needs. Beyond implementing and obtaining results for both clinical trials and chest x-rays, this thesis studies the nature of the health data, the novelty of applying deep learning methods to obtain large-scale labeled medical datasets, and the relevance of its applications in medical research, which have contributed to its extramural diffusion and worldwide reach. This thesis describes this journey so that the reader is navigated across multiple disciplines, from engineering to medicine up to ethical considerations in artificial intelligence applied to medicine.

En reconnaissance d’écriture manuscrite, les architectures Transformer permettent de faibles taux d’erreur, mais sont difficiles à entraîner avec le peu de données annotées disponibles. Dans ce manuscrit, nous proposons des architectures Transformer légères adaptées aux données limitées. Nous introduisons une architecture rapide basée sur un encodeur Transformer, et traitant jusqu’à 60 pages par seconde. Nous proposons aussi des architectures utilisant un décodeur Transformer pour inclure l’apprentissage de la langue dans la reconnaissance des caractères. Pour entraîner efficacement nos architectures, nous proposons des algorithmes de génération de données synthétiques adaptées au style visuel des documents modernes et anciens. Nous proposons également des stratégies pour l’apprentissage avec peu de données spécifiques, et la réduction des erreurs de prédiction. Nos architectures, combinées à l’utilisation de données synthétiques et de ces stratégies, atteignent des taux d’erreur compétitifs sur des lignes de texte de documents modernes. Sur des documents anciens, elles parviennent à s’entraîner avec des nombres limités de données annotées, et surpassent les approches de l’état de l’art. En particulier, 500 lignes annotées sont suffisantes pour obtenir des taux d’erreur caractères proches de 5%<br>Transformer architectures deliver low error rates but are challenging to train due to limited annotated data in handwritten text recognition. We propose lightweight Transformer architectures to adapt to the limited amounts of annotated handwritten text available. We introduce a fast Transformer architecture with an encoder, processing up to 60 pages per second. We also present architectures using a Transformer decoder to incorporate language modeling into character recognition. To effectively train our architectures, we offer algorithms for generating synthetic data adapted to the visual style of modern and historical documents. Finally, we propose strategies for learning with limited data and reducing prediction errors. Our architectures, combined with synthetic data and these strategies, achieve competitive error rates on lines of text from modern documents. For historical documents, they train effectively with minimal annotated data, surpassing state-ofthe- art approaches. Remarkably, just 500 annotated lines are sufficient for character error rates close to 5%

碩士<br>國立雲林科技大學<br>資訊工程系<br>107<br>In recent years, GAN (Generative Adversarial Network) has been very popular, for example, it has been very successful in the application of continuous data such as images, but applications for discrete data (such as text data) still face some difficulties. For this purpose, this thesis applies the Policy Gradient method of reinforcement learning to the conditional sequence generation (Conditional Sequence GAN, CSeqGAN) model to implement text generation technology. Since CSeqGAN can produce more diverse sentences, we use it to implement ordering systems and machine translation, and the entire adversarial training is implemented through the reinforcement learning. We jointly train two models: a generative model and a discriminative model. Given a sequence, the generative model defines a generated dialog sequence, and then the discriminative model is used to distinguish between machine generated text and human generated text. This thesis further uses Policy Gradient Training to encourage the generative model to generate a sequence that makes discriminant model unable to distinguish between real and generated samples. The generator is designed in the form of a Sequence-to-sequence (Seq2Seq) model with Attention mechanism. The discriminator is a binary classifier in which the input is pre-text (history) and post-text (response), encoded by the Hierarchical Encoder. The final output is a probability. The higher the probability value, the closer the y is to the real sample. For discriminator, we further use Monte Carlo search to extend the training to the individual word-based Reward method and apply the teacher-compulsory technique to allow the generated model to be updated in the training with reliable scores. This thesis used the self-generated large amount of data set to train the ordering system, and the translation model was trained by the Insuranceqa-Corpus insurance question and answer corpus. It was found that the regular Seq2Seq model can only produce boring, repeated responses, but our model can generate more interactive and interesting sentences.

With the increase in computation power, coupled with the advancements in the field in the form of GANs and cGANs, Neural Networks have become an attractive proposition for content generation. This opened opportunities for Procedural Content Generation algorithms (PCG) to tap Neural Networks generative power to create tools that allow developers to remove part of creative and developmental burden imposed throughout the gaming industry, be it from investors looking for a return on their investment and from consumers that want more and better content, fast. This dissertation sets out to develop a PCG mixed-initiative tool, leveraging cGANs, to create authored 3D terrains, allowing users to directly influence the resulting generated content without the need for formal training on terrain generation or complex interactions with the tool to influence the generative output, as opposed to state of the art generative algorithms that only allow for random content generation or are needlessly complex. Testing done to 113 people online, as well as in-person testing done to 30 people, revealed that it is indeed possible to develop a tool that allows users from any level of terrain creation knowledge, and minimal tool training, to easily create a 3D terrain that is more realistic looking than those generated by state-of-the-art solutions such as Perlin Noise.<br>Com o aumento do poder de computação, juntamente com os avanços neste campo na forma de GANs e cGANs, as Redes Neurais tornaram-se numa proposta atrativa para a geração de conteúdos. Graças a estes avanços, abriram-se oportunidades para os algoritmos de Geração de Conteúdos Procedimentais(PCG) explorarem o poder generativo das Redes Neurais para a criação de ferramentas que permitam aos programadores remover parte da carga criativa e de desenvolvimento imposta em toda a indústria dos jogos, seja por parte dos investidores que procuram um retorno do seu investimento ou por parte dos consumidores que querem mais e melhor conteúdo, o mais rápido possível. Esta dissertação pretende desenvolver uma ferramenta de iniciativa mista PCG, alavancando cGANs, para criar terrenos 3D cocriados, permitindo aos utilizadores influenciarem diretamente o conteúdo gerado sem necessidade de terem formação formal sobre a criação de terrenos 3D ou interações complexas com a ferramenta para influenciar a produção generativa, opondo-se assim a algoritmos generativos comummente utilizados, que apenas permitem a geração de conteúdo aleatório ou que são desnecessariamente complexos. Um conjunto de testes feitos a 113 pessoas online e a 30 pessoas presencialmente, revelaram que é de facto possível desenvolver uma ferramenta que permita aos utilizadores, de qualquer nível de conhecimento sobre criação de terrenos, e com uma formação mínima na ferramenta, criar um terreno 3D mais realista do que os terrenos gerados a partir da solução de estado da arte, como o Perlin Noise, e de uma forma fácil.

碩士<br>國立中央大學<br>光電科學與工程學系<br>107<br>Neural networks have been successfully applied in many applications. With appropriate training data and fine design of the artificial neural networks structure, the neural networks can be trained to carry out specific tasks. In literature, to design the grating profile for specific diffraction efficiencies requires to solve the inverse Maxwell’s equation with the optimization methods such as the genetic algorithm. The optimization process is time-consuming. By using the neural networks method to perform the learning and testing processes, we could obtain the desired grating profile with a very short time around 0.2 second. In this study, we use the Rigorous Coupled-Wave Analysis method to obtain the diffraction efficiencies of the gratings with specific dimensions. The diffraction efficiencies and the grating profiles serve as the input and the target output respectively to train the neural networks. The grating profiles of the specific diffraction efficiencies can be generated within one second and can be applied for fast grating profile generation. In order to reduce the computation time when encoding the hologram on the metasurface, we also try to use the neural networks to design the metasurface hologram. In our work, we used three light beams from different directions to switch different diffraction patterns on the hologram.

碩士<br>國立中央大學<br>通訊工程研究所<br>100<br>The generated music from automatic music composition is not completely match the rule of music theory in the past research. This thesis proposed using fuzzy neural network (FNN) to training a repeating pattern melody which called refrain in pop music. A refrain usually repeats many times in the music objects. The proposed learning algorithm is based on fuzzy back propagation algorithm (FBP). The main goal of a fuzzy inference system is to model composer decision making within conceptual as the process of composing music. The music theory knowledge of consonance intervals and key signature were adopted to check and adjust the output melody to prevent incorrectly. The simulation results show that the proposed learning algorithm have a good learning ability and well performance.

碩士<br>國立雲林科技大學<br>資訊工程系<br>107<br>It is an interesting topic to give a textual narrative to synthesize a realistic picture, which can be achieved by designing a Sequence to Sequence Model. In recent years, the Generative Adversarial Network（GAN）has been successfully used to train good generators. The idea of generating a confrontational network is to jointly train two models: a generating model and a discriminant model. In the given text sequence, the generative model generates a sequence of pictures, and the discriminative model is used to distinguish the true and false of the generated picture sequence. Although existing synthesis techniques have been able to generate suitable images based on textual descriptions, they do not reflect more detailed parts. In this paper, we use the Stacked Generative Adversarial Network（StackGAN）to design the text-to-picture generation technology, the specific approach is to use the idea of "divide and conquer." In computer science, divide and conquer is used to break down a complex problem into smaller problems that are easier to solve. More specifically, we divide the model into two parts: the Stage-I and the Stage-II sub model. Stage-I is used to outline the original look and color to produce a lower resolution image. Stage-II takes the output of the first model and the text description as input, producing the final higher resolution image, which means that it corrects the defects in the first stage results and increases the detail in the photo. In addition, this paper also uses Conditioning Augmentation to enhance the vector of text description to stabilize the network training process and improve the diversity of generated images. We used the Oxford 102 Flowers dataset to generate high-resolution flower images and used Inception Score to evaluate model scores. The results show that the scores we get are significantly improved compared to the existing models. From the observations, we can also find that the generated images are very close to the real images.

Deep learning is a sub-area of automatic learning that attempts to model complex structures in the data through the application of different neural network architectures with multiple layers of processing. These methods have been successfully applied in areas ranging from image recognition and classification, natural language processing, and bioinformatics. In this work we intend to create methods for named-entity recognition (NER) in text using techniques of deep learning in order to identify genetic mutations.<br>Deep Learning é uma subárea de aprendizagem automática que tenta modelar estruturas complexas no dados através da aplicação de diferentes arquitecturas de redes neuronais com várias camadas de processamento. Estes métodos foram aplicados com sucesso em áreas que vão desde o reconhecimento de imagem e classificação, processamento de linguagem natural e bioinformática. Neste trabalho pretendemos criar métodos para reconhecimento de entidades nomeadas (NER) no texto usando técnicas de Deep Learning, a fim de identificar mutações genéticas.<br>Mestrado em Engenharia de Computadores e Telemática

碩士<br>國立中興大學<br>電機工程學系所<br>107<br>The thesis uses a fully connected neural network (FCRNN) as a kernel controller to control the forward walking gait of a biped robot, the NAO, through the multi-objective modified continuous ant colony optimization (MO-MCACO) algorithm. We control the five joints in each leg, and there are a total of ten degrees of freedom of control. After optimizing the FCRNN, we use it to control the hip pitch, hip roll, and knee pitch angle of one leg. The other seven angles are obtained through the symmetry of the forward walking posture. The FCRNN has two types. One is non-sensor feedback FCRNN, and the other one is sensor-feedback FCRNN. For the latter, sensor feedbacks from one foot or two feet are studied. For the feedbacks from one foot, we only send the feedbacks from the right foot. For the feedbacks from two feet, the two FCRNNs controlling the feet share the same weights, with the difference in the feedbacks from their respectively controlled feet. In the evaluation of the robot control performance, we measure the walking speed, tilting straight, oscillation, walking posture, and stability of the feet in terms of three objective functions. The multi-objective optimization problem is solved by using the MO-MCACO. The simulation environment used in this thesis is the Webots robot simulator. The software-designed FCRNNs are applied to control a real NAO robot to show their performances. Comparisons of the performances of the FCRNNs without sensor feedbacks and those with sensor feedbacks from one foot and two feet are also performed in the simulations.

碩士<br>國立臺灣科技大學<br>電機工程系<br>103<br>Load forecasting plays the extremely important role in power dispatch operations. Accurate load forecasting can provide the more accurate unit commitment and planning in order to improve quality of power supply. Especially, after a lot of renewable energy generations were integrated into power systems, improving the precision of load forecasting is important for increasing safety of system operation and reducing the cost. Although renewable energy is inexhaustible, it is difficult for stable and sustained supply. The power dispatch will be considerable uncertainty. Therefore, if traditional fossil fuel units work together with renewable energy units, we need a more accurate load forecasting method to reduce the risk and cost of electricity operations. This thesis is intended to consider the relevance among the renewable energy generation, weather information and feeder load in order to propose a more accurate load forecasting method to be applied in the same feeder which photovoltaic and wind generators are connected to. This thesis proposed to operate the hourly load forecasting in distribution feeders by the application of the neural network and calculate hourly generation forecasting when photovoltaic and wind generations are connected in the same feeder. The feeder information of Qiaocun substation in Yunlin and Shanjia substation in Miaoli which belong to Taipower Company was tested and the test results confirmed that the proposed methods can improve load forecasting and thus improve quality of power supply for feeders. The results also provide more flexible scheduling policy for power dispatching units.

In recent history, photovoltaic (PV) systems as a means of energy generation have risen in popularity due to the world’s decreasing reliance on fossil fuels and a stronger focus on combating the adverse effects of climate change. While PV systems have immense potential, their vulnerability to faults substantially threatens their efficiency and reliability, potentially reducing their positive impact on the environment and the world economy. Current PV system maintenance strategies are either reactive or preventive, with a limited focus on predictive methods that leverage advanced machine learning models for fault detection. This thesis addresses this research gap, focusing on the development and optimisation of machine learning algorithms for proactive PV system fault detection. This is accomplished through the analysis of various PV system data parameters such as voltage, current, power, energy delivered or received, performance ratio, and meteorological data, among others. This research investigation started with a data collection process from the Desert Knowledge Australia Solar Centre (DKASC), a facility dedicated to solar energy research. After collecting data from 10 of the most fault-prone sites, rigorous pre-processing steps, including cleaning, transforming, and balancing, were employed. Particular attention was given to inverter failures and inverter intermittent issues, as they were identified as the most common faults, significantly influencing PV system performance and reliability. A variety of machine learning algorithms were employed, including deep learning methods such as Artificial Neural Networks and Recurrent Neutral Networks. However, Kernel SVM and K Nearest Neighbours were found to be most effective in predicting the specific individual faults, inverter failures and inverter intermittent issues, respectively. Subsequent parameter optimisation efforts, including adjusting fault occurrence window sizes, running summary days, classifier hyperparameters, and validation methods, enabled differentiation between those two fault types in a combined faults dataset using the K Nearest Neighbours model. This research project makes two novel contributions to the field. First, it developed an adaptive method for predicting specific faults in PV systems. Second, through a parameter optimisation process, this research created an adaptive method for differentiation between two specific faults. Through these adaptive fault prediction methods, the most effective machine learning model can be selected to predict any particular fault or differentiate between any specific faults, enhancing their real-world utility and impact. The findings from this research have considerable implications for future work in this domain. They serve as a guide for further research and development efforts to inform predictive maintenance strategies for PV systems. Future directions include the investigation of other types of faults, expanding the dataset to include more diverse fault scenarios, exploring advanced feature engineering and selection methods, integrating the developed fault prediction models with practical maintenance scheduling systems, and assessing the economic impacts of these models on the efficiency and cost-effectiveness of PV systems. In summary, while this research does contribute to improving the reliability and efficiency of PV systems through enhanced fault prediction, it also provides direction for further research into developing robust predictive maintenance strategies. The findings of this research support the broader goal of making renewable energy more reliable, efficient, and cost-effective in pursuit of a more sustainable energy-driven future.

<div>The compressed air system is widely used throughout the industry. Air compressors are one of the most costly systems to operate in industrial plants in therms of energy consumption. Therefore, it becomes one of the primary target when it comes to electrical energy and load management practices. Load forecasting is the first step in developing energy management systems both on the supply and user side. A comprehensive literature review has been conducted, and there was a need to study if predicting compressed air system’s load is a possibility. </div><div><br></div><div>System’s load profile will be valuable to the industry practitioners as well as related software providers in developing better practice and tools for load management and look-ahead scheduling programs. Feed forward neural networks (FFNN) and long short-term memory (LSTM) techniques have been used to perform 15 minutes ahead prediction. Three cases of different sizes and control methods have been studied. The results proved the possibility of the forecast. In this study two control methods have been developed by using the prediction. The first control method is designed for variable speed driven air compressors. The goal was to decrease the maximum electrical load for the air compressor by using the system's full operational capabilities and the air receiver tank. This goal has been achieved by optimizing the system operation and developing a practical control method. The results can be used to decrease the maximum electrical load consumed by the system as well as assuring the sufficient air for the users during the peak compressed air demand by users. This method can also prevent backup or secondary systems from running during the peak compressed air demand which can result in more energy and demand savings. Load management plays a pivotal role and developing maximum load reduction methods by users can result in more sustainability as well as the cost reduction for developing sustainable energy production sources. The last part of this research is concentrated on reducing the energy consumed by load/unload controlled air compressors. Two novel control methods have been introduced. One method uses the prediction as input, and the other one doesn't require prediction. Both of them resulted in energy consumption reduction by increasing the off period with the same compressed air output or in other words without sacrificing the required compressed air needed for production.</div><div><br></div>