We present a framework for designing fast and monotonic algorithms for transmission tomography penalized- likelihood image reconstruction. The new algorithms are based on paraboloidal surrogate functions for the log likelihood. Due to the... more
We present a framework for designing fast and monotonic algorithms for transmission tomography penalized- likelihood image reconstruction. The new algorithms are based on paraboloidal surrogate functions for the log likelihood. Due to the form of the log-likelihood function it is possible to find low curvature surrogate functions that guarantee monotonicity. Unlike previous methods, the proposed surrogate functions lead to monotonic
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
ABSTRACT Forward-backward pursuit (FBP) is an iterative two stage thresholding method (TST) for sparse signal recovery. Due to the selection of more indices during the forward step than the ones pruned by the backward step, FBP... more
ABSTRACT Forward-backward pursuit (FBP) is an iterative two stage thresholding method (TST) for sparse signal recovery. Due to the selection of more indices during the forward step than the ones pruned by the backward step, FBP iteratively enlarges the support estimate. With this structure, FBP does not necessitate the sparsity level to be known a priori in contrast to other TST algorithms such as subspace pursuit (SP) or compressive sampling matching pursuit. In this work, we address optimal selection of forward and backward step sizes for FBP. We analyse the empirical recovery performance of FBP with different step sizes via phase transitions. Moreover, we compare phase transitions of FBP with those of basis pursuit, SP and orthogonal matching pursuit.
... Burada, e˘ger y = z ise, δ(y, z)=1 'dir, di˘ger durumlarda sıfırdır. Ay ve by ise, A matrisini ve b vektörünü N satırlı matris ve vektörler olarak parçaladı˘gımızda y'ninci alt matris ve vektöre denk gelir: wk = Ak w + bk.... more
... Burada, e˘ger y = z ise, δ(y, z)=1 'dir, di˘ger durumlarda sıfırdır. Ay ve by ise, A matrisini ve b vektörünü N satırlı matris ve vektörler olarak parçaladı˘gımızda y'ninci alt matris ve vektöre denk gelir: wk = Ak w + bk. ... [3] David H. Wolpert, Stacked generalization, Neural Netw., vol. ...
Research Interests:
B-uc aflhc inadequste prrfonnanw of speech recognition systems, an accurate confidence scoring mechanism should be employed to un- dentand the user requests correctly. To determine a confidence score fora hypothesis, cemin confidence... more
B-uc aflhc inadequste prrfonnanw of speech recognition systems, an accurate confidence scoring mechanism should be employed to un- dentand the user requests correctly. To determine a confidence score fora hypothesis, cemin confidence features are combined. In this work the performance offiller-model based confidence features ham bccn in- vertigtted. Five types of filler model networks were defined: mphonc- netwark phone-network, phane-elass
Research Interests:
Straightforward combination of tree search with matching pursuits, which was suggested in 2001 by Cotter and Rao, and then later developed by some other authors, has been revisited recently as multipath matching pursuit (MMP). In this... more
Straightforward combination of tree search with matching pursuits, which was suggested in 2001 by Cotter and Rao, and then later developed by some other authors, has been revisited recently as multipath matching pursuit (MMP). In this comment, we would like to point out some major issues regarding this publication. First, the idea behind MMP is not novel, and the related literature has not been properly referenced. MMP has not been compared to closely related algorithms such as A* orthogonal matching pursuit (A*OMP). The theoretical analyses do ignore the pruning strategies applied by the authors in practice. All these issues have the potential to mislead the reader and lead to misinterpretation of the results. With this short paper, we intend to clarify the relation of MMP to existing literature in the area and compare its performance with A*OMP.
Heuristic search has recently been utilized for compressed sensing signal recovery problem by the A* Orthogonal Matching Pursuit (A*OMP) algorithm. A*OMP employs A* search on a tree with an OMP-based evaluation of the branches, where the... more
Heuristic search has recently been utilized for compressed sensing signal recovery problem by the A* Orthogonal Matching Pursuit (A*OMP) algorithm. A*OMP employs A* search on a tree with an OMP-based evaluation of the branches, where the search is terminated when the desired path length is achieved. The algorithm employs effective pruning techniques and cost models which make the tree search practical. Here, we propose two important extensions of A*OMP: We first introduce a novel dynamic cost model that reduces the search time. Second, we modify the termination criterion by stopping the search when ℓ2 norm of the residue is small enough. Following the restricted isometry property, this termination criterion is more appropriate for our purposes. We demonstrate the improvements in terms of both reconstruction accuracy and computation times via a wide range of simulations.
Research Interests:
Research Interests:
Research Interests:
Linear Discriminant Analysis (LDA) followed by a diagonalizing maximum likelihood linear transform (MLLT) applied to spliced static MFCC features yields important performance gains as compared to MFCC+dynamic features in most speech... more
Linear Discriminant Analysis (LDA) followed by a diagonalizing maximum likelihood linear transform (MLLT) applied to spliced static MFCC features yields important performance gains as compared to MFCC+dynamic features in most speech recognition tasks. It is reasonable to regularize LDA transform computation for stability. In this paper, we regularize LDA and heteroschedastic LDA transforms using two methods: (1) Statistical priors for
Research Interests:
Research Interests:
Determining the assignment of signals received from experiments (peaks) to specific nuclei of the target molecule in Nuclear Magnetic Resonance (NMR) spectroscopy is an important challenge. Nuclear Vector Replacement (NVR) is a framework... more
Determining the assignment of signals received from experiments (peaks) to specific nuclei of the target molecule in Nuclear Magnetic Resonance (NMR) spectroscopy is an important challenge. Nuclear Vector Replacement (NVR) is a framework for structure-based assignments which combines multiple types of NMR data such as chemical shifts, residual dipolar couplings, and NOEs. NVR-BIP is a tool which utilizes a scoring function with a Binary Integer Programming (BIP) model to perform the assignments. In this paper, Support Vector Machines (SVM) and boosting are employed to combine the terms in NVR-BIP’s scoring function by viewing the assignment as a classification problem. The assignment accuracies obtained using this approach show that boosting improves the assignment accuracy of NVR-BIP on our data set when RDCs are not available and outperforms SVMs. With RDCs, boosting and SVMs offer mixed results.
Research Interests:
... Ancak, Stauffer-Grimson'in tek a*amali e,iklemesi ve buna ek olarak yapilan ikincil e,ikleme ile elde edilen onplan aday goruntuleri ile birlikte incelendiginde, onplan kenar bilgisi, onplan nesnesinin tespitini... more
... Ancak, Stauffer-Grimson'in tek a*amali e,iklemesi ve buna ek olarak yapilan ikincil e,ikleme ile elde edilen onplan aday goruntuleri ile birlikte incelendiginde, onplan kenar bilgisi, onplan nesnesinin tespitini iyile,tirmektedir. 5. Deneyler ve Tartilma ...
Research Interests:
Research Interests:
ABSTRACT A* Orthogonal Matching Pursuit (A*OMP) utilizes best-first search for recovery of sparse signals from reduced dimensions. It applies A* search on a tree whose branches are extended similar to Orthogonal Matching Pursuit (OMP).... more
ABSTRACT A* Orthogonal Matching Pursuit (A*OMP) utilizes best-first search for recovery of sparse signals from reduced dimensions. It applies A* search on a tree whose branches are extended similar to Orthogonal Matching Pursuit (OMP). A*OMP makes a tractable tree search possible via effective complexity-accuracy trade-off parameters. Here, we concentrate on the effects of these parameters on the recovery performance. Via empirical comparison of complexity and performance, we demonstrate the effects of the search parameters on search size and recovery. We also compare A*OMP with well-known compressed sensing (CS) recovery techniques to reveal the improvement in the reconstruction.
Research Interests:
Research Interests:
Research Interests:
ABSTRACT In recent years, the interest in human authentication has been increasing. Biometrics are one of the easy authentication schemes, however, security and privacy problems limit their widespread usage. Following the interest in... more
ABSTRACT In recent years, the interest in human authentication has been increasing. Biometrics are one of the easy authentication schemes, however, security and privacy problems limit their widespread usage. Following the interest in privacy protecting biometric authentication, template protection schemes for biometric modalities has increased significantly in order to cope with security and privacy issues. BioHashing, which is based on transforming the biometric template using pseudo-random projections that are generated using a user-specified key or token, has received much attention as it improves verification accuracies over using only the biometric data, allows template revocation and preserves privacy. In our work, we develop a new BioHashing scheme for fingerprints. A fixed-length feature vector is required in order to design a BioHashing scheme. In the literature, most of the studies on fingerprint BioHashing uses features extracted from fingerprint texture. On the other hand, our new BioHashing scheme is based on minutia based feature vectors. We use the spectral minutiae representation for obtaining a fixed-length feature vector for a fingerprint sample. Then, we use a random projection matrix, which is generated from user's key/token, in order to generate a BioHash vector. We propose to randomly project each column of the spectral minutiae feature matrix via a single matrix which allows fast bit string extraction and adaptive quantization. Experiments on FVC2002 databases show the promise of the proposed system for fast and secure verification.
ABSTRACT In this paper, we propose a novel face image hashing method based on an optimal linear transformation. In the proposed method, first, we apply a feature extraction method. Then, we define an optimal linear transformation matrix... more
ABSTRACT In this paper, we propose a novel face image hashing method based on an optimal linear transformation. In the proposed method, first, we apply a feature extraction method. Then, we define an optimal linear transformation matrix based on within-class covariance matrix which is the maximum likelihood estimate of the variations of the biometric data belonging to the same user. Next, we reduce the dimension of the feature vector by using this transform. Finally, we apply quantization and obtain a face image hash vector. We test the performance of the proposed method with AT&T and M2VTS face databases and compare the results with the random projection based biometric hashing methods. We perform the simulations by taking into account two scenarios: 1) Secret key is not known by attacker, 2) Attacker illegally acquires the secret key. The simulation results show the proposed method has better performance especially when the secret key has been compromised.
Research Interests:
The binary quantization method which is used in random projection based biometric hashing systems reduces the authentication performance of these systems. In this paper, we propose new statistical quantization methods for biometric... more
The binary quantization method which is used in random projection based biometric hashing systems reduces the authentication performance of these systems. In this paper, we propose new statistical quantization methods for biometric hashing systems. Our proposed quantization methods use Gaussian mixture model and Gaussian distributions to determine quantization threshold value. Therefore, we improve the authentication performance of the biometric hashing
Research Interests:
In this paper I we propose a turn-based language modeling (TurnLM) technique for spoken dialog systems. This technique utilizes the time dependent nature of a dialog aimed at accomplishing a task. As opposed to the dialog state based... more
In this paper I we propose a turn-based language modeling (TurnLM) technique for spoken dialog systems. This technique utilizes the time dependent nature of a dialog aimed at accomplishing a task. As opposed to the dialog state based language modeling techniques which depend on the information in the system prompt, TurnLM does not require any information from the dialog manager.
Research Interests:
... Literatürde, bir kısım araştırmacılar da biyometrik tabanlı bu uygulamaların güvenlik ve mahremiyet problemleri üzerine çalışmalar yapmaktadır. Biyometrik verilerinin güvenlik ve mahremiyet problemleri bu tip sistemlerin yaygın... more
... Literatürde, bir kısım araştırmacılar da biyometrik tabanlı bu uygulamaların güvenlik ve mahremiyet problemleri üzerine çalışmalar yapmaktadır. Biyometrik verilerinin güvenlik ve mahremiyet problemleri bu tip sistemlerin yaygın kullanımı önündeki en büyük engellerden biridir. ...
Research Interests:
We present a framework for designing fast and monotonic algorithms for transmission tomography penalized- likelihood image reconstruction. The new algorithms are based on paraboloidal surrogate functions for the log likelihood. Due to the... more
We present a framework for designing fast and monotonic algorithms for transmission tomography penalized- likelihood image reconstruction. The new algorithms are based on paraboloidal surrogate functions for the log likelihood. Due to the form of the log-likelihood function it is possible to find low curvature surrogate functions that guarantee monotonicity. Unlike previous methods, the proposed surrogate functions lead to monotonic
Research Interests:
Page 1. KONUS¸ MA TANIMA ˙ IC¸˙IN NOMA ˙ ILE TEK-KANALDA KONUS¸ MA-M ¨UZ˙IK AYRIS¸ TIRMA SINGLE-CHANNEL SPEECH-MUSIC SEPARATION USING NMF FOR AUTOMATIC SPEECH RECOGNITION Cemil Demir1 ...
Research Interests:
NIST Speaker Recognition Evaluations (NIST SRE), played a major role in the progress of text-independent speaker verifica- tion algorithms. Tubitak UEKAE - Sabanci University jointly participated in NIST SRE 2010. In this paper, the... more
NIST Speaker Recognition Evaluations (NIST SRE), played a major role in the progress of text-independent speaker verifica- tion algorithms. Tubitak UEKAE - Sabanci University jointly participated in NIST SRE 2010. In this paper, the Gaussian Mixture Model Supervector - Support Vector Machine based speaker verification system developed for NIST SRE 2010 has been presented. A recipe has been given for