Search

Das maschinelle Lernen hat seinen Erfolg bei verschiedenen Problemen in vielen unterschiedlichen Bereichen bewiesen. Verschiedene Algorithmen für maschinelles Lernen verwenden unterschiedliche Ansätze, um die zugrunde liegenden Muster in den Daten zu erfassen. Auch wenn die Menge der Daten bei den verschiedenen Algorithmen für maschinelles Lernen unterschiedlich ist, benötigen sie doch eine ausreichende Menge an Daten, um diese Muster zu erkennen. Eine der einfachsten Möglichkeiten, diesen Bedarf der Algorithmen für maschinelles Lernen zu decken, ist die Verwendung mehrerer Quellen, die die gleiche Art von Daten erzeugen. Eine solche Lösung ist machbar, wenn man bedenkt, dass die Geschwindigkeit der Datengenerierung und die Anzahl der Quellen, die diese Daten generieren, parallel zu den Entwicklungen in der Technologie gestiegen sind. Der Wunsch der Algorithmen des maschinellen Lernens nach Daten kann mit Hilfe dieser Quellen leicht erfüllt werden. Dies kann jedoch zu einer Beeinträchtigung der Privatsphäre führen. Die von diesen Quellen erzeugten Daten können sensible Informationen enthalten, die für unerwünschte Zwecke verwendet werden können. Obwohl die Algorithmen für maschinelles Lernen Daten benötigen, sind die Quellen daher möglicherweise nicht bereit, ihre Daten weiterzugeben. Ein ähnliches Dilemma tritt auf, wenn der/die Dateneigentümer*in mit Hilfe von Algorithmen für maschinelles Lernen nützliche Informationen aus den Daten extrahieren möchte, aber nicht über genügend Rechenleistung oder Wissen verfügt. In diesem Fall kann diese Aufgabe möglicherweise an externe Parteien ausgelagert werden, die Algorithmen für maschinelles Lernen als Dienstleistung anbieten. Auch in diesem Fall können die sensiblen Informationen in den Daten der entscheidende Faktor für den/die Eigentümer*in sein, sich nicht für eine Auslagerung zu entscheiden, was dann dazu führt, dass die Daten für den/die Eigentümer*in nicht genutzt werden. Um diese Art von Dilemmata und Problemen anzugehen, zielt diese Arbeit darauf ab, eine vollständige Pipeline für maschinelles Lernen unter Wahrung der Privatsphäre zu entwickeln. Es werden mehrere Studien vorgestellt, die sich mit verschiedenen Phasen der Pipeline befassen, so dass alle Phasen eines Algorithmus für maschinelles Lernen unter Wahrung der Privatsphäre durchgeführt werden können. Eine dieser Phasen, die in dieser Arbeit behandelt wird, ist das Training eines maschinellen Lernalgorithmus. Das Training von kernbasierten maschinellen Lernalgorithmen unter Wahrung der Privatsphäre wird in verschiedenen Arbeiten mit unterschiedlichen kryptographischen Techniken behandelt, von denen eine ein von aus entwickeltes neuartiges Verschlüsselungsverfahren ist. Diese haben jeweils unterschiedliche Vorteile gegenüber den anderen. Darüber hinaus werden in dieser Arbeit Studien vorgestellt, die sich mit der Testphase nicht nur kernelbasierter maschineller Lernalgorithmen befassen, sondern auch mit einem speziellen Typ rekurrenter neuronaler Netze, nämlich den rekurrenten Kernnetzen, das die erste Studie ist, die eine solche Inferenz durchführt, ohne die Privatsphäre zu gefährden. Um eine datenschutzkonforme Inferenz auf rekurrenten Kernnetzen zu ermöglichen, wird in dieser Arbeit ein Framework mit dem Namen CECILIA eingeführt, das zwei neuartige Funktionen enthält, nämlich die Exponentialfunktion und die inverse Quadratwurzel der Gram-Matrix, sowie effiziente Versionen etablierter Funktionen, Multiplexer und least significant bit. Unter Verwendung dieses Frameworks und anderer Ansätze in den entsprechenden Studien ist es möglich, datenschutzkonforme Inferenzen für verschiedene vortrainierte Algorithmen des maschinellen Lernens durchzuführen. Neben dem Training und Testen von maschinellen Lernalgorithmen unter Wahrung der Privatsphäre wird in dieser Arbeit auch eine Studie vorgestellt, die darauf abzielt, die Leistung von maschinellen Lernalgorithmen zu bewerten, ohne die Privatsphäre zu gefährden. In dieser Arbeit wird CECILIA eingesetzt, um die Berechnung der Fläche unter der Kurve für zwei verschiedene kurrenbasierte Auswertungen, nämlich die Receiver-Operating-Characteristic-Kurve und die Precision-Recall-Kurve, auf eine datenschutzfreundliche Weise zu realisieren. Alle vorgeschlagenen Ansätze werden anhand verschiedener Aufgaben des maschinellen Lernens auf ihre Korrektheit geprüft und auf ihre Skalierbarkeit mit den Parametern des entsprechenden Systems/Algorithmus unter Verwendung synthetischer Daten untersucht. Die Ergebnisse zeigen, dass das Training und Testen von kernbasierten maschinellen Lernalgorithmen unter Wahrung der Privatsphäre mit verschiedenen Einstellungen möglich ist und, dass die Inferenz mit einem vortrainierten rekurrenten Kernnetzwerk unter Verwendung von CECILIA möglich ist. Darüber hinaus ermöglicht CECILIA auch die exakte Berechnung der Fläche unter der Kurve, um die Leistung eines maschinellen Lernalgorithmus zu bewerten, ohne die Privatsphäre zu beeinträchtigen. Machine learning has proven its success on various problems from many different domains. Different machine learning algorithms use different approaches to capture the underlying patterns in the data. Even though the amount varies between the machine learning algorithms, they require sufficient amounts of data to recognize those patterns. One of the easiest ways to meet this need of the machine learning algorithms is to use multiple sources generating the same type of data. Such a solution is feasible considering that the speed of data generation and the number of sources generating these data have been increasing in parallel to the developments in technology. One can easily satisfy the desire of the machine learning algorithms for data using these sources. However, this can cause a privacy leakage. The data generated by these sources may contain sensitive information that can be used for undesirable purposes. Therefore, although the machine learning algorithms demand for data, the sources may not be willing or even allowed to share their data. A similar dilemma occurs when the data owner wants to extract useful information from the data by using machine learning algorithms but it does not have enough computational power or knowledge. In this case, the data source may want to outsource this task to external parties that offer machine learning algorithms as a service. Similarly, in this case, the sensitive information in the data can be the decisive factor for the owner not to choose outsourcing, which then ends up with non-utilized data for the owner. In order to address these kinds of dilemmas and issues, this thesis aims to come up with a complete privacy preserving machine learning pipeline. It introduces several studies that address different phases of the pipeline so that all phases of a machine learning algorithm can be performed privately. One of these phases addressed in this thesis is training of a machine learning algorithm. The privacy preserving training of kernel-based machine learning algorithms are addressed in several different works with different cryptographic techniques, one of which is a our newly developed encryption scheme. The different techniques have different advantages over the others. Furthermore, this thesis introduces our study addressing the testing phase of not only the kernel-based machine learning algorithms but also a special type of recurrent neural network, namely recurrent kernel networks, which is the first study performing such an inference, without compromising privacy. To enable the privacy preserving inference on recurrent kernel networks, this thesis introduces a framework, called CECILIA, with two novel functions, which are the exponential and the inverse square root of the Gram matrix, and efficient versions of the existing functions, which are the multiplexer and the most significant bit. Using this framework and other approaches in the corresponding studies, it is possible to perform privacy preserving inference on various pre-trained machine learning algorithms. Besides the training and testing of machine learning algorithms in a privacy preserving way, this thesis also presents a work that aims to evaluate the performance of machine learning algorithms without sacrificing privacy. This work employs CECILIA to realize the area under curve calculation for two different curve-based evaluations, namely the receiver operating characteristic curve and the precision-recall curve, in a privacy preserving manner. All the proposed approaches are shown to be correct using several machine learning tasks and evaluated for the scalability of the parameters of the corresponding system/algorithm using synthetic data. The results show that the privacy preserving training and testing of kernel-based machine learning algorithms is possible with different settings and the privacy preserving inference on a pre-trained recurrent kernel network is feasible using CECILIA. Additionally, CECILIA also allows the exact area under curve computation to evaluate the performance of a machine learning algorithm without compromising privacy.

This study revealed the meaning of the experiences of Roma mothers, having children studying at the primary education level, regarding their children's schools to contribute to the determination and solution of the educational problems faced by Romas in Turkey. This study uses a phenomenological design, which is a qualitative research method. The participants were 11 Roma mothers living in a Roma neighborhood in Istanbul. The participants were selected using the criterion sampling technique. Research data were collected through face-to-face and focus group interviews and analyzed using content analysis. The results revealed that Romani mothers experienced symbolic violence, discrimination, and microassault behaviors in their children's schools. Symbolic violence is institutionally practiced. School administrators and teachers as well as non-Roma parents and students practice discriminatory and microassault behaviors. Roma mothers, perhaps out of desperation, continue to see education as a means to escape from their poor lives., Bu araştırmada Türkiye’de Romanların yaşadıkları eğitim sorunlarının tespiti ve çözümüne katkıda bulunabilmek için ilköğretim kademesinde çocuğu bulunan Roman annelerin veli olarak çocuklarının okullarıyla ilgili deneyimlerinin anlamını ortaya koymak amaçlanmıştır. Araştırmada nitel araştırma yöntemlerinden olan fenomenolojik desen kullanılmıştır. Katılımcılar İstanbul’da bir Roman mahallesinde yaşayan on bir Roman annedir. Katılımcılar ölçüt örnekleme tekniği kullanılarak belirlenmiştir. Araştırma verileri yüz yüze ve odak grup görüşmeler ile toplanmış ve içerik analizi ile analiz edilmiştir. Araştırma sonucunda Roman annelerin, çocuklarının okulunda sembolik şiddet, ayrımcı ve mikro saldırgan davranışları deneyimledikleri sonucuna ulaşılmıştır. Sembolik şiddet kurumsal olarak, ayrımcı ve mikro saldırgan davranışlar hem okul yönetimi ve öğretmenler hem de Roman olmayan veli ve öğrenciler tarafından uygulanmaktadır. Roman anneler belki de çaresizlikten, eğitimi yaşadıkları yoksul hayattan kurtulmanın yolu olarak görmeye umutsuzca devam etmektedirler.

Background: The COVID-19 pandemic process, which endangers the world, continues to threaten all countries. In this process, many activities were suspended due to social distance and isolation. This research aims to have information about the efficiency of the process by analyzing the opinions of students regarding online education.Methods: The Logic model was used to build the evaluation. The evaluation components are resources, activities, outputs, outcomes and impact. The feedback of 521 students in 2019-2020 period were evaluated. Students’ questionnaire form consisted of 17 questions. Also, data from students was collected by a questionnaire composed of open-ended questions. Results: Logic model is an useful model for program development and evaluation. . In this process, exam reliability has been increased. Students think that face-to-face education is more beneficial than online education.Conclusion: The results of this evaluation showed useful information for the curriculum., Giriş: Dünyayı tehlike altına alan COVİD-19 pandemi süreci tüm ülkeleri tehdit etmeye devam etmektedir. Bu süreç içinde sosyal mesafe ve izolasyon sebebiyle birçok faaliyete ara verilmiştir. Bu araştırma, söz konusu pandemi sürecinde yüz yüze eğitime ara verilmesi sonucunda Akdeniz Üniversitesinde uygulanan uzaktan eğitime ilişkin değerlendirilmesini amaçlamaktadır. Yöntem: Değerlendirmede Logic model kullanılmıştır. Logic model, kaynak, etkinlikler, ürün ve çıktılardan oluşmaktadır. 2019-2020 yılları arasında 521 öğrenciden alınan geribildirimler değerlendirilmiştir. Öğrenci anketi 17 sorudan oluşmaktadır. Ayrıca öğrencilerin verileri “açık uçlu sorulardan oluşan bir anket” yoluyla toplanmıştır.Bulgular: Çalışmamızda yapılan program değerlendirilmesi ile Covid 19 salgını süresince Akdeniz Üniversitesi Tıp Fakültesinde 2020-2021 yılları arasında planlı uzaktan eğitim uygulamalarının uygulanabilirliğini gösterilmiştir. Bu dönemde sınav güvenliği arttırılmıştır. Öğrenciler yüz yüze eğitimin çevrimiçi eğitime göre daha yararlı olacağını düşünmektedir. Sonuç: Bu değerlendirmenin sonuçları programın bütünü açısından planlanacak eğitimler için yararlı bilgiler içermektedir.

© 2022, UHOD - Uluslararasi Hematoloji Onkoloji Dergisi. All rights reserved.Diffuse large B-cell lymphoma(DLBCL) is the prototype of aggressive lymphomas. The abnormalities such as translocation, deletion, duplication are stated to affect the progress and the prognosis of the disease. This study investigates the presence and impact on the prognosis of K-ras mutation and C-myc overexpression in patients with DLBCL. Twenty eight patients diagnosed with DLBCL and 29 healthy controls, whose diagnostic lymph node biopsy was reactive, were enrolled in this study. Paraffin blocks were analyzed for K-ras mutations and C-myc expressions. The effects of C-myc and K-ras on overall and disease-free survival were evaluated. Of the 28 patients, three (10.7%) K-ras mutations were detected. K-ras mutation (+) cases have an average survival time of 28 months, while those without the mutation were 49 months. The average duration of disease-free survival was 17.6 months in patients with K-ras mutation, while it has been identified as 45 months in patients without mutations (p= 0.072). C-myc overexpression in 17 patients(61%) was detected. Median overall survival was 33 months in male patients, while it was 65 months in female patients (p= 0.006). The median disease-free survival time was 23 months in male patients, although 64 months in female patients (p= 0.003). Complete remission rate was 30% in C-myc overexpression group, while 69% in remaining group with the cut-off 2.48 (p= 0.02). The presence of K-ras and C-myc overexpression in patients with DLBCL does not affect overall and disease-free survival. C-myc overexpression harms complete remission. Gender was found to be influential on the overall and disease-free survival; meanwhile, the female gender was a better prognosis.

© 2022 by Turkish Society of Hematology Turkish Journal of Hematology, Published by Galenos Publishing House.Objective: This study aimed to evaluate the efficacy and safety of eltrombopag (ELT) in the treatment of thrombocytopenia following hematopoietic stem cell transplantation (HSCT). Materials and Methods: Forty-eight patients treated with ELT for thrombocytopenia after allogeneic or autologous transplantation at the Erciyes University Bone Marrow Transplantation Center between July 2017 and July 2021 were evaluated retrospectively. Results: Forty-eight HSCT recipients were included in this study. Thirty (62.5%) patients were evaluated as having experienced delayed platelet recovery (DPR) and 18 (37.5%) patients as having experienced secondary failure of platelet recovery (SFPR). The median platelet count before ELT treatment was 13x109/L (range: 3-20x109/L). Twenty-three patients responded to treatment and the cumulative incidence of successful platelet recovery was 48%. Patients with both DPR and SFPR responded, but patients with DPR had a higher response rate (50% vs. 44%). The median platelet count of the 23 responding patients was 12x109/L (5-19x109/L) before treatment and 68x109/L (52-266x109/L) after treatment (p

Objective: Nutritional status assessment, dysphagia evaluation and enteral feeding decision are important determinants of prognosis in acute neurovascular diseases. Materials and Methods: NöroTek is a point prevalence study conducted with the participation of 87 hospitals spread across all health sub regions of Turkey conducted on 10-May-2018 (World Stroke Awareness Day). A total of 972 hospitalized neurovascular patients [female: 53%, age: 69±14; acute ischemic stroke in 845; intracerebral hematoma (ICH) in 119 and post-resuscitation encephalopathy (PRE) in 8] with complete data were included in this sub-study. Results: Gastrostomy was inserted in 10.7% of the patients with ischemic stroke, 10.1% of the patients with ICH and in 50% of the patients with PRE. Independent predictors of percutaneous endoscopic gastrostomy (PEG) administration were The National Institutes of Health Stroke Scale score at admission [exp (β): 1.09 95% confidence interval (CI): 1.05-1.14, per point] in ischemic stroke; and mechanical ventilation in ischemic [exp (β): 6.18 (95% CI: 3.16-12.09)] and hemorrhagic strokes [exp (β): 26.48 (95% CI: 1.36-515.8)]. PEG was found to be a significant negative indicator of favorable (modified Rankin’s scale score 0-2) functional outcome [exp (β): 0.032 (95% CI: 0.004-0.251)] but not of in-hospital mortality [exp (β): 1.731 (95% CI: 0.785-3.829)]. Nutritional and swallowing assessments were performed in approximately two-thirds of patients. Of the nutritional assessments 69% and 76% of dysphagia assessments were completed within the first 2 days. Tube feeding was performed in 39% of the patients. In 83.5% of them, tube was inserted in the first 2 days; 28% of the patients with feeding tube had PEG later. Conclusion: The NöroTek study provided the first reliable and large-scale data on key quality metrics of nutrition practice in acute stroke in Turkey. In terms of being economical and accurate it makes sense to use the point prevalence method. Amaç: Akut nörovasküler hastalıklarda nütrisyonel durum ve disfaji değerlendirmesi ve enteral beslenme kararı önemli prognoz belirleyicilerindendir. Gereç ve Yöntem: NöroTek, 10 Mayıs 2018’de (Dünya İnme Farkındalık Günü) Türkiye’nin tüm sağlık alt bölgelerine yayılmış 87 hastanenin katılımıyla gerçekleştirilen bir nokta prevalans çalışmasıdır. Hastanede yatan ve bu alt çalışma için toplanan verisi tam olan toplam 972 nörovasküler hasta (kadın: %53, yaş: 69±14 yıl; 845’i akut iskemik inme; 119’u intraserebral hematom ve 8’i post-resüsitasyon ensefalopatisi) analiz edildi. Bulgular: Gastrostomi iskemik inmeli hastaların %10,7, intraserebral kanamalıların %10,1 ve post-resusitasyon ensefalopatisi olanların %50’sine uygulanmıştır. Perkütan endoskopik gastrostomi (PEG) gereksiniminin bağımsız belirleyicileri, iskemik inme grubunda kabul NIHSS [exp (β): 1,09, %95 güven aralığı (GA): 1,05-1,14, puan başına] ile hem iskemik hem de hemorajik inmelerde mekanik ventilasyon uygulanmış olmasıdır [iskemik için: exp (β): 6,18, %95 GA: 3,16- 12,09] ve hemorajik inme için: [exp (β): 26,48, 95% GA: 1,36-515,8]. İnme olgularında PEG uygulaması hastane içi mortalite için bağımsız belirleyici değildi [exp (β): 1,731, 95% GA: 0,785-3,829]. Ancak, PEG uygulanmış olması taburculuk esnasında iyi prognoza (modifiye Rankin skoru 0-2) sahip olabilme için anlamlı bir negatif etmen olarak bulundu [exp (β): 0,032, %95 GA: 0,004-0,251]. Hastanede yatan nörovasküler hastaların yaklaşık üçte ikisinde malnütrisyon ve yutma bozukluğu açısından değerlendirme yapılmıştı. Nutrisyonel status değerlendirmesinin %69’u ve disfaji değerlendirmesinin %76’sı ilk 48 saat içinde gerçekleştirilmişti. Tüple enteral nütrisyon uygulama oranı %39’du. Beslenme tüplerinin %83,5’i ilk 2 gün içinde yerleştirilirken beslenme tüpü olan hastaların %28’ine daha sonra PEG açılmıştı. Sonuç: NöroTek çalışması ile Türkiye’de hastanede yatan akut inme hastalarında nutrisyonel uygulamaların temel kalite ölçütlerine ilişkin ilk güvenilir ve büyük ölçekli veri sağlanmıştır. Ekonomik olması ve doğruluğu açısından nokta yaygınlık yönteminin bu tip verilerin temini için daha fazla kullanılması mantıklıdır.

© 2022 by Turkish Society of Hematology Turkish Journal of Hematology, Published by Galenos Publishing House.Objective: Studies comparing the efficacy and safety of prophylactic regimens for central nervous system (CNS) involvement in acute lymphoblastic leukemia (ALL) are scarce in adults. This multicenter retrospective study aimed to compare the efficacy of prophylactic regimens with and without CNS irradiation on the development of CNS relapse during follow-up. Materials and Methods: This was a multicenter comparative cohort study. A total of 203 patients were included from four tertiary care centers in Turkey. Patients were divided into two groups according to whether they received CNS irradiation or not. The groups were analyzed retrospectively regarding patient and disease characteristics, with the main focus being CNS relapse. Results: While 105 patients received chemotherapy-based prophylaxis, 98 patients received additional CNS irradiation. These groups were statistically comparable in terms of demographic characteristics and risk factors for CNS involvement. In the irradiation group, patients were younger and had more stem cell transplants. In a median of 23.8 (11.1-62.4) months, there was no difference between the two groups regarding CNS relapse-free survival (log-rank p=0.787). Conclusion: Craniospinal irradiation may not be indispensable for every adult patient with ALL, similarly to pediatric patients. It is crucial to avoid the long-term toxicities of radiation, especially in patients with long life expectancy. Craniospinal irradiation may be reserved for therapeutic use in cases of CNS relapse and prophylaxis for some high-risk patients.