Research Outputs
Permanent URI for this communityhttps://hdl.handle.net/20.500.14288/2
Browse
643 results
Search Results
Publication Metadata only 3D articulated shape segmentation using motion information(Institute of Electrical and Electronics Engineers (IEEE), 2010) Department of Computer Engineering; N/A; Department of Computer Engineering; Yemez, Yücel; Kalafatlar, Emre; Faculty Member; Master Student; College of Engineering; Graduate School of Sciences and Engineering; 107907; N/AWe present a method for segmentation of articulated 3D shapes by incorporating the motion information obtained from time-varying models. We assume that the articulated shape is given in the form of a mesh sequence with fixed connectivity so that the inter-frame vertex correspondences, hence the vertex movements, are known a priori. We use different postures of an articulated shape in multiple frames to constitute an affinity matrix which encodes both temporal and spatial similarities between surface points. The shape is then decomposed into segments in spectral domain based on the affinity matrix using a standard K-means clustering algorithm. The performance of the proposed segmentation method is demonstrated on the mesh sequence of a human actor.Publication Metadata only 3D isometric shape correspondence(IEEE, 2010) Department of Computer Engineering; Department of Computer Engineering; Yemez, Yücel; Sahillioğlu, Yusuf; Faculty Member; PhD Student; College of Engineering; Graduate School of Sciences and Engineering; 107907; 215195We address the problem of correspondence between 3D isometric shapes. We present an automatic method that finds the optimal correspondence between two given (nearly) isometric shapes by minimizing the amount of deviation from isometry. We optimize the isometry error in two steps. In the first step, the 3D points uniformly sampled from the shape surfaces are transformed into spectral domain based on geodesic affinity, where the isometry errors are minimized in polynomial time by complete bipartite graph matching. The second step of optimization, which is well-initialized by the resulting correspondence of the first step, explicitly minimizes the isometry cost via an iterative greedy algorithm in the original 3D Euclidean space. Our method is put to test using (nearly) isometric pairs of shapes and its performance is measured via ground-truth correspondence information when available.Publication Metadata only 3D shape correspondence by isometry-driven greedy optimization(IEEE Computer Soc, 2010) N/A; Department of Computer Engineering; Department of Computer Engineering; Sahillioğlu, Yusuf; Yemez, Yücel; PhD Student; Faculty Member; Graduate School of Sciences and Engineering; College of Engineering; 215195; 107907We present an automatic method that establishes 3D correspondence between isometric shapes. Our goal is to find an optimal correspondence between two given (nearly) isometric shapes, that minimizes the amount of deviation from isometry. We cast the problem as a complete surface correspondence problem. Our method first divides the given shapes to be matched into surface patches of equal area and then seeks for a mapping between the patch centers which we refer to as base vertices. Hence the correspondence is established in a fast and robust manner at a relatively coarse level as imposed by the patch radius. We optimize the isometry cost in two steps. in the first step, the base vertices are transformed into spectral domain based on geodesic affinity, where the isometry errors are minimized in polynomial time by complete bipartite graph matching. the resulting correspondence serves as a good initialization for the second step of optimization in which we explicitly minimize the isometry cost via an iterative greedy algorithm in the original 3D Euclidean space. We demonstrate the performance of our method on various isometric (or nearly isometric) pairs of shapes for some of which the ground-truth correspondence is available.Publication Open Access 3D spatial organization and network-guided comparison of mutation profiles in Glioblastoma reveals similarities across patients(Public Library of Science, 2019) Dinçer, Cansu; Kaya, Tuğba; Tunçbağ, Nurcan; Department of Chemical and Biological Engineering; Department of Computer Engineering; Department of Chemical and Biological Engineering; Department of Computer Engineering; Keskin, Özlem; Gürsoy, Attila; Faculty Member; Koç University Research Center for Translational Medicine (KUTTAM) / Koç Üniversitesi Translasyonel Tıp Araştırma Merkezi (KUTTAM); College of Engineering; 26605; 8745Glioblastoma multiforme (GBM) is the most aggressive type of brain tumor. Molecular heterogeneity is a hallmark of GBM tumors that is a barrier in developing treatment strategies. In this study, we used the nonsynonymous mutations of GBM tumors deposited in The Cancer Genome Atlas (TCGA) and applied a systems level approach based on biophysical characteristics of mutations and their organization in patient-specific subnetworks to reduce inter-patient heterogeneity and to gain potential clinically relevant insights. Approximately 10% of the mutations are located in "patches" which are defined as the set of residues spatially in close proximity that are mutated across multiple patients. Grouping mutations as 3D patches reduces the heterogeneity across patients. There are multiple patches that are relatively small in oncogenes, whereas there are a small number of very large patches in tumor suppressors. Additionally, different patches in the same protein are often located at different domains that can mediate different functions. We stratified the patients into five groups based on their potentially affected pathways, revealed from the patient-specific subnetworks. These subnetworks were constructed by integrating mutation profiles of the patients with the interactome data. Network-guided clustering showed significant association between each group and patient survival (P-value = 0.0408). Also, each group carries a set of signature 3D mutation patches that affect predominant pathways. We integrated drug sensitivity data of GBM cell lines with the mutation patches and the patient groups to analyze the therapeutic outcome of these patches. We found that Pazopanib might be effective in Group 3 by targeting CSF1R. Additionally, inhibiting ATM that is a mediator of PTEN phosphorylation may be ineffective in Group 2. We believe that from mutations to networks and eventually to clinical and therapeutic data, this study provides a novel perspective in the network-guided precision medicine.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; Department of Computer Engineering; Qararyah, Fareed Mohammad; Erten, Didem Unat; PhD Student; Faculty Member; 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 Metadata only A containerized proof-of-concept implementation of LightChain system(Ieee, 2020) N/A; N/A; Department of Computer Engineering; N/A; Department of Computer Engineering; Department of Computer Engineering; Department of Computer Engineering; Hassanzadeh-Nazarabadi, Yahya; Nayal, Nazir; Hamdan, Shadi Sameh; Özkasap, Öznur; Küpçü, Alptekin; PhD Student; Faculty Member; Master Student; Faculty Member; Faculty Member; Graduate School of Sciences and Engineering; College of Engineering; Graduate School of Sciences and Engineering; College of Engineering; College of Engineering; N/A; N/A; N/A; 113507; 168060LightChain is the first Distributed Hash Table (DHT)-based blockchain with a logarithmic asymptotic message and memory complexity. In this demo paper, we present the software architecture of our open-source implementation of LightChain, as well as a novel deployment scenario of the entire LightChain system on a single machine aiming at results reproducibility.Publication Metadata only A corpus for computational research of Turkish makam music(Association for Computing Machinery, 2014) Uyar, Burak; Atli, Hasan Sercan; Şentürk, Sertan; Serra, Xavier; Department of Computer Engineering; Department of Computer Engineering; Bozkurt, Barış; Faculty Member; College of Engineering; N/AEach music tradition has its own characteristics in terms of melodic, rhythmic and timbral properties as well as semantic understandings. To analyse, discover and explore these culture-specific characteristics, we need music collections which are representative of the studied aspects of the music tradition. For Turkish makam music, there are various resources available such as audio recordings, music scores, lyrics and Editorial material metadata. However, most of these resources are not typically suited for computational analysis, are hard to access, do not have sufficient quality or do not include adequate descriptive information. In this paper we present a corpus of Turkish makam music created within the scope of the CompMusic project. The corpus is intended for computational research and the primary considerations during the creation of the corpus reflect some criteria, namely, purpose, coverage, completeness, quality and re-usability. So far, we have gathered approximately 6000 audio recordings, 2200 music scores with lyrics and 27000 instances of Editorial material metadata related to Turkish makam music. The metadata include information about makams, recordings, scores, compositions, artists etc. as well as the interrelations between them. In this paper, we also present several test datasets of Turkish makam music. Test datasets contain manual annotations by experts and they provide ground truth for specific computational tasks to test, calibrate and improve the research tools. We hope that this research corpus and the test datasets will facilitate academic studies in several fields such as music information retrieval and computational musicology.Publication Metadata only A critical evaluation of recent deep generative sketch models from a human-centered perspective(Institute of Electrical and Electronics Engineers Inc., 2022) Department of Computer Engineering; N/A; Department of Computer Engineering; Sezgin, Tevfik Metin; Sabuncuoğlu, Alpay; Faculty Member; PhD Student; College of Engineering; Graduate School of Sciences and Engineering; 18632; N/ADrawing a sketch is a uniquely personal process that depends on previous knowledge, experiences, and current mood. Hence, the success of deep generative sketch models depends on user expectations. Yet, the unconditional generation ability of these models does not consider human-centered metrics in the training step. To achieve this kind of training process, we frst need to understand the factors behind human perception on successful generative examples. We designed a user study where we asked twenty-one people from different disciplines to determine these factors. In this study, participants ordered four recent generative models' (Autoencoder, DCGAN, SketchRNN, and Sketchformer) output sketches from most to least recognizable. The results suggest that success in representing the distinct feature of a category is more important than other attributes such as spatial proportions or stroke counts. We shared our code, the interactive notebooks, and feld study results to accelerate further analysis in the area.Publication Metadata only A criticism on popular sketch datasets(Institute of Electrical and Electronics Engineers Inc., 2022) Department of Computer Engineering; N/A; Department of Computer Engineering; Department of Computer Engineering; Sezgin, Tevfik Metin; Dede, Ezgi; Çelik, Birkan; Faculty Member; Master Student; Student; College of Engineering; Graduate School of Sciences and Engineering; College of Engineering; 18632; N/A; N/ASketching is a tool that people can use without any training and benefit from when communicating, thinking or keeping records. The wide range of uses of sketching has made it a high-potential, promising research topic for human-computer interaction researchers. The first step for the researchers who were working for this purpose was developing sketch recognition models. However, in order to continue these studies, they needed a large amount of sketch data. Creating these datasets is a costly task. For this reason, the cheapest methods that enable to produce a large number of sketches quickly were preferred in the research. Although the required amount of sketching data has been collected thanks to these methods, it is necessary to question their quality and similarity to the sketches created during daily life interactions. In this article, a critical comparison of the most widely used sketch datasets in the literature with the sketches we create during daily life interactions is made. In addition, a new dataset which consists of sketches that are created during human-human interactions is introduced. The study showed that popular sketch datasets do not reflect the quality of sketches we create in our daily life.Publication Metadata only A deep learning approach for data driven vocal tract area function estimation(IEEE, 2018) N/A; Department of Computer Engineering; Department of Computer Engineering; Asadiabadi, Sasan; Erzin, Engin; PhD Student; Faculty Member; Graduate School of Sciences and Engineering; College of Engineering; N/A; 34503In this paper we present a data driven vocal tract area function (VTAF) estimation using Deep Neural Networks (DNN). We approach the VTAF estimation problem based on sequence to sequence learning neural networks, where regression over a sliding window is used to learn arbitrary non-linear one-to-many mapping from the input feature sequence to the target articulatory sequence. We propose two schemes for efficient estimation of the VTAF; (1) a direct estimation of the area function values and (2) an indirect estimation via predicting the vocal tract boundaries. We consider acoustic speech and phone sequence as two possible input modalities for the DNN estimators. Experimental evaluations are performed over a large data comprising acoustic and phonetic features with parallel articulatory information from the USC-TIMIT database. Our results show that the proposed direct and indirect schemes perform the VTAF estimation with mean absolute error (MAE) rates lower than 1.65 mm, where the direct estimation scheme is observed to perform better than the indirect scheme.