Arama Sonuçları
Listeleniyor 1 - 10 / 12
Yayın Raylı sistemlerde yüksek gerilim aksamının otomatik denetimi(IEEE, 2014-04-23) Ağdoğan, Didem; Babacan, Veysel Karani; Eskil, Mustafa TanerRaylı sistemlerde yolculugun sorunsuz tamamla-nabilmesi için sistem bütünlüğü kritik öneme sahiptir. Sistem bütünlüğü, lokomotif ve vagonlar haricinde katener (yüksek gerilim) hattı, pantograf ve raylara bağlıdır. Katener hattı ve pantograf, lokomotife elektrik iletimini sağlarken rayların seviyesi pantografın elektrik hattına düzenli temasına etki eder. Raylarda oluşabilecek çöküntüler katener hattı ile pantograf arasında ark (kıvılcım) oluşumuna neden olur. Katener hattının pantograf sınırları dışına çıkması, pantografta oluşabilecek çentikler ve ark oluşumu lokomotif üzerinden anlık izlenebilir. Bu çalışmada amacımız, bu üç ögeden kaynaklanabilecek hataları kameralı bir sistemle, gerçek zamanlı ve otomatik izleyerek tren yolculuğunun güvenli ve kesintisiz yapılmasına katkıda bulunmaktır.Yayın Numerical integration methods for simulation of mass-spring-damper systems(Springer-Verlag, 2012) Özgüz, Mete; Eskil, Mustafa TanerThe dynamics of a face are often implemented as a system of connected particles with various forces acting upon them. Animation of such a system requires the approximation of velocity and position of each particle through numerical integration. There are many numerical integrators that are commonly used in the literature. We conducted experiments to determine the suitability of numerical integration methods in approximating the particular dynamics of mass-spring-damper systems. Among Euler, semi-implicit Euler, Runge-Kutta and Leapfrog, we found that simulation with Leapfrog numerical integration characterizes a mass-spring-damper system best in terms of the energy loss of the overall system.Yayın Palmprint verification using SIFT majority voting(Springer-Verlag, 2012) Abeysundera, Hasith Pasindu; Eskil, Mustafa TanerIn this paper we illustrate the implementation of a robust, real-time biometric system for identity verification based on palmprint images. The palmprint images are preprocessed to align the major axes of hand shapes and to extract the palm region. We extract features using Scale Invariant Feature Transform (SIFT). Classification of individual SIFT features is done through KNN. The class of the hand image is decided by a majority based voting among its classified SIFT features. We demonstrate on the CASIA and PolyU datasets that the proposed system achieves authentication accuracy comparable to other state of the art algorithms.Yayın Integrating vendors into cooperative design practices(Taylor & Francis Ltd, 2009) Eskil, Mustafa Taner; Sticklen, JonThis paper describes a new approach to cooperative design using distributed, off-the-shelf design components. The ultimate goal is to enable assemblers to rapidly design their products and perform simulations using parts that are offered by a global network of suppliers. The obvious way to realise this goal would be to transfer desired component models to the client computer. However, in order to protect proprietary data, manufacturers are reluctant to share their design models without non-disclosure agreements, which can take in the order of months to put in place. Due to bandwidth limitations, it is also impractical to keep the models at the manufacturer site and do simulations by simple message passing. To deal with these impediments in e-commerce the modular distributed modelling (MDM) methodology is leveraged, which enables transfer of component models while hiding proprietary implementation details. MDM methodology with routine design (RD) methods are augmented to realise a platform (RD-MDM) that enables automatic selection of secured off-the-shelf design components over the Internet, integration of these components in an assembly, running simulations for design testing and publishing the approved product model as a secured MDM agent. This paper demonstrates the capabilities of the RD-MDM platform on a fuel cell-battery hybrid vehicle design example.Yayın Extraction and selection of muscle based features for facial expression recognition(IEEE Computer Soc, 2014-12-04) Benli, Kristin Surpuhi; Eskil, Mustafa TanerIn this study we propose a new set of muscle activity based features for facial expression recognition. We extract muscular activities by observing the displacements of facial feature points in an expression video. The facial feature points are initialized on muscular regions of influence in the first frame of the video. These points are tracked through optical flow in sequential frames. Displacements of feature points on the image plane are used to estimate the 3D orientation of a head model and relative displacements of its vertices. We model the human skin as a linear system of equations. The estimated deformation of the wireframe model produces an over-determined system of equations that can be solved under the constraint of the facial anatomy to obtain muscle activation levels. We apply sequential forward feature selection to choose the most descriptive set of muscles for recognition of basic facial expressions.Yayın İfade tanıma için yüz anatomisine dayalı öznitelikler(IEEE, 2014-04-23) Benli, Kristin Surpuhi; Eskil, Mustafa TanerBu çalışmada yüz ifadesi tanıma için kas kuvvetlerine dayalı yeni öznitelikler öneriyoruz. Yüz üzerinde seçtiğimiz noktaların video üzerindeki hareketlerini izleyerek kas kuvvetlerini çözüyoruz. Yüz noktaları, ilk video çerçevesi üzerinde, kas kuvvet alanları üzerinde ilklendirilir. Bu noktalar optik akış algoritması ile izlenir. Noktaların devinimleri yüzün 3 boyutlu yönelimi ve yüz ifadesine dayalı bağıl devinimleri kestirmek için kullanılır. İnsan yüzünü yaylarla, artık-belirtilmiş doğrusal bir denklem sistemi olarak modelliyoruz. Bu sistemi yüz anatomisi kısıtı altında, kas kuvvetleri için çözüyoruz. Ardışık ileri seçim yaparak, temel yüz ifadeleri için en betimleyici kas kümesini belirliyoruz.Yayın Subset selection for tuning of hyper-parameters in artificial neural networks(IEEE, 2017) Aki, K.K.Emre; Erkoç, Tuğba; Eskil, Mustafa TanerHyper-parameters of a machine learning architecture define its design. Tuning of hyper-parameters is costly and for large data sets outright impractical, whether it is performed manually or algorithmically. In this study we propose a Neocognitron based method for reducing the training set to a fraction, while keeping the dynamics and complexity of the domain. Our approach does not require processing of the entire training set, making it feasible for larger data sets. In our experiments we could successfully reduce the MNIST training data set to less than 2.5% (1,489 images) by processing less than 10% of the 60K images. We showed that the reduced data set can be used for tuning of number of hidden neurons in a multi-layer perceptron.Yayın Semi-automatic adaptation of high-polygon wireframe face models through inverse perspective projection(Springer-Verlag, 2012) Benli, Kristin Surpuhi; Ağdoğan, Didem; Özgüz, Mete; Eskil, Mustafa TanerPrecise registration of a generic 3D face model with a subject's face is a critical stage for model based analysis of facial expressions. In this study we propose a semi-automatic model fitting algorithm to fit a high-polygon wireframe model to a single image of a face. We manually mark important landmark points both on the wireframe model and the face image. We carry out an initial alignment by translating and scaling the wireframe model. We then translate the landmark vertices in the 3D wireframe model so that they coincide with inverse perspective projections of image landmark points. The vertices that are not manually labeled as landmark are translated with a weighted sum of vectorial displacement of k neighboring landmark vertices, inversely weighted by their 3D distances to the vertex under consideration. Our experiments indicate that we can fit a high-polygon model to the subject's face with modest computational complexity.Yayın Unreasonable effectiveness of last hidden layer activations for adversarial robustness(Institute of Electrical and Electronics Engineers Inc., 2022) Tuna, Ömer Faruk; Çatak, Ferhat Özgür; Eskil, Mustafa TanerIn standard Deep Neural Network (DNN) based classifiers, the general convention is to omit the activation function in the last (output) layer and directly apply the softmax function on the logits to get the probability scores of each class. In this type of architectures, the loss value of the classifier against any output class is directly proportional to the difference between the final probability score and the label value of the associated class. Standard White-box adversarial evasion attacks, whether targeted or untargeted, mainly try to exploit the gradient of the model loss function to craft adversarial samples and fool the model. In this study, we show both mathematically and experimentally that using some widely known activation functions in the output layer of the model with high temperature values has the effect of zeroing out the gradients for both targeted and untargeted attack cases, preventing attackers from exploiting the model's loss function to craft adversarial samples. We've experimentally verified the efficacy of our approach on MNIST (Digit), CIFAR10 datasets. Detailed experiments confirmed that our approach substantially improves robustness against gradient-based targeted and untargeted attack threats. And, we showed that the increased non-linearity at the output layer has some ad-ditional benefits against some other attack methods like Deepfool attack.Yayın Exploiting epistemic uncertainty of the deep learning models to generate adversarial samples(Cornell Univ, 2021-02-13) Tuna, Ömer Faruk; Çatak, Ferhat Özgür; Eskil, Mustafa TanerDeep neural network architectures are considered to be robust to random perturbations. Nevertheless, it was shown that they could be severely vulnerable to slight but carefully crafted perturbations of the input, termed as adversarial samples. In recent years, numerous studies have been conducted in this new area called "Adversarial Machine Learning" to devise new adversarial attacks and to defend against these attacks with more robust DNN architectures. However, almost all the research work so far has been concentrated on utilising model loss function to craft adversarial examples or create robust models. This study explores the usage of quantified epistemic uncertainty obtained from Monte-Carlo Dropout Sampling for adversarial attack purposes by which we perturb the input to the areas where the model has not seen before. We proposed new attack ideas based on the epistemic uncertainty of the model. Our results show that our proposed hybrid attack approach increases the attack success rates from 82.59% to 85.40%, 82.86% to 89.92% and 88.06% to 90.03% on MNIST Digit, MNIST Fashion and CIFAR-10 datasets, respectively.
