Research Outputs
Permanent URI for this communityhttps://hdl.handle.net/20.500.14288/2
Browse
66 results
Search Results
Publication Metadata only A computational-graph partitioning method for training memory-constrained DNNs(Elsevier, 2021) Wahib, Mohamed; Dikbayir, Doga; Belviranli, Mehmet Esat; N/A; Department of Computer Engineering; Qararyah, Fareed Mohammad; Erten, Didem Unat; PhD Student; Faculty Member; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of Engineering; N/A; 219274Many state-of-the-art Deep Neural Networks (DNNs) have substantial memory requirements. Limited device memory becomes a bottleneck when training those models. We propose ParDNN, an automatic, generic, and non-intrusive partitioning strategy for DNNs that are represented as computational graphs. ParDNN decides a placement of DNN's underlying computational graph operations across multiple devices so that the devices' memory constraints are met and the training time is minimized. ParDNN is completely independent of the deep learning aspects of a DNN. It requires no modification neither at the model nor at the systems level implementation of its operation kernels. ParDNN partitions DNNs having billions of parameters and hundreds of thousands of operations in seconds to few minutes. Our experiments with TensorFlow on 16 GPUs demonstrate efficient training of 5 very large models while achieving superlinear scaling for both the batch size and training throughput. ParDNN either outperforms or qualitatively improves upon the related work.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 A gated fusion network for dynamic saliency prediction(Institute of Electrical and Electronics Engineers (IEEE), 2022) Kocak, Aysun; Erdem, Erkut; Department of Computer Engineering; Erdem, Aykut; Faculty Member; Department of Computer Engineering; College of Engineering; 20331Predicting saliency in videos is a challenging problem due to complex modeling of interactions between spatial and temporal information, especially when ever-changing, dynamic nature of videos is considered. Recently, researchers have proposed large-scale data sets and models that take advantage of deep learning as a way to understand what is important for video saliency. These approaches, however, learn to combine spatial and temporal features in a static manner and do not adapt themselves much to the changes in the video content. In this article, we introduce the gated fusion network for dynamic saliency (GFSalNet), the first deep saliency model capable of making predictions in a dynamic way via the gated fusion mechanism. Moreover, our model also exploits spatial and channelwise attention within a multiscale architecture that further allows for highly accurate predictions. We evaluate the proposed approach on a number of data sets, and our experimental analysis demonstrates that it outperforms or is highly competitive with the state of the art. Importantly, we show that it has a good generalization ability, and moreover, exploits temporal information more effectively via its adaptive fusion scheme.Publication Metadata only A new dataset of protein-protein interfaces(Cell Press, 2007) Güney, Emre; Nussinov, Ruth; Tsai, C. J.; Department of Computer Engineering; Department of Chemical and Biological Engineering; Gürsoy, Attila; Keskin, Özlem; Tunçbağ, Nurcan; Faculty Member; Faculty Member; PhD Student; Department of Computer Engineering; Department of Chemical and Biological Engineering; College of Engineering; College of Engineering; 8745; 26605; 245513Publication Metadata only A novel economic-based scheduling heuristic for computational grids(Sage Publications Ltd, 2007) N/A; Department of Computer Engineering; Sönmez, Ömer Ozan; Gürsoy, Attila; Master Student; Faculty Member; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of Engineering; N/A; 8745In the economic-based computational grids we need effective schedulers not only to minimize the makespan but also to minimize the costs that are spent for the execution of the jobs. in this work, A novel economy driven job scheduling heuristic is proposed and a simulation application is developed by using GridSim toolkit to investigate the performance of the heuristic. the simulation-based experiments demonstrate the effectiveness of the proposed heuristic both in terms of parameter sweep and sequential workflow type of applications.Publication Metadata only A volumetric fusion technique for surface reconstruction from silhouettes and range data(academic Press inc Elsevier Science, 2007) Department of Computer Engineering; N/A; N/A; Yemez, Yücel; Wetherilt, Can James; Faculty Member; Master Student; Department of Computer Engineering; College of Engineering; 107907; N/AOptical triangulation, An active reconstruction technique, is known to be an accurate method but has several shortcomings due to occlusion and laser reflectance properties of the object surface, that often lead to holes and inaccuracies on the recovered surface. Shape from silhouette, on the other hand, As a passive reconstruction technique, yields robust, hole-free reconstruction of the visual hull of the object. in this paper, A hybrid surface reconstruction method that fuses geometrical information acquired from silhouette images and optical triangulation is presented. Our motivation is to recover the geometry from silhouettes on those parts of the surface which the range data fail to capture. a volumetric octree representation is first obtained from the silhouette images and then carved by range points to amend the missing cavity information. an isolevel value on each surface cube of the carved octree structure is accumulated using local surface triangulations obtained separately from range data and silhouettes. the MARChing cubes algorithm is then applied for triangulation of the volumetric representation. the performance of the proposed technique is demonstrated on several real objects.Publication Metadata only Analysis of hot region organization in hub proteins(Springer, 2010) N/A; Department of Computer Engineering; Department of Chemical and Biological Engineering; Çukuroğlu, Engin; Gürsoy, Attila; Keskin, Özlem; PhD Student; Faculty Member; Faculty Member; Department of Computer Engineering; Department of Chemical and Biological Engineering; Graduate School of Sciences and Engineering; College of Engineering; College of Engineering; N/A; 8745; 26605Protein interaction maps constructed from binary interactions reveal that some proteins are highly connected to others (acting as hub proteins), whereas some others have a few interactions (at the edges of the map). This paper addresses hub proteins from a structural point: interfaces. It investigates how hot spots are organized in hub proteins (hot regions). We annotate interfaces as the ones between two date-hubs (DD), two party hubs (PP), and two non-hubs (NN). We investigate the physico-chemical properties of these three types of interfaces focusing on the accessible surface area distribution, hot region organization, and amino acid composition differences. Results reveal that there are significant differences between DD and PP interfaces. More of the hot spots are organized into the hot regions in DD interfaces compared to PP ones. A high fraction of the interfaces are covered by hot regions in DD interfaces. There are more distinct hot regions in DDs. Since the same (or overlapping) DD interfaces should be used repeatedly, different hot regions can be used to bind to different partners. Further, these hot region characteristics can be used to predict whether a given hub interface is involved in a DD or a PP interface type with 80% accuracy.Publication Open Access Artificial bandwidth extension of spectral envelope along a Viterbi path(Elsevier, 2013) Department of Computer Engineering; Yağlı, Can; Turan, Mehmet Ali Tuğtekin; Erzin, Engin; Master Student; Faculty Member; Department of Computer Engineering; College of Engineering; N/A; N/A; 34503In this paper, we propose a hidden Markov model (HMM)-based wideband spectral envelope estimation method for the artificial bandwidth extension problem. The proposed HMM-based estimator decodes an optimal Viterbi path based on the temporal contour of the narrowband spectral envelope and then performs the minimum mean square error (MMSE) estimation of the wideband spectral envelope on this path. Experimental evaluations are performed to compare the proposed estimator to the state-of-the-art HMM and Gaussian mixture model based estimators using both objective and subjective evaluations. Objective evaluations are performed with the log-spectral distortion (LSD) and the wideband perceptual evaluation of speech quality (PESQ) metrics. Subjective evaluations are performed with the A/B pair comparison listening test. Both objective and subjective evaluations yield that the proposed wideband spectral envelope estimator consistently improves performances over the state-of-the-art estimators. (C) 2012 Elsevier B.V. All rights reserved.Publication Metadata only Automatic classification,of musical genres using inter-genre similarity(IEEE-Inst Electrical Electronics Engineers Inc, 2007) Bağcı, Ulaş; Department of Computer Engineering; Erzin, Engin; Faculty Member; Department of Computer Engineering; College of Engineering; 34503Musical genre classification is an essential tool for music information retrieval systems and it has potential to become a highly demanded application in various media platforms. Two important problems of the automatic musical genre classification are feature extraction and classifier design. In this letter, we propose two novel classifiers using inter-genre similarity (IGS) modeling and investigate the use of dynamic timbral texture features in order to improve automatic musical genre classification performance. Inter-genre similarity is modeled over hard-to-classify samples of the musical genre feature space. In the classification, samples within inter-genre similarity class are eliminated to reduce inter-genre confusion and to improve genre classification performance. Experimental results show that the proposed classifiers provide better classification rates than the existing methods.Publication Open Access Can a smartband be used for continuous implicit authentication in real life(Institute of Electrical and Electronics Engineers (IEEE), 2020) Ekiz, Deniz; Dardağan Yağmur Ceren; Ersoy, Cem; Department of Computer Engineering; Can, Yekta Said; Department of Computer Engineering; College of EngineeringThe use of cloud services that process privacy-sensitive information such as digital banking, pervasive healthcare, smart home applications requires an implicit continuous authentication solution, which will make these systems less vulnerable to the spoofing attacks. Physiological signals can be used for continuous authentication due to their uniqueness. Ubiquitous wrist-worn wearable devices are equipped with photoplethysmogram sensors, which enable us to extract heart rate variability (HRV) features. In this study, we show that these devices can be used for continuous physiological authentication for enhancing the security of the cloud, edge services, and IoT devices. A system that is suitable for the smartband framework comes with new challenges such as relatively low signal quality and artifacts due to placement, which were not encountered in full lead electrocardiogram systems. After the artifact removal, cleaned physiological signals are fed to the machine learning algorithms. In order to train our machine learning models, we collected physiological data using off-the-shelf smartbands and smartwatches in a real-life event. By applying a minimum quality threshold, we achieved a 3.96% Equal Error Rate. Performance evaluation shows that HRV is a strong candidate for continuous unobtrusive implicit physiological authentication.