Publications with Fulltext
Permanent URI for this collectionhttps://hdl.handle.net/20.500.14288/6
Browse
14 results
Search Results
Publication Open Access Online failure diagnosis in interdependent networks(Springer Nature, 2021) Akbari, Vahid; N/A; Shiri, Davood; PhD Student; Graduate School of Sciences and EngineeringIn interdependent networks, nodes are connected to each other with respect to their failure dependency relations. As a result of this dependency, a failure in one of the nodes of one of the networks within a system of several interdependent networks can cause the failure of the entire system. Diagnosing the initial source of the failure in a collapsed system of interdependent networks is an important problem to be addressed. We study an online failure diagnosis problem defined on a collapsed system of interdependent networks where the source of the failure is at an unknown node (v). In this problem, each node of the system has a positive inspection cost and the source of the failure is diagnosed when v is inspected. The objective is to provide an online algorithm which considers dependency relations between nodes and diagnoses v with minimum total inspection cost. We address this problem from worst-case competitive analysis perspective for the first time. In this approach, solutions which are provided under incomplete information are compared with the best solution that is provided in presence of complete information using the competitive ratio (CR) notion. We give a lower bound of the CR for deterministic online algorithms and prove its tightness by providing an optimal deterministic online algorithm. Furthermore, we provide a lower bound on the expected CR of randomized online algorithms and prove its tightness by presenting an optimal randomized online algorithm. We prove that randomized algorithms are able to obtain better CR compared to deterministic algorithms in the expected sense for this online problem.Publication Open Access Self-supervised monocular scene decomposition and depth estimation(IEEE Computer Society, 2021) Department of Computer Engineering; N/A; Güney, Fatma; Safadoust, Sadra; Department of Computer Engineering; Koç Üniversitesi İş Bankası Yapay Zeka Uygulama ve Araştırma Merkezi (KUIS AI)/ Koç University İş Bank Artificial Intelligence Center (KUIS AI); College of Engineering; Graduate School of Sciences and Engineering; 187939; N/ASelf-supervised monocular depth estimation approaches either ignore independently moving objects in the scene or need a separate segmentation step to identify them. We propose MonoDepthSeg to jointly estimate depth and segment moving objects from monocular video without using any ground-truth labels. We decompose the scene into a fixed number of components where each component corresponds to a region on the image with its own transformation matrix representing its motion. We estimate both the mask and the motion of each component efficiently with a shared encoder. We evaluate our method on three driving datasets and show that our model clearly improves depth estimation while decomposing the scene into separately moving components.Publication Open Access A diversity combination model incorporating an inward bias for interaural time-level difference cue integration in sound lateralization(Multidisciplinary Digital Publishing Institute (MDPI), 2020) N/A; Department of Computer Engineering; Mojtahedi, Sina; Erzin, Engin; Ungan, Pekcan; Faculty Member; Faculty Member; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of Engineering; School of Medicine; N/A; 34503; N/AA sound source with non-zero azimuth leads to interaural time level differences (ITD and ILD). Studies on hearing system imply that these cues are encoded in different parts of the brain, but combined to produce a single lateralization percept as evidenced by experiments indicating trading between them. According to the duplex theory of sound lateralization, ITD and ILD play a more significant role in low-frequency and high-frequency stimulations, respectively. In this study, ITD and ILD, which were extracted from a generic head-related transfer functions, were imposed on a complex sound consisting of two low- and seven high-frequency tones. Two-alternative forced-choice behavioral tests were employed to assess the accuracy in identifying a change in lateralization. Based on a diversity combination model and using the error rate data obtained from the tests, the weights of the ITD and ILD cues in their integration were determined by incorporating a bias observed for inward shifts. The weights of the two cues were found to change with the azimuth of the sound source. While the ILD appears to be the optimal cue for the azimuths near the midline, the ITD and ILD weights turn to be balanced for the azimuths far from the midline.Publication Open Access Neovascularization of engineered tissues for clinical translation: where we are, where we should be?(American Institute of Physics (AIP) Publishing, 2021) Department of Chemical and Biological Engineering; N/A; Nazeer, Muhammad Anwaar; Karaoğlu, İsmail Can; Albayrak, Cem; Kızılel, Seda; Özer, Onur; PhD Student; PhD Student; Faculty Member; Department of Chemical and Biological Engineering; Koç University Research Center for Translational Medicine (KUTTAM) / Koç Üniversitesi Translasyonel Tıp Araştırma Merkezi (KUTTAM); Graduate School of Sciences and Engineering; College of Engineering; N/A; N/A; N/A; 28376; N/AOne of the key challenges in engineering three-dimensional tissue constructs is the development of a mature microvascular network capable of supplying sufficient oxygen and nutrients to the tissue. Recent angiogenic therapeutic strategies have focused on vascularization of the constructed tissue, and its integration in vitro; these strategies typically combine regenerative cells, growth factors (GFs) with custom-designed biomaterials. However, the field needs to progress in the clinical translation of tissue engineering strategies. The article first presents a detailed description of the steps in neovascularization and the roles of extracellular matrix elements such as GFs in angiogenesis. It then delves into decellularization, cell, and GF-based strategies employed thus far for therapeutic angiogenesis, with a particularly detailed examination of different methods by which GFs are delivered in biomaterial scaffolds. Finally, interdisciplinary approaches involving advancement in biomaterials science and current state of technological development in fabrication techniques are critically evaluated, and a list of remaining challenges is presented that need to be solved for successful translation to the clinics.Publication Open Access Acetic acid conversion to ketene on Cu2O(1 0 0): reaction mechanism deduced from experimental observations and theoretical computations(Elsevier, 2021) Tissot, H.; Halldin Stenlid, J.; Wang, C.; Brinck, T.; Sassa, Y.; Johansson, F. O. L.; Weissenrieder, J.; Department of Chemistry; N/A; Kaya, Sarp; Panahi, Mohammad; Faculty Member; PhD Student; Department of Chemistry; Koç University Tüpraş Energy Center (KUTEM) / Koç Üniversitesi Tüpraş Enerji Merkezi (KÜTEM); College of Sciences; Graduate School of Sciences and Engineering; 116541; N/AKetene, a versatile reagent in production of fine and specialty chemicals, is produced from acetic acid. We investigate the synthesis of ketene from acetic acid over the (3,0;1,1) surface of Cu2O(1 0 0) through analysis of the adsorption and desorption characteristics of formic and acetic acids. The results allow us to establish a reaction mechanism for ketene formation. Observations from x-ray photoelectron spectroscopy (XPS), scanning tunneling microscopy, and temperature programmed desorption (TPD), supported by a comparison with formic acid results, suggest that acetic acid reacts with Cu2O through deprotonation to form acetate species coordinated to copper sites and hydroxylation of nearby surface oxygen sites. For formic acid the decomposition of adsorbed formate species results in desorption of CO2 and CO while, for acetic acid, high yields of ketene are observed at temperature >500 K. Modeling by density functional theory (DFT) confirms the strong interaction of acetic acid with the (3,0;1,1) surface and the spontaneous dissociation into adsorbed acetate and hydrogen atom species, the latter forming an OH-group. In an identified reaction intermediate ketene binds via all C and O atoms to Cu surface sites, in agreement with interpretations from XPS. In the vicinity of the adsorbate the surface experiences a local reorganization into a c(2 × 2) reconstruction. The total computed energy barrier for ketene formation is 1.81 eV in good agreement with the 1.74 eV obtained from TPD analysis. Our experimental observations and mechanistic DFT studies suggests that Cu2O can operate as an efficient catalyst for the green generation of ketene from acetic acid.Publication Open Access Strong light-matter interactions in Au plasmonic nanoantennas coupled with Prussian Blue Catalyst on BiVO(4) for photoelectrochemical water splitting(Wiley, 2020) Ghobadi, T. G. U.; Ghobadi, A.; Soydan, M. C.; Karadaş, F.; Özbay, E.; N/A; Department of Chemistry; Barzgarvishlaghi, Mahsa; Kaya, Sarp; Faculty Member; Department of Chemistry; Koç University Tüpraş Energy Center (KUTEM) / Koç Üniversitesi Tüpraş Enerji Merkezi (KÜTEM); Graduate School of Sciences and Engineering; College of Sciences; N/A; 116541A facial and large-scale compatible fabrication route is established, affording a high-performance heterogeneous plasmonic-based photoelectrode for water oxidation that incorporates a CoFe-Prussian blue analog (PBA) structure as the water oxidation catalytic center. For this purpose, an angled deposition of gold (Au) was used to selectively coat the tips of the bismuth vanadate (BiVO4 ) nanostructures, yielding Au-capped BiVO4 (Au-BiVO4 ). The formation of multiple size/dimension Au capping islands provides strong light-matter interactions at nanoscale dimensions. These plasmonic particles not only enhance light absorption in the bulk BiVO4 (through the excitation of Fabry-Perot (FP) modes) but also contribute to photocurrent generation through the injection of sub-band-gap hot electrons. To substantiate the activity of the photoanodes, the interfacial electron dynamics are significantly improved by using a PBA water oxidation catalyst (WOC) resulting in an Au-BiVO4 /PBA assembly. At 1.23 V (vs. RHE), the photocurrent value for a bare BiVO4 photoanode was obtained as 190 μA cm-2 , whereas it was boosted to 295 μA cm-2 and 1800 μA cm-2 for Au-BiVO4 and Au-BiVO4 /PBA, respectively. Our results suggest that this simple and facial synthetic approach paves the way for plasmonic-based solar water splitting, in which a variety of common metals and semiconductors can be employed in conjunction with catalyst designs.Publication Open Access Prototyping products using web-based AI tools: designing a tangible programming environment with children(Association for Computing Machinery (ACM), 2022) N/A; Sezgin, Tevfik Metin; Sabuncuoğlu, Alpay; Faculty Member; Koç Üniversitesi İş Bankası Yapay Zeka Uygulama ve Araştırma Merkezi (KUIS AI)/ Koç University İş Bank Artificial Intelligence Center (KUIS AI); College of Engineering; Graduate School of Sciences and Engineering; 18632; N/AA wide variety of children's products such as mobile apps, toys, and assistant systems now have integrated smart features. Designing such AI-powered products with children, the users, is essential. Using high-fidelity prototypes can be a means to reveal children's needs and behaviors with AI-powered systems. Yet, a prototype that can show unpredictable features similar to the final AI-powered product can be expensive. A more manageable and inexpensive solution is using web-based AI prototyping tools such as Teachable Machine. In this work, we developed a Teachable Machine-powered game-development environment to inform our tangible programming environment's design decisions. Using this kind of an AI-powered high-fidelity prototype in the research process allowed us to observe children in a very similar setting to our final AI-powered product and extract design considerations. This paper reports our experience of prototyping AI-powered solutions with children and shares our design considerations for children's self-made tangible representations.Publication Open Access To augment or not to augment? a comparative study on text augmentation techniques for low-resource NLP(MIT Press, 2022) N/A; Şahin, Gözde Gül; Faculty Member; College of Engineering; 366984Data-hungry deep neural networks have established themselves as the de facto standard for many NLP tasks, including the traditional sequence tagging ones. Despite their state-of-the-art performance on high-resource languages, they still fall behind their statistical counterparts in low-resource scenarios. One methodology to counterattack this problem is text augmentation, that is, generating new synthetic training data points from existing data. Although NLP has recently witnessed several new textual augmentation techniques, the field still lacks a systematic performance analysis on a diverse set of languages and sequence tagging tasks. To fill this gap, we investigate three categories of text augmentation methodologies that perform changes on the syntax (e.g., cropping sub-sentences), token (e.g., random word insertion), and character (e.g., character swapping) levels. We systematically compare the methods on part-of-speech tagging, dependency parsing, and semantic role labeling for a diverse set of language families using various models, including the architectures that rely on pretrained multilingual contextualized language models such as mBERT. Augmentation most significantly improves dependency parsing, followed by part-of-speech tagging and semantic role labeling. We find the experimented techniques to be effective on morphologically rich languages in general rather than analytic languages such as Vietnamese. Our results suggest that the augmentation techniques can further improve over strong baselines based on mBERT, especially for dependency parsing. We identify the character-level methods as the most consistent performers, while synonym replacement and syntactic augmenters provide inconsistent improvements. Finally, we discuss that the results most heavily depend on the task, language pair (e.g., syntactic-level techniques mostly benefit higher-level tasks and morphologically richer languages), and model type (e.g., token-level augmentation provides significant improvements for BPE, while character-level ones give generally higher scores for char and mBERT based models).Publication Open Access A machine learning approach for implementing data-driven production control policies(Taylor _ Francis, 2021) Department of Business Administration; N/A; Tan, Barış; Khayyati, Siamak; Faculty Member; Department of Business Administration; College of Administrative Sciences and Economics; Graduate School of Sciences and Engineering; 28600; N/AGiven the extensive data being collected in manufacturing systems, there is a need for developing a systematic method to implement data-driven production control policies. For an effective implementation, first, the relevant information sources must be selected. Then, a control policy that uses the real-time signals collected from these sources must be implemented. We analyse the production control policy implementation problem in three levels: choosing the information sources, forming clusters of information signals to be used by the policy and determining the optimal policy parameters. Due to the search-space size, a machine-learning-based framework is proposed. Using machine learning speeds up optimisation and allows utilising the collected data with simulation. Through two experiments, we show the effectiveness of this approach. In the first experiment, the problem of selecting the right machines and buffers for controlling the release of materials in a production/inventory system is considered. In the second experiment, the best dispatching policy based on the selected information sources is identified. We show that selecting the right information sources and controlling a production system based on the real-time signals from the selected sources with the right policy improve the system performance significantly. Furthermore, the proposed machine learning framework facilitates this task effectively.Publication Open Access X-Ray chest image classification by a small-sized convolutional neural network(Institute of Electrical and Electronics Engineers (IEEE), 2019) Dokur, Zümray; Ölmez, Tamer; N/A; Kesim, Ege; Graduate School of Sciences and EngineeringConvolutional Neural Networks are widely used in image classification problems due to their high performances. Deep learning methods are also used recently in the classification of medical signals or images. It is observed that well-known pre-trained large networks are used in the classification of X-ray chest images. The performances of these networks on the training set are satisfactory, but their practical use includes some difficulties. The usage of the different imaging modalities in the training process decreases the generalization ability of these networks. And also, due to their large sizes, they are not suitable for real-time applications. In this study, new network structures and the size of the input image are investigated for the classification of X-ray chest images. It is observed that chest images are assigned to twelve classes with approximately 86% success rate by using the proposed network, and the training is carried out in a short time due to the small network structure. The proposed network is run as a real time application on an embedded system with a camera and it is observed that the classification result is produced in less than one second.