Publications without Fulltext
Permanent URI for this collectionhttps://hdl.handle.net/20.500.14288/3
Browse
617 results
Filters
Advanced Search
Filter by
Settings
Search Results
Publication Metadata only Self-supervised object-centric learning for videos(Neural information processing systems foundation, 2023) Xie, Weidi; Department of Computer Engineering; Aydemir, Görkay; Güney, Fatma; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of EngineeringUnsupervised multi-object segmentation has shown impressive results on images by utilizing powerful semantics learned from self-supervised pretraining. An additional modality such as depth or motion is often used to facilitate the segmentation in video sequences. However, the performance improvements observed in synthetic sequences, which rely on the robustness of an additional cue, do not translate to more challenging real-world scenarios. In this paper, we propose the first fully unsupervised method for segmenting multiple objects in real-world sequences. Our object-centric learning framework spatially binds objects to slots on each frame and then relates these slots across frames. From these temporally-aware slots, the training objective is to reconstruct the middle frame in a high-level semantic feature space. We propose a masking strategy by dropping a significant portion of tokens in the feature space for efficiency and regularization. Additionally, we address over-clustering by merging slots based on similarity. Our method can successfully segment multiple instances of complex and high-variety classes in YouTube videos.Publication Metadata only Spatial and thermal aware methods for efficient workload management in distributed data centers(Elsevier B.V., 2024) N/A; Department of Computer Engineering; Ali, Ahsan; Özkasap, Öznur; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of EngineeringGeographically distributed data centers provide facilities for users to fulfill the demand of storage and computations, where most of the operational cost is due to electricity consumption. In this study, we address the problem of energy consumption of cloud data centers and identify key characteristics of techniques proposed for reducing operational costs, carbon emissions, and financial penalties due to service level agreement (SLA) violations. By considering computer room air condition (CRAC) units that utilize outside air for cooling purposes as well as temperature and space-varying properties, we propose the energy cost model which takes into account temperature ranges for cooling purposes and operations of CRAC units. Then, we propose spatio-thermal-aware algorithms to manage workload using the variation of electricity price, locational outside and within the data center temperature, where the aim is to schedule the incoming workload requests with minimum SLA violations, cooling cost, and energy consumption. We analyzed the performance of our proposed algorithms and compared the experimental results with the benchmark algorithms for metrics of interest including SLA violations, cooling cost, and overall operations cost. Modeling, experiments, and verification conducted on CloudSim with realistic data center scenarios and workload traces show that the proposed algorithms result in reduced SLA violations, save between 15% to 75% of cooling cost and between 3.89% to 39% of the overall operational cost compared to the existing solutions.Publication Metadata only Hyperspectral image denoising via self-modulating convolutional neural networks(Elsevier B.V., 2024) Torun, Orhan; Yuksel, Seniha Esen; Imamoglu, Nevrez; Department of Computer Engineering; Erdem, Aykut; Department of Computer Engineering; College of EngineeringCompared to natural images, hyperspectral images (HSIs) consist of a large number of bands, with each band capturing different spectral information from a certain wavelength, even some beyond the visible spectrum. These characteristics of HSIs make them highly effective for remote sensing applications. That said, the existing hyperspectral imaging devices introduce severe degradation in HSIs. Hence, hyperspectral image denoising has attracted lots of attention by the community lately. While recent deep HSI denoising methods have provided effective solutions, their performance under real-life complex noise remains suboptimal, as they lack adaptability to new data. To overcome these limitations, in our work, we introduce a self-modulating convolutional neural network which we refer to as SM-CNN, which utilizes correlated spectral and spatial information. At the core of the model lies a novel block, which we call spectral self-modulating residual block (SSMRB), that allows the network to transform the features in an adaptive manner based on the adjacent spectral data, enhancing the network's ability to handle complex noise. In particular, the introduction of SSMRB transforms our denoising network into a dynamic network that adapts its predicted features while denoising every input HSI with respect to its spatio-spectral characteristics. Experimental analysis on both synthetic and real data shows that the proposed SM-CNN outperforms other state-of-the-art HSI denoising methods both quantitatively and qualitatively on public benchmark datasets. Our code will be available at https://github.com/orhan-t/SM-CNN.Publication Metadata only The role of endometrial sampling before Hysterectomy in premenopausal women with abnormal uterine bleeding(MDPI, 2024) Kuru, Oguzhan; Erkan, Ipek Betul Ozcivit; Saricoban, Cansu Turker; Akgor, Utku; Ilvan, Sennur; Department of Computer Engineering; İnan, Neslihan Gökmen; Department of Computer Engineering; College of EngineeringBackground/Objectives: An endometrial sampling is recommended for patients experiencing abnormal uterine bleeding above the age of 40 or 45. Valid risk prediction models are needed to accurately assess the risk of endometrial cancer and avoid an unnecessary endometrial biopsy in premenopausal women. We aimed to assess the necessity and usefulness of preoperative endometrial sampling by evaluating premenopausal women who underwent hysterectomy for abnormal uterine bleeding after preoperative endometrial sampling at our clinic. Methods: A retrospective analysis was conducted on 339 patients who underwent preoperative endometrial sampling and subsequently underwent hysterectomy due to abnormal uterine bleeding. Detailed gynecologic examinations, patient histories, and reports of endometrial sampling and hysterectomy were recorded. Cohen's Kappa (kappa) statistic was utilized to evaluate the concordance between histopathological results from an endometrial biopsy and hysterectomy. Results: The mean age of the cohort was 47 +/- 4 years. Endometrial biopsies predominantly revealed benign findings, with 137 (40.4%) cases showing proliferative endometrium and 2 (0.6%) cases showing endometrial cancer. Following hysterectomy, final pathology indicated proliferative endometrium in 208 (61.4%) cases, with 7 (2.1%) cases showing endometrioid cancer. There was a statistically significant but low level of concordance between histopathological reports of endometrial biopsy and hysterectomy results (Kappa = 0.108; p < 0.001). Significant differences were observed only in the body mass index of patients based on hysterectomy results (p = 0.004). When demographic characteristics were compared with cancer incidence, smoking status and preoperative endometrial biopsy findings showed statistically significant differences (p = 0.042 and p = 0.010, respectively). Conclusions: The concordance between the pathological findings of a preoperative endometrial biopsy and hysterectomy is low. Body mass index is an important differentiating factor between benign histopathologic findings of endometrium and endometrial neoplasia. Moreover, adenomyosis was found to be associated with endometrial cancer cases. The current approach to premenopausal women with abnormal uterine bleeding, which includes a routine endometrial biopsy, warrants re-evaluation by international societies and experts.Publication Metadata only Object and relation centric representations for push effect prediction(Elsevier, 2024) Tekden, Ahmet E.; Asfour, Tamim; Uğur, Emre; Department of Computer Engineering; Erdem, Aykut; Department of Computer Engineering; College of EngineeringPushing is an essential non -prehensile manipulation skill used for tasks ranging from pre -grasp manipulation to scene rearrangement, reasoning about object relations in the scene, and thus pushing actions have been widely studied in robotics. The effective use of pushing actions often requires an understanding of the dynamics of the manipulated objects and adaptation to the discrepancies between prediction and reality. For this reason, effect prediction and parameter estimation with pushing actions have been heavily investigated in the literature. However, current approaches are limited because they either model systems with a fixed number of objects or use image -based representations whose outputs are not very interpretable and quickly accumulate errors. In this paper, we propose a graph neural network based framework for effect prediction and parameter estimation of pushing actions by modeling object relations based on contacts or articulations. Our framework is validated both in real and simulated environments containing different shaped multi -part objects connected via different types of joints and objects with different masses, and it outperforms image -based representations on physics prediction. Our approach enables the robot to predict and adapt the effect of a pushing action as it observes the scene. It can also be used for tool manipulation with never -seen tools. Further, we demonstrate 6D effect prediction in the lever -up action in the context of robot -based hard -disk disassembly.Publication Metadata only HiSEG: Human assisted instance segmentation(Elsevier Ltd, 2024) Department of Computer Engineering; Sezgin, Tevfik Metin; Department of Computer Engineering; College of Engineering; Graduate School of Sciences and EngineeringInstance segmentation is a form of image detection which has a range of applications, such as object refinement, medical image analysis, and image/video editing, all of which demand a high degree of accuracy. However, this precision is often beyond the reach of what even state-of-the-art, fully automated instance segmentation algorithms can deliver. The performance gap becomes particularly prohibitive for small and complex objects. Practitioners typically resort to fully manual annotation, which can be a laborious process. In order to overcome this problem, we propose a novel approach to enable more precise predictions and generate higher-quality segmentation masks for high-curvature, complex and small-scale objects. Our human-assisted segmentation method, HiSEG, augments the existing Strong Mask R-CNN network to incorporate human-specified partial boundaries. We also present a dataset of hand-drawn partial object boundaries, which we refer to as “human attention maps”. In addition, the Partial Sketch Object Boundaries (PSOB) dataset contains hand-drawn partial object boundaries which represent curvatures of an object's ground truth mask with several pixels. Through extensive evaluation using the PSOB dataset, we show that HiSEG outperforms state-of-the art methods such as Mask R-CNN, Strong Mask R-CNN, Mask2Former, and Segment Anything, achieving respective increases of +42.0, +34.9, +29.9, and +13.4 points in APMask metrics for these four models. We hope that our novel approach will set a baseline for future human-aided deep learning models by combining fully automated and interactive instance segmentation architectures.Publication Metadata only Developing a multimodal classroom engagement analysis dashboard for higher-education(Association for Computing Machinery, 2023) Sabuncuoglu, Alpay; Department of Computer Engineering; Sezgin, Tevfik Metin; Department of Computer Engineering; College of EngineeringDeveloping learning analytics dashboards (LADs) is a growing research interest as online learning tools have become more accessible in K-12 and higher education settings. This paper reports our multimodal classroom engagement data analysis and dashboard design process and the resulting engagement dashboard. Our work stems from the importance of monitoring classroom engagement, which refers to students' active physical and cognitive involvement in learning that influences their motivation and success in a given course. To monitor this vital facade of learning, we developed an engagement dashboard using an iterative and user-centered process. We first created a multimodal machine learning model that utilizes face and pose features obtained from recent deep learning models. Then, we created a dashboard where users can view their engagement over time and discover their learning/teaching patterns. Finally, we conducted user studies with undergraduate and graduate-level participants to obtain feedback on our dashboard design. Our paper makes three contributions by (1) presenting a student-centric, open-source dashboard, (2) demonstrating a baseline architecture for engagement analysis using our open-Access data, and (3) presenting user insights and design takeaways to inspire future LADs. We expect our research to guide the development of tools for novice teacher education, student self-evaluation, and engagement evaluation in crowded classrooms.Publication Metadata only Factors affecting Turkish medical students' pursuit of a career in neurosurgery: a single center survey study(Elsevier Inc., 2024) Çalış, Fatih; Şimşek, Abdullah Talha; Topyalın, Nur; Adam, Baha E.; Elias, Çimen; Aksu, Muhammed Emin; Aladdam, Mohammed; Gültekin, Güliz; Sorkun, Muhammet Hüseyin; Tez, Müjgan; Balak, Naci; Department of Computer Engineering; İnan, Neslihan Gökmen; Department of Computer Engineering; College of EngineeringBackground: Statistics show that over the past 2 decades, even in high-income countries, fewer and fewer students have listed neurosurgery as their top career option. Literature on medical students' pursuit of neurosurgical careers in middle- and low-income countries are scarce. The aim of this research, conducted in Turkey with a middle-income economy, was to contribute insights relevant to medical education and neurosurgery across the world. Methods: A survey was conducted with a target sample of fourth-year (167 students), fifth-year (169 students), and sixth-year (140 students) medical students (476 in total) from the Medical School at Istanbul Medeniyet University in Turkey. The response rates of the fourth-, fifth-, and sixth-year students were 62% (104/167), 53% (90/169), and 50% (70/140), respectively (in total, 266, including 147 female and 119 male). Results: In terms of the genuine intention, only 2.5% of men and 2.7% of women were committed to specializing in neurosurgery. This study further revealed that possible reasons for these students' low motivation to specialize in neurosurgery were their beliefs that in neurosurgery, the physical and psychological demands were high, and the night shifts were intense, meaning they would not have a social life or spare time for their hobbies; that morbidity/mortality were high; and that financial incentives were insufficient, especially in public institutions. Conclusion: Turkish medical students did not rank neurosurgery at the top of their career choices. Possible reasons for this are socioeconomic factors and the inadequate introduction of neurosurgery to medical students. © 2024 Elsevier Inc.Publication Metadata only Longitudinal attacks against iterative data collection with local differential privacy(Tubitak Scientific & Technological Research Council Turkey, 2024) Department of Computer Engineering; Gürsoy, Mehmet Emre; Department of Computer Engineering; College of EngineeringLocal differential privacy (LDP) has recently emerged as an accepted standard for privacy -preserving collection of users' data from smartphones and IoT devices. In many practical scenarios, users' data needs to be collected repeatedly across multiple iterations. In such cases, although each collection satisfies LDP individually by itself, a longitudinal collection of multiple responses from the same user degrades that user's privacy. To demonstrate this claim, in this paper, we propose longitudinal attacks against iterative data collection with LDP. We formulate a general Bayesian adversary model, and then individually show the application of this adversary model on six popular LDP protocols: GRR, BLH, OLR, RAPPOR, OUE, and SS. We experimentally demonstrate the effectiveness of our attacks using two metrics, three datasets, and various privacy and domain parameters. The effectiveness of our attacks highlights the privacy risks associated with longitudinal data collection in a practical and quantifiable manner and motivates the need for appropriate countermeasures.Publication Metadata only Large language models as a rapid and objective tool for pathology report data extraction(Federation Turkish Pathology Soc., 2024) Department of Computer Engineering; Bolat, Beyza; Eren, Özgür Can; Dur Karasayar, Ayşe Hümeyra; Meriçöz, Çisel Aydın; Demir, Çiğdem Gündüz; Kulaç, İbrahim; Department of Computer Engineering; Koç Üniversitesi İş Bankası Enfeksiyon Hastalıkları Uygulama ve Araştırma Merkezi (EHAM) / Koç University İşbank Center for Infectious Diseases (KU-IS CID); Koç Üniversitesi İş Bankası Yapay Zeka Uygulama ve Araştırma Merkezi (KUIS AI)/ Koç University İş Bank Artificial Intelligence Center (KUIS AI); Koç University Research Center for Translational Medicine (KUTTAM) / Koç Üniversitesi Translasyonel Tıp Araştırma Merkezi (KUTTAM); School of Medicine; Graduate School of Health Sciences; College of EngineeringMedical institutions continuously create a substantial amount of data that is used for scientific research. One of the departments with a great amount of archived data is the pathology department. Pathology archives hold the potential to create a case series of valuable rare entities or large cohorts of common entities. The major problem in creation of these databases is data extraction which is still commonly done manually and is highly laborious and error prone. For these reasons, we offer using large language models to overcome these challenges. Ten pathology reports of selected resection specimens were retrieved from electronic archives of Ko & ccedil; University Hospital for the initial set. These reports were de-identified and uploaded to ChatGPT and Google Bard. Both algorithms were asked to turn the reports in a synoptic report format that is easy to export to a data editor such as Microsoft Excel or Google Sheets. Both programs created tables with Google Bard facilitating the creation of a spreadsheet from the data automatically. In conclusion, we propose the use of AI-assisted data extraction for academic research purposes, as it may enhance efficiency and precision compared to manual data entry.