Dept. of Computer Engineering - Master's degree
Permanent URI for this collection
Browse
Browsing Dept. of Computer Engineering - Master's degree by Title
Now showing 1 - 20 of 582
Results Per Page
Sort Options
Item Open Access A 3D garment design and simulation system(Bilkent University, 2004) Durupınar, FundaIn this thesis study, a 3D graphics environment for virtual garment design and simulation is presented. The proposed system enables the three dimensional construction of a garment from its two dimensional cloth panels, for which the underlying structure is a mass-spring model. Construction of the garment is performed through cutting, boundary smoothing , seaming and scaling. Afterwards, it is possible to do fitting on virtual mannequins like in the real life as if in a tailor’s workshop. The behavior of cloth under different environmental conditions is implemented applying a physically-based approach. As well as the simulation of the draping of garments, efficient and realistic visualization of garments is an important issue in cloth modelling. There are various material types and reflectance properties for fabrics. We have implemented a number of material and rendering options such as knitwear, woven cloth and standard shading methods such as Gouraud shading. Performance results of the system are presented at the end.Item Open Access 3D hair design and key frame animation in real time(Bilkent University, 2008) Başarankut, BarkınComputer generated animations of humans, animals and all other kinds of objects have been studied extensively during the last two decades. The key for creating good animations has been to correctly imitate the behaviors of real objects and reflect these into computer generated images. With the rapid development of computer technology, creating realistic simulations has become possible, and the most striking components of these realistic animations happen to be the most dynamic (moving) parts; hair, in the case of human animations. With the development of high quality hair animations, the concern is not only creating physically correct animations, but also controlling these animations. An implementation of a key frame hair animation creation system, supported by a hair design tool, helping to model and animate hair easily, and provide these functionalities in real time is the aim of the proposed system. This work reviews several hair animation and sketching techniques, and proposes a system that provides a complete level of control (capable of controlling even the individual hair strands) of key frame animation and hair design in real time.Item Open Access 3D hair sketching for real-time hair modeling and dynamic simulations(Bilkent University, 2008) Aras, RıfatHair has been an active research area in computer graphics society for over a decade. Different approaches have been proposed for different aspects of hair research such as modeling, simulating, animating and rendering. In this thesis, we introduce a sketch-based tool making use of direct manipulation interfaces to create hair models and furthermore simulate the created hair models under physically based constraints in real-time. Throughout the thesis, the created tool will be analyzed with respect to different aspects of the problem such as hair modeling, hair simulation, hair sketching and hair rendering.Item Open Access 3D mesh animation system targeted for multi-touch environments(Bilkent University, 2009) Ceylan, DuyguFast developments in computer technology have given rise to different application areas such as multimedia, computer games, and Virtual Reality. All these application areas are based on animation of 3D models of real world objects. For this purpose, many tools have been developed to enable computer modeling and animation. Yet, most of these tools require a certain amount of experience about geometric modeling and animation principles, which creates a handicap for inexperienced users. This thesis introduces a solution to this problem by presenting a mesh animation system targeted specially for novice users. The main approach is based on one of the fundamental model representation concepts, Laplacian framework, which is successfully used in model editing applications. The solution presented perceives a model as a combination of smaller salient parts and uses the Laplacian framework to allow these parts to be manipulated simultaneously to produce a sense of movement. The interaction techniques developed enable users to carry manipulation and global transformation actions at the same time to create more pleasing results. Furthermore, the approach utilizes the multi-touch screen technology and direct manipulation principles to increase the usability of the system. The methods described are experimented by creating simple animations with several 3D models; which demonstrates the advantages of the proposed solution.Item Open Access Accelerating pagerank with a heterogeneous two phase CPU-FPGA algorithm(Bilkent University, 2020-12) Usta, FurkanPageRank is a network analysis algorithm that is used to measure the importance of each vertex in a graph. Fundamentally it is a Sparse Matrix-Vector multiplication problem and suffers from the same bottlenecks, such as irregular memory access and low computation-to-communication ratio. Moreover, the existing Field Programmable Gate Array (FPGA) accelerators for PageRank algorithm either require large portions of the graph to be in-memory, which is not suitable for big data applications or cannot fully utilize the memory bandwidth. Recently published Propagation Blocking(PB) methodology improves the performance of PageRank by dividing the execution into binning and accumulation phases. In this paper, we propose a heterogeneous high-throughput implementation of the PB algorithm where the binning phase executed on the FPGA while accumulation is done on a CPU. Unlike prior solutions, our design can handle graphs of any sizes with no need for an on-board FPGA memory. We also show that despite the low frequency of our device, compared to the CPU, by offloading random writes to an accelerator we can still improve the performance significantly. Experimental results show that with our proposed accelerator, PB algorithm can gain up to 40% speedup.Item Open Access Accelerator design for graph analytics(Bilkent University, 2016-06) Yeşil, ŞerifWith the increase in data available online, data analysis became a significant problem in today’s datacenters. Moreover, graph analytics is one of the significant application domains in big data era. However, traditional architectures such as CPUs and Graphics Processing Units (GPUs) fail to serve the needs of graph applications. Unconventional properties of graph applications such as irregular memory accesses, load balancing, and irregular computation challenge current computing systems which are either throughput oriented or built on top of traditional locality based memory subsystems. On the other hand, an emerging technique hardware customization, can help us to overcome these problems since they are expected to be energy efficient. Considering the power wall, hardware customization becomes more desirable. In this dissertation, we propose a hardware accelerator framework that is capable of handling irregular, vertex centric, and asynchronous graph applications. Developed high level SystemC models gives an abstraction to the programmer allowing to implement the hardware without extensive knowledge about the underlying architecture. With the given template, programmers are not limited to a single application since they can develop any graph application as long as it fits to the given template abstract. Besides the ability to develop different applications, the given template also decreases the time spent on developing and testing different accelerators. Additionally, an extensive experimental study shows that the proposed template can outperform a high-end 24 core CPU system up to 3x with up to 65x power efficiency.Item Open Access Active learning methods based on statistical leverage scores(Bilkent University, 2016-08) Orhan, CemIn many real-world machine learning applications, unlabeled data are abundant whereas the class labels are expensive and/or scarce. An active learner aims to obtain a model with high accuracy with as few labeled instances as possible by effectively selecting useful examples for labeling. We propose two novel active learning approaches for pool-based active learning setting: ALEVS for querying single example at each iteration and DBALEVS for querying a batch of examples. ALEVS and DBALEVS select the most in uential instance(s) based on statistical leverages scores of examples. The rank-k statistical leverage score of i-th row of an n x n kernel matrix K is the squared norm of the i-th row of the matrix U whose columns are the top-k eigenvectors of K. Statistical leverage scores are shown to be useful in matrix approximation algorithms in finding in uential rows of a matrix. ALEVS and DBALEVS assess the in uence of the examples by the statistical leverage scores of kernel matrix computed on the examples of the pool. Additionally, through maximizing a submodular set function at each iteration DBALEVS selects a diverse a set of examples that are highly in uential but are dissimilar to selected labeled set. Extensive experiments on diverse datasets show that the proposed methods, ALEVS and DBALEVS offer more effective strategies in comparison to other single and batch mode active learning approaches, respectively.Item Open Access Active node determination for correlated data gathering in wireless sensor networks(Bilkent University, 2009) Karasabun, EfeIn wireless sensor network applications where data gathered by different sensor nodes is correlated, not all sensor nodes need to be active for the wireless sensor network to be functional. However, the sensor nodes that are selected as active should form a connected wireless network in order to transmit the collected correlated data to the data gathering node. The problem of determining a set of active sensor nodes in a correlated data environment for a fully operational wireless sensor network can be formulated as an instance of the connected correlation-dominating set problem. In this work, our contribution is twofold; we propose an effective and runtime efficient iterative improvement heuristic to solve the active sensor node determination problem and a benefit function that aims to minimize the number of active sensor nodes while maximizing the residual energy levels of the selected active sensor nodes. Extensive simulations we performed show that the proposed approach can achieve a good performance in terms of both network lifetime and runtime efficiency.Item Open Access Active set partitioning scheme for extending the lifetime of large wireless sensor networks(Bilkent University, 2010) Kalkan, MustafaWireless Sensor Networks consist of spatially distributed and energy-constrained autonomous devices called sensors to cooperatively monitor physical or environmental conditions such as temperature, sound, vibration, pressure or pollutants at different locations. Because these sensor nodes have limited energy supply, energy efficiency is a critical design issue in wireless sensor networks. Having all the nodes simultaneously work in the active mode, results in an excessive energy consumption and packet collisions because of high node density in the network. In order to minimize energy consumption and extend network life-time, this thesis presents a centralized graph partitioning approach to organize the sensor nodes into a number of active sensor node sets such that each active set maintains the desired level of sensing coverage and forms a connected network to perform sensing and communication tasks successfully. We evaluate our proposed scheme via simulations under different network topologies and parameters in terms of network lifetime and run-time efficiency and observe approximately 50% improvement in the number of obtained active node sets when compared with different active node set selection mechanisms.Item Open Access Activity analysis for assistive systems(Bilkent University, 2014) İşcen, AhmetAlthough understanding and analyzing human actions is a popular research topic in computer vision, most of the research has focused on recognizing ”ordinary” actions, such as walking and jumping. Extending these methods for more specific domains, such as assistive technologies, is not a trivial task. In most cases, these applications contain more fine-grained activities with low inter-class variance and high intra-class variance. In this thesis, we propose to use motion information from snippets, or small video intervals, in order to recognize actions from daily activities. Proposed method encodes the motion by considering the motion statistics, such as the variance and the length of trajectories. It also encodes the position information by using a spatial grid. We show that such approach is especially helpful for the domain of medical device usage, which contains actions with fast movements Another contribution that we propose is to model the sequential information of actions by the order in which they occur. This is especially useful for fine-grained activities, such as cooking activities, where the visual information may not be enough to distinguish between different actions. As for the visual perspective of the problem, we propose to combine multiple visual descriptors by weighing their confidence values. Our experiments show that, temporal sequence model and the fusion of multiple descriptors significantly improve the performance when used together.Item Open Access An actuated flexible spinal mechanism for a bounding quadrupedal robot(Bilkent University, 2012) Çulha, UtkuEvolution and experience based learning have given animals body structures and motion capabilities to survive in the nature by achieving many complicated tasks. Among these animals, legged vertebrates use their musculoskeletal bodies up to the limits to achieve actions involving high speeds and agile maneuvers. Moreover the flexible spine plays a very important role in providing auxiliary power and dexterity for such dynamic behaviors. Robotics research tries to imitate such dynamic abilities on mechanical platforms. However, most existing robots performing these dynamic motions does not include such a flexible spine architecture. In this thesis we investigate how quadrupedal bounding can be achieved with the help of an actuated flexible spine. Depending upon biological correspondences we first present a novel quadruped robot model with an actuated spine and relate it with our proposed new bounding gait controller model. By optimizing our model and a standard stiff backed model via repetitive parametric methods, we investigate the role of spinal actuation on the performance enhancements of the flexible model. By achieving higher ground speeds and hopping heights we discuss the relations between flexible body structure and stride properties of a dynamic bounding gait. Furthermore, we present an analytical model of the proposed robot structure along with the spinal architecture and analyze the dynamics and active forces on the overall system. By gathering simulation results we question how such a flexible spine system can be improved to achieve higher performances during dynamic gaits.Item Open Access Adaptation of multiway-merge sorting algorithm to MIMD architectures with an experimental study(Bilkent University, 2002) Cantürk, LeventSorting is perhaps one of the most widely studied problems of computing. Numerous asymptotically optimal sequential algorithms have been discovered. Asymptotically optimal algorithms have been presented for varying parallel models as well. Parallel sorting algorithms have already been proposed for a variety of multiple instruction, multiple data streams (MIMD) architectures. In this thesis, we adapt the multiwaymerge sorting algorithm that is originally designed for product networks, to MIMD architectures. It has good load balancing properties, modest communication needs and well performance. The multiway-merge sort algorithm requires only two all-to-all personalized communication (AAPC) and two one-to-one communications independent from the input size. In addition to evenly distributed load balancing, the algorithm requires only size of 2N/P local memory for each processor in the worst case, where N is the number of items to be sorted and P is the number of processors. We have implemented the algorithm on the PC Cluster that is established at Computer Engineering Department of Bilkent University. To compare the results we have implemented a sample sort algorithm (PSRS Parallel Sorting by Regular Sampling) by X. Liu et all and a parallel quicksort algorithm (HyperQuickSort) on the same cluster. In the experimental studies we have used three different benchmarks namely Uniformly, Gaussian, and Zero distributed inputs. Although the multiwaymerge algorithm did not achieve better results than the other two, which are theoretically cost optimal algorithms, there are some cases that the multiway-merge algorithm outperforms the other two like in Zero distributed input. The results of the experiments are reported in detail. The multiway-merge sort algorithm is not necessarily the best parallel sorting algorithm, but it is expected to achieve acceptable performance on a wide spectrum of MIMD architectures.Item Open Access Adaptive source routing and route generation for multicomputers(Bilkent University, 1995) Aydoğan, YücelScalable multicomputers are based upon interconnection networks that typically provide multiple communication routes between any given pair of processor nodes. In such networks, the selection of the routes is an important problem because of its impact on the communication performance. We propose the adaptive source routing (ASR) scheme which combines adaptive routing and source routing into one which has the advantages of both schemes. In ASR, the degree of adaptivity of each packet is determined at the source processor. Every packet can be routed in a fully adaptive or partially adaptive or nonadaptive manner, all within the same network at the same time. The ASR scheme permits any network topology to be used provided that deadlock constraints are satisfied. We evaluate and compare performance of the adaptive source routing and non-adaptive randomized routing by simulations. Also we propose an algorithm to generate adaptive routes for all pairs of processors in any multistage interconnection network. Adaptive routes are stored in a route table in each processor’s memory and provide high bandwidth and reliable interprocessor communication. We evaluate the performance of the algorithm on IBM SP2 networks in terms of obtained bandwidth, time to fill in the route tables, and efficiency exploited by the parallel execution of the algorithm.Item Open Access Adaptive thread and memory access schelduling in chip multiprocessors(Bilkent University, 2013) Aktürk, İsmailThe full potential of chip multiprocessors remains unexploited due to architecture oblivious thread schedulers used in operating systems, and thread-oblivious memory access schedulers used in off-chip main memory controllers. For the thread scheduling, we introduce an adaptive cache-hierarchy-aware scheduler that tries to schedule threads in a way that inter-thread contention is minimized. A novel multi-metric scoring scheme is used that specifies the L1 cache access characteristics of a thread. The scheduling decisions are made based on multi-metric scores of threads. For the memory access scheduling, we introduce an adaptive compute-phase prediction and thread prioritization scheme that efficiently categorize threads based on execution characteristics and provides fine-grained prioritization that allows to differentiate threads and prioritize their memory access requests accordingly.Item Open Access Affect and personality aware analysis of speech content for automatic estimation of depression severity(Bilkent University, 2023-09) Gönç, KaanThe detection of depression has gained a significant amount of scientific attention for its potential in early diagnosis and intervention. In light of this, we propose a novel approach that places exclusive emphasis on textual features for depression severity estimation. The proposed method seamlessly integrates affect (emotion and sentiment), and personality features as distinct yet interconnected modalities within a transformer-based architecture. Our key contribution lies in a masked multimodal joint cross-attention fusion, which adeptly combines the information gleaned from these different text modalities. This fusion approach empowers the model not only to discern subtle contextual cues within textual data but also to comprehend intricate interdependencies between the modalities. A comprehensive experimental evaluation is undertaken to meticulously assess the individual components comprising the proposed architecture, as well as extraneous ones that are not inherent to it. The evaluation additionally includes the assessments conducted in a unimodal setting where the impact of each modality is examined individually. The findings derived from these experiments substantiate the self-contained efficacy of our architecture. Furthermore, we explore the significance of individual sentences within speech content, offering valuable insights into the contribution of specific textual cues and we perform a segmented evaluation of the proposed method for different ranges of depression severity. Finally, we compare our method with existing state-of-the-art studies utilizing different combinations of auditory, visual, and textual features. The final results demonstrate that our method achieves promising results in depression severity estimation, outperforming the other methods.Item Open Access Age and gender normalization in kinship verification(Bilkent University, 2021-09) Çalıkkasap, OğuzhanKinship veri cation from facial images using deep learning is an interesting problem that is unsolved and gains growing attention of the research community. However, the most recent kinship veri cation systems su er from age- and genderrelated facial attributes that cause problems in kinship veri cation between subjects of di erent age and gender. In this study, we propose various methods to reduce the negative e ect of the age- and gender-related facial attributes in kinship veri cation to achieve a more robust veri cation model. The proposed approach utilizes the comprehensive modeling capabilities of the recent generative adversarial network architectures to model the age and gender of subjects and reduce their e ect in kinship veri cation, if not remove entirely. Furthermore, we conduct a thorough analysis over individual and combined e ects of age and gender normalization, performed in both image and latent space of the generative models. Lastly, we investigate the impact of additional emphasis on the facial identity information during the normalization process. Taking one of the most recent kinship veri cation models as our baseline, we show that gender normalization has reduced the veri cation performance gap between subject pairs with the same and di erent gender, up to 6%. Furthermore, joint normalization of age and gender improves the kinship veri cation accuracy up to 5% and 10% on two di erent in-the-wild kinship datasets. Therefore, this thesis proposes generic approaches to improve the reliability and robustness of kinship veri cation by normalizing the age and gender attributes without making changes in the core architecture of the employed kinship veri cation system.Item Open Access Algorithms for effective querying of graph-based pathway databases(Bilkent University, 2007) Çetintaş, AhmetAs the scientific curiosity shifts toward system-level investigation of genomicscale information, data produced about cellular processes at molecular level has been accumulating with an accelerating rate. Graph-based pathway ontologies and databases have been in wide use for such data. This representation has made it possible to programmatically integrate cellular networks as well as investigating them using the well-understood concepts of graph theory to predict their structural and dynamic properties. In this regard, it is essential to effectively query such integrated large networks to extract the sub-networks of interest with the help of efficient algorithms and software tools. Towards this goal, we have developed a querying framework along with a number of graph-theoretic algorithms from simple neighborhood queries to shortest paths to feedback loops, applicable to all sorts of graph-based pathway databases from PPIs to metabolic pathways to signaling pathways. These algorithms can also account for compound or nested structures present in the pathway data, and have been implemented within the querying components of Patika (Pathway Analysis Tools for Integration and Knowledge Acquisition) tools and have proven to be useful for answering a number of biologically significant queries for a large graph-based pathway database.Item Open Access Algorithms for sink mobility in wireless sensor networks to improve network lifetime(Bilkent University, 2008) Koç, MetinA wireless sensor network (WSN) consists of hundreds or thousands of sensor nodes organized in an ad-hoc manner to achieve a predefined goal. Although WSNs have limitations in terms of memory and processor, the main constraint that makes WSNs different from traditional networks is the battery problem. Since sensor nodes are generally deployed to areas with harsh environmental conditions, replacing the exhausted batteries become practically impossible. This requires to use the energy very carefully in both node and network level. Different approaches are proposed in the literature for improving network lifetime, including data aggregation, energy efficient routing schemes and MAC protocols, etc. Main motivation for these approaches is to prolong the network lifetime without sacrificing service quality. Sink (data collection node) mobility is also one of the effective solutions in the literature for network lifetime improvement. In this thesis, we focus on the controlled sink mobility and present a set of algorithms for different parts of the problem, like sink sites determination, and movement decision parameters. Moreover, a load balanced topology construction algorithm is given as another component of network lifetime improvement. Experiment results are presented which compare the performance of different components of the mobility scheme with other approaches in the literature, and the whole sink mobility scheme with random movement and static sink cases. As a result, it is observed that our algorithms perform better than random movement and static cases for different scenarios.Item Open Access Algorithms for the discovery of large genomic inversions using pooled clone sequencing(Bilkent University, 2015) Rasekh, Marzieh EslamiAn inversion is a chromosomal rearrangement in which an internal segment of a chromosome has been broken twice, flipped 180 degrees, and rejoined. Most known examples of large inversions were found indirectly from studies on human disease where inversions have no detectable effect in parents, but increase the risk of a disease-associated rearrangement in the offspring. The development of a map of inversion polymorphisms will provide valuable information regarding their distribution and frequency in the human genome and will help unravel how inversions and the segmental duplications architecture associated with inverted haplotypes contribute to genomic susceptibility to disease rearrangements. The 1000 Genomes Project spearheaded the development of several methods to identify inversions, however, they are limited to relatively short inversions, and there are currently no available algorithms to discover large inversions using high throughput sequencing technologies (HTS). This is mainly because the breakpoints of such events typically lie within segmental duplications and common repeats, reducing the mappability of short reads. We propose using pooled clone sequencing (PCS), a method originally developed to improve haplotype phasing, to characterize large genomic inversions. PCS merges the advantages of clone based sequencing approaches with the speed and cost efficiency of HTS technologies. Using this sequencing data, we developed a novel algorithm, dipSeq for discovering large inversions (>500 Kbp) following the observation that clones that span the inversion breakpoint will be split into two sections, split clones, when mapped to the reference genome. We evaluate the performance of dipSeq on 3 sets of simulated data, demonstrating its correctness and robustness to structural duplications and other types of structural variations. We further applied dipSeq to the genome of a HapMap individual (NA12878). dipSeq was able to accurately discover all previously known and experimentally validated large inversions. We also identified a new inversion and confirmed using fluorescent in situ hybridization. Although dipSeq displays a relatively high false positive rate using real data, it performed better with simulated data, suggesting that the performance with the NA12878 genome may be improved with higher depth of coverage.Item Open Access An evaluation of transaction management issues in mobile real-time database systems(Bilkent University, 1998-01) Kayan, ErsanThe integration of issues from real-time database systems and mobile computing systems is a new research area that aims to provide support for real-time requirements of database applications in a mobile computing environment. Due to certain constraints of a mobile computing environment, such as resource-poor mobile computers, low-bandwidth and unreliable wireless links, and mobility of users/computers, various research issues in traditional real-time database systems need to be reconsidered. In this thesis, we investigate some of these issues using a detailed simulation model of a mobile real-time database management system. We propose a transaction execution model with two alternative execution strategies and evaluate the performance of the system considering various mobile system characteristics. Performance results are provided in terms of the fraction of transaction that satisfy their deadlines.