5916 research outputs found
Sort by
Federated semi-supervised medical image classification using improved inter-client relation matching
No full textTraditional centralized machine learning faces chal-lenges with 'data label scarcity' and 'data privacy', especially in medical data classification. Semi-supervised learning leverages unlabeled data, while federated learning protects data privacy by decentralizing model training. In this study, we introduce an improved inter-client relation matching algorithm for semi-supervised federated medical image classification (iFedIRM). The FedIRM algorithm defines a loss function to transfer the knowledge of disease relationships from labeled clients to unlabeled clients to extract discriminative information from unlabeled data using per-category mean feature vectors. Our approach enhances the FedIRM algorithm by incorporating per-category feature covariance matrices for better feature representation and using group-based averaging for model aggregation. We utilize the ConvNeXt architecture as the backbone, and apply a confidence threshold to filter unreliable pseudo-labels. Experiments on three medical datasets show improved accuracy, F1, and AUC metrics. Code will be available at https://github.com/msoz7/iFedIRM
Numerical simulation of combustion in a glass melting furnace: Influence of spectral radiation models
No full textThermal radiation serves as the main heat transfer mechanism in the glass melting process and plays a critical role in shaping flame behavior within the combustion chamber. To achieve accurate simulations of glass melting furnaces, precise modeling of radiation is crucial. This research examines the influence of the Rank-Correlated Spectral Line-based Weighted Sum of Gray Gases (RC-SLW) model on flame characteristics and radiative heat transfer in an industrial glass furnace. The combustion process is simulated using a steady partially premixed diffusion flamelet model, the Moss-Brookes mechanism for predicting soot formation and oxidation, and the k-ω SST turbulence model. The RCSLW spectral radiation model is incorporated into ANSYS Fluent through a user-defined function to evaluate the impact of spectral radiation on furnace performance. Two distinct cases are analyzed: in one set, a constant glass surface temperature is assumed, while the other incorporates a one-dimensional model to simulate the behavior of the molten glass layer. This comparative approach enables a more comprehensive assessment of the radiative and convective heat transfer within the furnace. Results show that RC-SLW models outperform WSGG models by capturing the measured temperature profiles but in simulations energy efficiency of the WSGG models are better than RC-SLW. Outlet heat flux increases in the RC-SLW compared to WSGG models. Temperature profiles from the RC-SLW cases better align with probe measurements, particularly near mid-furnace positions, emphasizing the model’s improved accuracy in capturing radiative heat transfer and soot radiation effects.TÜBİTA
A self-organizing winner-takes-all mechanism based on heterosynaptic plasticity
No full textBu çalışmada, biyolojik sinir sistemlerinden esinlenerek tasarlanan, heterosinaptik plastisiteye dayalı özdüzenleyici bir Kazanan-Her-Şeyi-Alır mekanizması sunulmaktadır. Önerilen modelde, sinaptik ögrenme, ateşleme zamanlamasına bağlı plastisite yöntemleri yerine kalsiyum-temelli Sinaptik Etiketleme ve Yakalama mekanizması ile gerçekleştirilmiştir. Bu sayede sinaptik güncellemeler, yalnızca lokal etkinliklere degil, aynı zamanda hücre düzeyinde sentezlenen proteinlerin seçici baglanmasına bağlı hale getirilmiştir. Oluşturulan model üç farklı örüntüyü yüksek dogrulukla sınıflandırabilen ateşleyen sinir agları üzerinde test edildi. Eğitim sürecinde postsinaptik nöronlar belirli örüntülere özgüleşmiş ve ateşleme-temelli etiketleme yöntemiyle örüntü tahmini yapılmıştır. Sonuçlar, denetimsiz, biyolojik olarak anlamlı bir ögrenme mekanizmasıyla Kazanan-Her-Şeyi-Alır benzeri ağ yapılarının etkin biçimde oluşturulabilecegini ortaya koymaktadır.In this study, a self-organizing Winner-Takes-All mechanism inspired by biological neural systems and based on heterosynaptic plasticity is proposed. Unlike spike-timingdependent plasticity, synaptic learning in the proposed model is governed by a calcium-based Synaptic Tagging and Capture mechanism. This enables synaptic updates to depend not only on local activity, but also on the selective binding of proteins synthesized at the cellular level. The model was tested on spiking neural networks capable of classifying three distinct patterns with high accuracy. During training, postsynaptic neurons became specialized for specific patterns, and pattern prediction was performed using a spike-based labeling strategy. The results demonstrate that biologically plausible, unsupervised learning mechanisms can effectively lead to the emergence of Winner- Takes-All-like network structures. © 2025 IEEE
Between right and resistance: A quantitative analysis of women's self-defense and legal legitimacy in turkey
No full textWomen's legal right to self-defense in Turkey is limited in practice, particularly under chronic abuse. This study examines how legal knowledge, attitudes, and behaviors shape self-defense. A cross-sectional survey of 392 women was analyzed to assess the impact of socio-demographic and cognitive factors. Attitudes significantly predicted behavior, while legal knowledge alone did not. Results highlight the influence of institutional distrust, stigma, and normative pressure. The study calls for feminist-informed legal reform and recognition of self-defense education as a collective right to resist gender-based violence, emphasizing structural inequality and the gap between legal rights and real-life agency
An approach to AI-supported learning in architectural education: Case of speculative space design
No full textThe study discusses integrating text-to-image artificial intelligence (AI) tools into the architectural studio. If the integration of AI is realized through narrative production, which is a combination of written and visual media, the student’s interaction with the tools can be strengthened. This approach was tested through a workshop called “AI-Supported Speculative Space Production Workshop” conducted by the authors of the article as a case study. The workshop included 12 participants and lasted for 10 days. The expected output at the end of the workshop was a storyboard consisting of sequences that narrate and visualize the designed space. The data of the case study was collected through observation, diaries written by the students and submission of all productions in the process. The case study process was evaluated and presented according to the reflectivity between the participants’ productions in the design process by using the visual analysis method. In addition, the outputs of the case study were assessed by design experts according to three criteria that are related to the research, sketches, the narrative, and the integration of them. It was observed that especially the students who used the AI tool in relation to other representations in the design process achieved more successful results. In this way, inferences were made about how text-to-image AI tools can be integrated into the architectural design studio process while understanding their limitations and potential. These approaches are expected to contribute to the effective utilization of AI in the studio.Publisher versio
Joint analysis of sQTL and Hi-C reveals spatial proximity between sQTLs and target genes in cancer tissues
No full textGene expression and regulation with or without alternative splicing are crucial for tissues and cells to function correctly. They have been studied from three almost independent perspectives at the genome level: 1- Recognition of splicing quantitative trait loci (sQTLs), 2- Expression quantitative trait loci (eQTLs) recognition, and 3- Recognition of longer-range physical chromatin interactions between genome segments which model 3D dynamics of cells and tissues. Even though the associations between eQTLs and longer-range chromatin interactions have been previously studied, a similar relationship between sQTLs and chromatin interactions has yet to be analyzed. In this case, examining whether sQTLs control the alternative splicing of their target genes' mRNA via physically interacting genome segments is crucial. We have jointly analyzed high-throughput chromatin conformation capture (Hi-C) and sQTL datasets over eight human cancer tissues. We have discovered a positive association between the number of genes having sQTLs and chromatin interaction frequency. Such a positive association still exists when we also control for eQTLs. Additionally, sQTLs and their target genes generally exist inside identical topologically associating domains (TADs). These findings are observed over the whole set of analyzed cancer types and functional subsets of the sQTL dataset, such as survival-related sQTLs. Furthermore, tissue-specific sQTLs are statistically enriched in tissue-specific frequently interacting regions (FIREs) in 6 out of 8 human cancer tissues (Chronic Myeloid Leukemia, Colon Adenocarcinoma, Acute Myeloid Leukemia, Lung Adenocarcinoma, Prostate Cancer, Sarcoma). Our sQTL and Hi-C datasets have shown the existence of closer spatial distance between sQTLs and their target genes with possible alternative splicing across several different cancer types in humans. Such a closer spatial distance also exists, independent of whether we integrate eQTLs into the analysis. We found that sQTLs regulate alternative splicing through chromatin interactions. Source code of the analysis in this research is available on https://github.com/seferlab/sqtlhic .Publisher versio
An adaptive emotion-aware strategy for human-agent negotiation: Insights from real-world human-robot experiments
No full textNegotiation is pivotal for conflict resolution in human-agent interactions, where emotional and behavioral dynamics can significantly shape the outcomes. However, many existing strategies prioritize time- or behavior-based tactics and overlook the dynamic role of emotional awareness. This paper presents the Solver Agent, which integrates real-time facial expression recognition into a hybrid strategy incorporating time- and behavior-based approaches. It is deployed on a humanoid robot with multimodal interaction capabilities (speech, gestures, facial expression analysis) to dynamically refine its bidding and concession strategies based on an opponent's emotional cues and negotiation patterns. In user studies with 28 participants, the Solver Agent achieved higher agent scores, improved socialwelfare, and faster agreements than a baseline hybrid strategy without compromising participant satisfaction. Participants also viewed the Solver Agent as more attuned to their preferences and goals. These findings highlight that embodied emotion-aware negotiation can foster equitable and efficient collaboration, pointing to new opportunities in human-agent interaction research.TÜBİTAKPublisher versio
Workplace microaggressions against lgbti plus employees in Turkey: A thematic analysis of environmental and interpersonal discrimination
No full textPurposeThe current study aimed to determine the extent and scope of microaggressions in the workplace directed towards LGBTI+ employees in Turkey.Design/methodology/approachThe research used both quantitative and qualitative data based on 664 statements made by individuals in the "The Situation of LGBTI+ Employees in the Private and Public Sector in Turkey" survey conducted between 2015 and 2020 (n = 2,695). The quantitative data consist of frequencies and the qualitative data center on answers to nine open ended questions regarding LGBTI+ individuals' experiences of discrimination in the workplace. We employed the taxonomy proposed by Nadal et al. (2010) to determine which actions constituted microaggressions and the form they took. We also conducted a critical discourse analysis of the open-ended questions where individuals described their experiences of microaggressions.FindingsMicroaggressions directed at LGBTI+ employees are pervasive in Turkey. Microaggressions largely follow the taxonomy created by Nadal et al. (2010) although we did not find microaggressions in all of the taxonomy's categories. We found that microaggressions mostly take the form of phobic language and mockery followed by heteronormativity, exoticization and disapproval. Two further categories, othering and threatening behaviors, emerged from our data.Originality/valueThis study addresses a significant gap in the literature on workplace microaggressions against LGBTI+ individuals, particularly in non-Western contexts. To our knowledge, it is the first study of its kind conducted in a non-Western Muslim-majority country. The research uniquely captures and critically analyzes the lived experiences of LGBTI+ employees through their own narratives, examining how microaggressions manifest as discriminatory discourses in the workplace.Friedrich Naumann Foundatio
Development of the usage possibilities of adobe with computational design
No full textIn addition to improving the physical material properties of adobe, the ability to use it with today's design approach also plays an important role in this material being considered a contemporary building material. The development of computer-aided design technology not only changes the architectural design concept but also improves the usage possibilities of traditional building materials. The parametric structures created with computational design allow the use of traditional materials in different ways, leading to the emergence of innovative construction methods. With its easily accessible, economical, and sustainable features, adobe is a preferred material for contemporary designs, and it meets today's building production needs. It is a necessity of our age to investigate the adobe material, which increases the indoor air quality and creates healthy spaces, as a building material of today, as well as a material of the future. This study aims to consider how the usage possibilities and production methods of Adobe material can be improved by examining the innovations brought by computational design to Adobe material via parametrically designed Adobe building projects and structural elements. It is also important to do a benchmark in this study by examining the usage of other building materials used in computational design projects and establishing a relationship between these techniques and adobe