L'antigène de prolifération cellulaire Ki-67 est exprimé de manière constitutive dans les cellules de mammifères. Ki-67 est régulièrement utilisé en tant que marqueur de prolifération cellulaire pour classer les tumeurs. Cependant, malgré son utilisation fréquente en histopathologie, ses fonctions sont encore mal caractérisées. Mes travaux de thèse ont eu pour objectif d'améliorer la compréhension des fonctions biologiques de Ki-67 ainsi que d’étudier l’importance de son expression dans l’initiation et la progression des cancers. Nous avons montré que Ki-67 était dispensable à la prolifération cellulaire. Quant aux souris mutantes Ki-67, elles ne présentaient aucune anomalie de développement, étaient fertiles et vieillissaient normalement. Néanmoins, l’expression de Ki-67 s’est révélée être requise pour l’organisation de l'hétérochromatine dans les cellules prolifératives. En étudiant le contrôle de l'expression de Ki-67, nous avons pu mettre en évidence que les différents niveaux d’expression de Ki-67, souvent observés dans les lignées cellulaires transformées ou non, les tissus et les échantillons de tumeurs des patients, seraient expliqués par une régulation via la machinerie du cycle cellulaire.En utilisant nos souris mutantes Ki-67, nous avons également montré que l’absence de Ki-67 permettait de protéger les souris contre la carcinogenèse intestinale dans les deux différents modèles expérimentaux utilisés. De plus, l'analyse de la conséquence de l'ablation de Ki-67 dans la lignée tumorale murine, 4T1, a révélé que Ki-67 est requis pour le maintien des propriétés souches de ces cellules cancéreuses. En outre, la déplétion de Ki-67 a fortement affecté la croissance des tumeurs et la formation de métastases pulmonaires chez les souris. De façon similaire, l'absence de Ki-67 a fortement altéré le développement des xénogreffes de la lignée MDA-MB-231 dans des souris immuno-déficientes. De plus, le séquençage de l'ARN dans les cellules 4T1 a révélé l’existence d’altérations importantes au niveau transcriptomique, suite à la déplétion de Ki-67.L’ensemble de ces résultats suggère une implication spécifique de Ki-67 dans l'initiation et la progression tumorale et que Ki-67 serait une cible thérapeutique potentielle et intéressante dans le traitement du cancer. / The cell proliferation antigen Ki-67 is constitutively expressed in cycling mammalian cells and is widely used as a cell proliferation marker to grade tumours. Despite its use in cancer histo-pathology its functions are poorly understood. The aim of this project is to improve understanding of Ki-67 functions and its requirements in cancer initiation and progression. We found that Ki-67 is dispensable for cell proliferation and Ki-67 mutant mice did not exhibit any developmental abnormalities, and were fertile and aged well. Although Ki-67 was uncoupled from cell proliferation, Ki-67 was found to promote heterochromatin organization in proliferating cells. Studying Ki-67 expression control, we have found that cell cycle regulation accounts for Ki-67 variability levels in normal human cells, proliferating tissues in mice, human cancer cell lines and caner patients.Using our Ki-67 mutant mice, we found that Ki-67 depletion can protect mice from intestinal carcinogenesis in two different experimental models used. Moreover, analysis of the consequence of Ki-67 ablation in the mouse breast cancer cell line, 4T1 has revealed its requirements for the maintenance of the stem-like proprieties of these cancer cells. More importantly, Ki-67 depletion strongly affects 4T1 tumour growth and formation of lung metastases in vivo. Similarly, Ki-67 absence strongly impaired the development of the TNBC-derived MDA-MB-231 xenografts in vivo. Moreover, comparison of Ki-67 dependent alterations in gene expression in 4T1 cells by RNA sequencing revealed widespread transcriptome changes following Ki-67 depletion. Together, these results suggest a specific involvement of Ki-67 in cancer initiation and progression and may constitute a potential therapeutic target in cancer therapy.

Cette thèse porte sur les médiums (tâng-ki, « l’enfant de divination »), leurs différentes pratiques et évolution au cours de l’histoire dans la société taiwanaise. L’enjeu est d’étudier ce domaine médiumnique en employant des approches de sociologie historique et anthropologie religieuse mettant en œuvre une démarche comparative de l’image du médium et de celle perçue après une enquête de terrain. Afin de saisir « le vrai visage » du médium. D’un point de vue méthodologique, cette thèse s’appuie sur une démarche qualitative et l’usage des entretiens directs avec les médiums et médiums apprentis. Le corpus se compose de cinq chapitres. Au chapitre I il y a les travaux de recherche sur les médiums à Taïwan d’où l’exemple du professeur Lin Fu-shih. Les évènements après la libération de l’île de Taiwan ont été relatés ainsi qu’une étude sur l’histoire de la religion à Taiwan basée sur les faits historiques marquants. Pour le chapitre II, la thèse s’articule autour des éléments du terrain où il a été présenté le récit de l’enquête que l’auteur lui-même a mené ; il s’agit de la structuration des temples : exemples de temples publics et privés qui se terminent par l’entrée au temple Tshú-nai. Le chapitre III se consacre principalement à l’univers du médium aux quatre terrains enquêtés marqués par des commentaires et interprétation. Au chapitre IV l’intérêt est porté sur la présentation et description du temple Tshú-nai. Au dernier chapitre, il y a les étapes d’initiation des pratiquants de tsáo-lîng-san (se rendre sur la montagne des esprits) et les histoires annexes des médiums apprentis. / This thesis deals with mediums in taiwanese community (tâng-ki, « the child of divination »), their different practices and how they evolved with time. It aims at studying the medium field using some historical social facts and religious anthropology and a comparative approach to the general image of mediums and that obtained through a field survey to let people know the “real face” of a medium. Methodologically, this thesis relies on a qualitative approach and direct questioning of mediums and mediums to be. The corps of this paper is divided into five chapters. Chapter I is about researches on mediums in Taiwan with reference to Professor Lin Fu-shih. It also presents events which happen after the liberation of the Taiwan Island and a study on relevant historical religious facts in Taiwan. Chapter II deals with the results of a survey. The story of this survey lead by the author himself is reported in this chapter. It is mainly about how temples are structured. The study of some public and private temples will lead to the presentation of Tshú-nai temple. Chapter III specifically covers the medium’s universe discovered in the four areas of the survey followed by commentaries and interpretations. In chapter IV, the thesis focuses on the presentation and description of Tshú-nai temple. The last chapter reveals the different steps of the tsáo-lîng-san believers’ initiation (which includes going to the Spirits' Mountains) and a supplementary history of mediums to be.

O estudo das alterações neoplásicas do sistema nervoso (SN) de cães apresenta enorme desafio devido suas particularidades, entre as espécies de animais domésticos o cão Canis familiares é o que apresenta a maior ocorrência destes neoplasmas. O desenvolvimento deste trabalho visa estabelecer o estudo anátomopatológico, molecular e epidemiológico em neurooncologia veterinária, avaliando pela técnica de imuno-histoquímica marcadores da proliferação celular PCNA e Ki-67, expressão da proteína p53 e marcadores de diferenciação celular. Foram utilizados neste estudo 18 animais, da espécie canina, que deram entrada ao Serviço de Patologia vinculado ao Hospital veterinário (HOVET) e ao Departamento de Patologia da Faculdade de Medicina Veterinária e Zootecnia da Universidade de São Paulo, e que tiveram diagnóstico de neoplasia intracraniana primária confirmado, entre eles (8) meningiomas e (3) astrocitomas. Não houve predileção sexual e a idade média do aparecimento tumoral foi de 9 anos. O valor médio do índice proliferativo (IP) de meningiomas benignos para PCNA é 4,8% e Ki-67 é 2,8%, no astrocitoma fibrilar o valor do IP é para PCNA 1% e Ki-67 1%, no astrocitoma anaplásico o valor do IP é para PCNA 10% e Ki-67 5% e no xantoastrocitoma pleomórfico o valor do IP é para PCNA 20% e Ki-67 4%.Não foi detectada alterações em p53 e o IP foi útil para definir o comportamento tumoral benigno ou maligno e dois tipos tumorais que não haviam sido descritos em cães: xantoastrocitoma pleomórfico e tumor de parênquima pineal de diferenciação intermediária puderam ser diagnosticados graças ao conjunto de informações coletadas, morfologia celular, marcadores de diferenciação celular e índice proliferativo. / The study of neoplastic alterations of dogs nervous system (NS) presents huge challenges due their particularities, between domestic animals species the dog Canis familiars presents the highest of these neoplasms. The development of this research aims to establish the anatomical-pathological, molecular and epidemiology study in veterinary neuro-oncology, evaluating by Immunohistochemistry technique markers of cell proliferation PCNA and Ki-67, expression of p53 protein and markers of cell differentiation. In this study were used 18 animals, dogs, which had entry at the Pathology Service of the Veterinary Hospital (HOVET) and at the Department of pathology from Faculdade de Medicina Veterinária e Zootecnia of University of São Paulo, which had confirmed diagnose of primary intracranial neoplasia, among them (8) meningiomas and (3) astrocytomas. There wasn\'t sexual preference and the mean age of tumor manifestation was 9 years old. The mean value of labelling index (LI) of benign meningiomas for PCNA is 4,8% and Ki-67 is 2,8%, in fibrillary astrocytoma the mean value of LI is 1% for PCNA and 1% for Ki-67, in anaplastic astrocytoma the mean value of LI is 10% for PCNA and 5% for Ki-67 and in pleomorphic xanthoastrocytoma the mean value of LI is 20% for PCNA and 4% for Ki-67. The p53 alterations wasn\'t detected and the LI was useful to set the benign or malign tumor behavior and two tumor types which had not been described in dogs: pleomorphic xanthoastrocytoma and pineal parenchymal tumour of intermediate differentiation could be diagnosed through the set of information collected, cell morphology, markers of cell differentiation and labelling index.

O estudo das alterações neoplásicas do sistema nervoso (SN) de cães apresenta enorme desafio devido suas particularidades, entre as espécies de animais domésticos o cão Canis familiares é o que apresenta a maior ocorrência destes neoplasmas. O desenvolvimento deste trabalho visa estabelecer o estudo anátomopatológico, molecular e epidemiológico em neurooncologia veterinária, avaliando pela técnica de imuno-histoquímica marcadores da proliferação celular PCNA e Ki-67, expressão da proteína p53 e marcadores de diferenciação celular. Foram utilizados neste estudo 18 animais, da espécie canina, que deram entrada ao Serviço de Patologia vinculado ao Hospital veterinário (HOVET) e ao Departamento de Patologia da Faculdade de Medicina Veterinária e Zootecnia da Universidade de São Paulo, e que tiveram diagnóstico de neoplasia intracraniana primária confirmado, entre eles (8) meningiomas e (3) astrocitomas. Não houve predileção sexual e a idade média do aparecimento tumoral foi de 9 anos. O valor médio do índice proliferativo (IP) de meningiomas benignos para PCNA é 4,8% e Ki-67 é 2,8%, no astrocitoma fibrilar o valor do IP é para PCNA 1% e Ki-67 1%, no astrocitoma anaplásico o valor do IP é para PCNA 10% e Ki-67 5% e no xantoastrocitoma pleomórfico o valor do IP é para PCNA 20% e Ki-67 4%.Não foi detectada alterações em p53 e o IP foi útil para definir o comportamento tumoral benigno ou maligno e dois tipos tumorais que não haviam sido descritos em cães: xantoastrocitoma pleomórfico e tumor de parênquima pineal de diferenciação intermediária puderam ser diagnosticados graças ao conjunto de informações coletadas, morfologia celular, marcadores de diferenciação celular e índice proliferativo. / The study of neoplastic alterations of dogs nervous system (NS) presents huge challenges due their particularities, between domestic animals species the dog Canis familiars presents the highest of these neoplasms. The development of this research aims to establish the anatomical-pathological, molecular and epidemiology study in veterinary neuro-oncology, evaluating by Immunohistochemistry technique markers of cell proliferation PCNA and Ki-67, expression of p53 protein and markers of cell differentiation. In this study were used 18 animals, dogs, which had entry at the Pathology Service of the Veterinary Hospital (HOVET) and at the Department of pathology from Faculdade de Medicina Veterinária e Zootecnia of University of São Paulo, which had confirmed diagnose of primary intracranial neoplasia, among them (8) meningiomas and (3) astrocytomas. There wasn\'t sexual preference and the mean age of tumor manifestation was 9 years old. The mean value of labelling index (LI) of benign meningiomas for PCNA is 4,8% and Ki-67 is 2,8%, in fibrillary astrocytoma the mean value of LI is 1% for PCNA and 1% for Ki-67, in anaplastic astrocytoma the mean value of LI is 10% for PCNA and 5% for Ki-67 and in pleomorphic xanthoastrocytoma the mean value of LI is 20% for PCNA and 4% for Ki-67. The p53 alterations wasn\'t detected and the LI was useful to set the benign or malign tumor behavior and two tumor types which had not been described in dogs: pleomorphic xanthoastrocytoma and pineal parenchymal tumour of intermediate differentiation could be diagnosed through the set of information collected, cell morphology, markers of cell differentiation and labelling index.

Passive loading of a specimen is a relatively cheap method to use in fracture mechanical testing compared to an actively loaded specimen. For stress corrosion cracking testing it is easier to use a passively loaded specimen since the specimen easily can be placed in a specific corrosive environment. The passive method lacks information about the crack growth over time and the load can not be regulated during the test to ensure crack growth. This thesis work was mainly about finding a specimen with a region of constant KI to ensure crack growth without the need of controlling the load and to find a way to estimate the crack growth over time. The work is based on Linear Elastic Fracture Mechanics and the Finite Element Method. The thesis work resulted in a specimen with constant KI in the region 23/50 ≤ a/W ≤ 33/50 of crack growth and an equation was found to describe the relation between the crack propagation and the strain measured on the specimens back face. / Passiv belastning av en provstav är en relativt billig metod att använda för brottmekanisk provning jämfört med en aktivt belastad provstav. En passivt belastad provstav är enklare att använda vid provning av spänningskorrosion då provstaven enkelt kan placeras i en specifik korrosiv miljö. Den passiva metoden är bristande i information om spricktillväxen över tid och belastningen på provstaven kan inte justeras för att upprätthålla fortsatt spricktillväxt under provningens gång. Detta examensarbete syftade till att ta fram en provstav med ett spricktillväxtomrade med konstant KI för fortsatt spricktillväxt vid konstant belastning samt att hitta en metod för att uppskatta spricktillväxten över tid. Arbetet är baserat på linjärelastisk brottmekanik och finita elementmetoden. Arbetet resulterade i en provstav med konstant KI i spricktillväxtområdet 23/50 ≤ a/W ≤ 33/50 och en relation mellan spricktillväxt och töjningen som mäts på provstavens baksida.

Objectives: A decrease in Ki67 during neoadjuvant therapy predicts response to tamoxifen. Previous trials have shown a decreased Ki67 in breast tumors with as little as two or more weeks of preoperative tamoxifen. Shortening the preoperative treatment time in window of opportunity clinical trials makes these trials more attractive to women. POWERPIINC examined the effect of 7 days of preoperative tamoxifen on breast tumor proliferation and patient symptoms. Methods: Women with untreated stage I/II, ER-positive, invasive breast cancer with no contraindications to tamoxifen were enrolled. Women received 20 mg of tamoxifen for 7 days up to the day of surgery. Proliferation was assessed by Ki67 immunohistochemistry before and after 7 days of tamoxifen. Symptoms and QOL were assessed by the FACT-ES and MENQOL. Adherence was measured by pill counts. Results: 52 women were enrolled, and 44 were evaluable for Ki67. The median age was 58.5 years, and the median tumor diameter was 1.2 cm. Most women (73%) were post-menopausal. Most tumors were PR positive (88%) and HER2-negative (92%). The Ki67 decreased by a geometric mean of 40% (95% CI 29%-63%), and 73% (95% CI 57%-85%) of women had tumors with decreased proliferation (p = 0.0001 by paired t-test). Adherence to taking tamoxifen during the preoperative period was 100%. Women reported minimal bother from psychosocial or physical symptoms at baseline or on the day of surgery. Conclusion: Seven days of tamoxifen showed a similar relative decrease in Ki67 as that reported for longer courses, was acceptable to women, and could be considered for window of opportunity studies.

CoordenaÃÃo de AperfeiÃoamento de NÃvel Superior / O objetivo deste estudo, à avaliar a expressÃo das proteÃnas p53 e Ki-67 em adenomas colorretais, suas relaÃÃes com caracterÃsticas clinico-patolÃgicas e avaliar a relaÃÃo entre as duas proteÃnas. A amostra consistiu de 50 pÃlipos adenomatosos encontrados em pacientes submetidos a exames colonoscÃpicos. ApÃs a realizaÃÃo de polipectomia, os pÃlipos eram conservados em soluÃÃo tamponada de formalina a 10% e submetidos à rotina de preparo de cortes e lÃminas e coloraÃÃo pela hematoxilina-eosina para confirmaÃÃo da natureza adenomatosa. Realizou-se imunohistoquÃmica especÃfica para as proteÃnas p53 e Ki-67 pelo mÃtodo imunoenzimÃtico da streptoavidina-biotina-peroxidase para cada adenoma. A proteÃna p53 foi positiva em 18% dos adenomas e a proteÃna Ki-67, expresso como Ãndice (i.Ki-67), obteve mÃdia de 0,49. Houve diferenÃa estatisticamente significante na expressÃo de p53 (p=0,0003) e Ki-67(p=0,02) entre os adenomas com alto e baixo grau de displasia, sendo maior no primeiro grupo. Encontrou-se, ainda maior expressÃo da proteÃna Ki-67 nos adenomas retais em relaÃÃo aos de localizaÃÃo cÃlica (p= 0,02). NÃo houve relaÃÃo entre a expressÃo das duas proteÃnas, na amostra / Ki-67 (p=0,02) expression between adenomas with high and low grade dysplasia, greater in the first group. There was greater expression of Ki-67 protein in the rectal adenomas than colic adenomas (p=0,02). There was no relation between the expression of the two proteins in the sample.

Made available in DSpace on 2014-06-12T23:01:51Z (GMT). No. of bitstreams: 2 arquivo4255_1.pdf: 4065278 bytes, checksum: c0dabbd5dc05e703429eae598cc70d8d (MD5) license.txt: 1748 bytes, checksum: 8a4605be74aa9ea9d79846c1fba20a33 (MD5) Previous issue date: 2009 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / As displasias da laringe são lesões precursoras do carcinoma escamocelular invasivo e constituem lesões pouco comuns, pois, a maior parte dos casos é diagnosticada como carcinoma invasor. Freqüentemente, utilizam-se os critérios de cérvix uterina para graduação das displasias na prega vocal. Todavia, existem diferenças histopatológicas, patogênicas e epidemiológicas entre as lesões intraepiteliais cervicais e laríngeas. O epitélio de transição na prega vocal pode ocasionar problemas diagnósticos. Os autores procuram caracterizar as displasias acentuadas/carcinoma in situ da prega vocal quanto aos aspectos histopatológicos, morfométricos (área do epitélio e diâmetro dos núcleos) e índice proliferativo (Ki-67), comparando estes dados com os obtidos no epitélio escamoso normal e de transição. Dentre 1400 casos de carcinoma de laringe (1994 a 2006), 5 casos (0,35%) de displasia acentuada/carcinoma in situ foram estudados. O grupo controle constituiu-se de dois indivíduos masculinos, não tabagistas e não etilistas de 52 e 78 anos. Observaram-se nos pacientes com displasia predomínio do sexo masculino, na 6º década e associação com tabagismo e etilismo. Comparando-se o epitélio displásico com o epitélio normal e de transição verificou-se que foi maior a área ocupada pelo primeiro, bem como o diâmetro do núcleo das células displásicas. O diâmetro dos núcleos por camada mostrou diferenças significativas no epitélio escamoso normal, mas não no epitélio displásico e de transição. O índice proliferativo foi maior no epitélio displásico que no escamoso normal e de transição com núcleos corados na camada basal/parabasal no epitélio escamoso normal e de transição e em todos os níveis no epitélio displásico

Cell proliferation plays an important role in multistage chemical carcinogenesis. Again, several reports demonstrated that upregulation of metallothionein (MT) expression is associated with increased cell proliferation that may contribute to the pathogenesis of preneoplastic phenotype to frank malignancy. In this study, we evaluated the roles of early DNA damage, altered expressions of liver MT and Ki-67 nuclear antigen, and altered hepatic levels of zinc (Zn) and copper (Cu) on cell proliferation and the progression of hepatocarcinogenesis through premalignant, late premalignant and malignant transformation phases in male Sprague-Dawley rats. We have further studied the association between MT expression and cell proliferation in hepatocarcinogenesis. There was substantial induction of DNA single-strand breaks (SSBs) (P < 0.001) and development of hepatocellular premalignant lesions along with significant decrease in hepatic levels of Zn and increase in Cu content following a single, necrogenic, intraperitoneal (i.p.) injection (200 mg/Kg body weight) of diethylnitrosamine (DEN) at week 4 of the experimental protocol. Moreover, DEN + phenobarbital (PB)-treatment significantly elevated MT-, Ki-67-, and BrdU-immunoexpressions along with their immunolabeling indices. Furthermore, positive correlations between MT- and Ki-67- labeling (P = 0.0006) at various time intervals, as well as, between MT immunoreactivity and 5'-bromo-2'-deoxyuridine-labeling index (BrdU-LI) (P = 0.0007) indicate that, MT expression might be associated with Ki-67 expression and cell proliferation thereby. The study suggests that DEN treatment may lead to alteration of Zn and Cu levels resulting in early DNA damage along with elevation of MT expression that may ultimately lead to hepatic cell proliferation. The results thus provide evidence in support of the role of MT as a potential positive regulator of cell growth during the early stages of hepatocellular transformation in rats.

A model's response for a given input pattern depends on the seen patterns in the training data. The larger the amount of training data, the more likely edge cases are covered during training. However, the more complex input patterns are, the larger the model has to be. For very simple use cases, a relatively small model can achieve very high test accuracy in a matter of minutes. On the other hand, a large model has to be trained for multiple days. The actual time to develop a model of that size can be considered to be even greater since often many different architecture types and hyper-parameter configurations have to be tried. An extreme case for a large model is the recently released GPT-3 model. This model consists of 175 billion parameters and was trained using 45 terabytes of text data. The model was trained to generate text and is able to write news articles and source code based only on a rough description. However, a model like this is only creatable for researchers with access to special hardware or immense amounts of data. Thus, it is desirable to find less resource-intensive training approaches to enable other researchers to create well performing models. This thesis investigates the use of pre-trained models. If a model has been trained on one dataset and is then trained on another similar data, it faster learns to adjust to similar patterns than a model that has not yet seen any of the task's pattern. Thus, the learned lessons from one training are transferred to another task. During pre-training, the model is trained to solve a specific task like predicting the next word in a sequence or first encoding an input image before decoding it. Such models contain an encoder and a decoder part. When transferring that model to another task, parts of the model's layers will be removed. As a result, having to discard fewer weights results in faster training since less time has to be spent on training parts of a model that are only needed to solve an auxiliary task. Throughout this thesis, the concept of siamese architectures will be discussed since when using that architecture, no parameters have to be discarded when transferring a model trained with that approach onto another task. Thus, the siamese pre-training approach positively impacts the need for resources like time and energy use and drives the development of new models in the direction of Green AI. The models trained with this approach will be evaluated by comparing them to models trained with other pre-training approaches as well as large existing models. It will be shown that the models trained for the tasks in this thesis perform as good as externally pre-trained models, given the right choice of data and training targets: It will be shown that the number and type of training targets during pre-training impacts a model's performance on transfer learning tasks. The use cases presented in this thesis cover different data from different domains to show that the siamese training approach is widely applicable. Consequently, researchers are motivated to create their own pre-trained models for data domains, for which there are no existing pre-trained models. / Die Vorhersage eines Models hängt davon ab, welche Muster in den während des Trainings benutzen Daten vorhanden sind. Je größer die Menge an Trainingsdaten ist, desto wahrscheinlicher ist es, dass Grenzfälle in den Daten vorkommen. Je größer jedoch die Anzahl der zu lernenden Mustern ist, desto größer muss jedoch das Modell sein. Für einfache Anwendungsfälle ist es möglich ein kleines Modell in wenigen Minuten zu trainieren um bereits gute Ergebnisse auf Testdaten zu erhalten. Für komplexe Anwendungsfälle kann ein dementsprechend großes Modell jedoch bis zu mehrere Tage benötigen um ausreichend gut zu sein. Ein Extremfall für ein großes Modell ist das kürzlich veröffentlichte Modell mit dem Namen GPT-3, welches aus 175 Milliarden Parametern besteht und mit Trainingsdaten in der Größenordnung von 45 Terabyte trainiert wurde. Das Modell wurde trainiert Text zu generieren und ist in der Lage Nachrichtenartikel zu generieren, basierend auf einer groben Ausgangsbeschreibung. Solch ein Modell können nur solche Forscher entwickeln, die Zugang zu entsprechender Hardware und Datenmengen haben. Es demnach von Interesse Trainingsvorgehen dahingehend zu verbessern, dass auch mit wenig vorhandenen Ressourcen Modelle für komplexe Anwendungsfälle trainiert werden können. Diese Arbeit beschäfigt sich mit dem Vortrainieren von neuronalen Netzen. Wenn ein neuronales Netz auf einem Datensatz trainiert wurde und dann auf einem zweiten Datensatz weiter trainiert wird, lernt es die Merkmale des zweiten Datensatzes schneller, da es nicht von Grund auf Muster lernen muss sondern auf bereits gelerntes zurückgreifen kann. Man spricht dann davon, dass das Wissen transferiert wird. Während des Vortrainierens bekommt ein Modell häufig eine Aufgabe wie zum Beispiel, im Fall von Bilddaten, die Trainingsdaten erst zu komprimieren und dann wieder herzustellen. Bei Textdaten könnte ein Modell vortrainiert werden, indem es einen Satz als Eingabe erhält und dann den nächsten Satz aus dem Quelldokument vorhersagen muss. Solche Modelle bestehen dementsprechend aus einem Encoder und einem Decoder. Der Nachteil bei diesem Vorgehen ist, dass der Decoder lediglich für das Vortrainieren benötigt wird und für den späteren Anwendungsfall nur der Encoder benötigt wird. Zentraler Bestandteil in dieser Arbeit ist deswegen das Untersuchen der Vorteile und Nachteile der siamesische Modellarchitektur. Diese Architektur besteht lediglich aus einem Encoder, was dazu führt, dass das Vortrainieren kostengünstiger ist, da weniger Gewichte trainiert werden müssen. Der wesentliche wissenschaftliche Beitrag liegt darin, dass die siamische Architektur ausführlich verglichen wird mit vergleichbaren Ansätzen. Dabei werden bestimmte Nachteile gefunden, wie zum Beispiel dass die Auswahl einer Ähnlichkeitsfunktion oder das Zusammenstellen der Trainingsdaten große Auswirkung auf das Modelltraining haben. Es wird erarbeitet, welche Ähnlichkeitsfunktion in welchen Kontexten empfohlen wird sowie wie andere Nachteile der siamischen Architektur durch die Anpassung der Trainingsziele ausgeglichen werden können. Die entsprechenden Experimente werden dabei auf Daten aus unterschiedlichen Domänen ausgeführt um zu zeigen, dass der entsprechende Ansatz universell anwendbar ist. Die Ergebnisse aus konkreten Anwendungsfällen zeigen außerdem, dass die innerhalb dieser Arbeit entwickelten Modelle ähnlich gut abschneiden wie extern verfügbare Modelle, welche mit großem Ressourcenaufwand trainiert worden sind. Dies zeigt, dass mit Bedacht erarbeitete Architekturen die benötigten Ressourcen verringern können.