Browsing by Subject "Neural networks"

Now showing 1 - 20 of 67

Open Access
A broad ensemble learning system for drifting stream classification
(Institute of Electrical and Electronics Engineers, 2023-08-21) Bakhshi, Sepehr; Ghahramanian, Pouya; Bonab, H.; Can, Fazlı
In a data stream environment, classification models must effectively and efficiently handle concept drift. Ensemble methods are widely used for this purpose; however, the ones available in the literature either use a large data chunk to update the model or learn the data one by one. In the former, the model may miss the changes in the data distribution, while in the latter, the model may suffer from inefficiency and instability. To address these issues, we introduce a novel ensemble approach based on the Broad Learning System (BLS), where mini chunks are used at each update. BLS is an effective lightweight neural architecture recently developed for incremental learning. Although it is fast, it requires huge data chunks for effective updates and is unable to handle dynamic changes observed in data streams. Our proposed approach, named Broad Ensemble Learning System (BELS), uses a novel updating method that significantly improves best-in class model accuracy. It employs an ensemble of output layers to address the limitations of BLS and handle drifts. Our model tracks the changes in the accuracy of the ensemble components and reacts to these changes. We present our mathematical derivation of BELS, perform comprehensive experiments with 35 datasets that demonstrate the adaptability of our model to various drift types, and provide its hyperparameter, ablation, and imbalanced dataset performance analysis. The experimental results show that the proposed approach outperforms 10 state-of-the-art baselines, and supplies an overall improvement of 18.59% in terms of average prequential accuracy.
Open Access
Activity recognition invariant to sensor orientation with wearable motion sensors
(MDPI AG, 2017) Yurtman, A.; Barshan, B.
Most activity recognition studies that employ wearable sensors assume that the sensors are attached at pre-determined positions and orientations that do not change over time. Since this is not the case in practice, it is of interest to develop wearable systems that operate invariantly to sensor position and orientation. We focus on invariance to sensor orientation and develop two alternative transformations to remove the effect of absolute sensor orientation from the raw sensor data. We test the proposed methodology in activity recognition with four state-of-the-art classifiers using five publicly available datasets containing various types of human activities acquired by different sensor configurations. While the ordinary activity recognition system cannot handle incorrectly oriented sensors, the proposed transformations allow the sensors to be worn at any orientation at a given position on the body, and achieve nearly the same activity recognition performance as the ordinary system for which the sensor units are not rotatable. The proposed techniques can be applied to existing wearable systems without much effort, by simply transforming the time-domain sensor data at the pre-processing stage. © 2017 by the authors. Licensee MDPI, Basel, Switzerland.
Open Access
Analog CMOS implementation of cellular neural networks
(IEEE, 1993) Baktır, I. A.; Tan, M. A.
The analog CMOS circuit realization of cellular neural networks with transconductance elements is presented. This realization can be easily adapted to various types of applications in image processing just by choosing the appropriate transconductance parameters according to the predetermined coefficients. The effectiveness of the designed circuits for connected component detection is shown by HSPICE simulations. For “fixed function” cellular neural network circuits the number of transistors are reduced further by using multi-input transconductance elements.
Open Access
Applying deep learning in augmented reality tracking
(IEEE, 2016-11-12) Akgül, Ömer; Penekli, H. I.; Genç, Y.
An existing deep learning architecture has been adapted to solve the detection problem in camera-based tracking for augmented reality (AR). A known target, in this case a planar object, is rendered under various viewing conditions including varying orientation, scale, illumination and sensor noise. The resulting corpus is used to train a convolutional neural network to match given patches in an incoming image. The results show comparable or better performance compared to state of art methods. Timing performance of the detector needs improvement but when considered in conjunction with the robust pose estimation process promising results are shown. © 2016 IEEE.
Open Access
Approximating the stochastic growth model with neural networks trained by genetic algorithms
(2006) Kıykaç, Cihan
In this thesis study, we present a direct numerical solution methodology for the onesector nonlinear stochastic growth model. Rather than parameterizing or dealing with the Euler equation, like other methods do, our method directly parameterizes the policy function with a neural network trained by a genetic algorithm. Since genetic algorithms are derivative free and the policy function is directly parameterized, there is no need for taking derivatives. While other methods are bounded by the existence of required derivatives in higher dimensional state spaces, our method preserves its functionality. As genetic algorithms are global search algorithms, our method’s results are robust whatever the search space is. In addition to the stochastic growth model, to observe the performance of the method under real conditions, we tested the method by adding capital adjustment costs to the model. Under all parameter configurations, the method performs quite well.
Open Access
Artificial neural network modeling and simulation of in-vitro nanoparticle-cell interactions
(American Scientific Publishers, 2014) Cenk, N.; Budak, G.; Dayanik, S.; Sabuncuoglu, I.
In this research a prediction model for the cellular uptake efficiency of nanoparticles (NPs), which is the rate that NPs adhere to a cell surface or enter a cell, is investigated via an artificial neural network (ANN) method. An appropriate mathematical model for the prediction of the cellular uptake rate of NPs will significantly reduce the number of time-consuming experiments to determine which of the thousands of possible variables have an impact on NP uptake rate. Moreover, this study constitutes a basis for targeted drug delivery and cell-level detection, treatment and diagnosis of existing pathologies through simulating NP-cell interactions. Accordingly, this study will accelerate nanomedicine research. Our research focuses on building a proper ANN model based on a multilayered feed-forward back-propagation algorithm that depends on NP type, size, surface charge, concentration and time for prediction of cellular uptake efficiency. The NP types for in-vitro NP-healthy cell interaction analysis are polymethyl methacrylate (PMMA), silica and polylactic acid (PLA), all of whose shapes are spheres. The proposed ANN model has been developed on MATLAB Programming Language by optimizing a number of hidden layers (HLs), node numbers and training functions. The datasets are obtained from in-vitro NP-cell interaction experiments conducted by Nanomedicine and Advanced Technology Research Center. The dispersion characteristics and cell interactions with different NPs in organisms are explored using an optimal ANN prediction model. Simulating the possible interactions of targeted NPs with cells via an ANN model will be faster and cheaper compared to the excessive experimentation currently necessary.
Open Access
Associative memory design using overlapping decompositions
(Pergamon Press, 2001) Akar, M.; Sezer, M. E.
This paper discusses the use of decomposition techniques in the design of associative memories via artificial neural networks. In particular, a disjoint decomposition which allows an independent design of lower-dimensional subnetworks and an overlapping decomposition which allows subnetworks to share common parts, are analyzed. It is shown by a simple example that overlapping decompositions may help in certain cases where design by disjoint decompositions fails. With this motivation, an algorithm is provided to synthesize neural networks using the concept of overlapping decompositions. Applications of the proposed design procedure to a benchmark example from the literature and to a pattern recognition problem indicate that it may improve the effectiveness of the existing methods.
Open Access
Automatic radar antenna scan type recognition in electronic warfare
(Institute of Electrical and Electronics Engineers, 2011-10-04) Barshan, B.; Eravci, B.
We propose a novel and robust algorithm for antenna scan type (AST) recognition in electronic warfare (EW). The stages of the algorithm are scan period estimation, preprocessing (normalization, resampling, averaging), feature extraction, and classification. Naive Bayes (NB), decision-tree (DT), artificial neural network (ANN), and support vector machine (SVM) classifiers are used to classify five different ASTs in simulation and real experiments. Classifiers are compared based on their accuracy, noise robustness, and computational complexity. DT classifiers are found to outperform the others.
Open Access
Autonomous air combat with reinforcement learning under different noise conditions
(IEEE - Institute of Electrical and Electronics Engineers, 2023-08-28) Taşbaş, A. S.; Serbest, S.; Şahin, Safa Onur; Üre, N. K.
The autonomous realization of air combat with reinforcement learning-based methods has recently become a prominent field of study. In this paper, we present a classifier architecture to solve the air combat problem in noisy environments, which is a sub-branch of this field. We collect data from environments with different noise levels using air combat simulation. Using these data, we train three different data sets with the number of state stacks 2, 4, and 8. We train neural network-based classifiers using these datasets. These classifiers adaptively estimate the noise level in the environment at each time step and activate the appropriate pre-trained reinforcement learning policy based on this estimate. In addition, we share the performance comparison of these classifiers in different state stacks.
Open Access
Balancing computation load and communication overhead with multilevel self organizing maps
(2001-07) Bıkmaz, Erdoğan
Open Access
BELS: a broad ensemble learning system for data stream classification
(2021-12) Bakhshi, Sepehr
Data stream classification has become a major research topic due to the increase in temporal data. One of the biggest hurdles of data stream classification is the development of algorithms that deal with evolving data, also known as concept drifts. As data changes over time, static prediction models lose their validity. Adapting to concept drifts provides more robust and better performing models. The Broad Learning System (BLS) is an effective broad neural architecture recently developed for incremental learning. BLS cannot provide instant response since it requires huge data chunks and is unable to handle concept drifts. We propose a Broad Ensemble Learning System (BELS) for stream classification with concept drift. BELS uses a novel updating method that greatly improves bestin- class model accuracy. It employs a dynamic output ensemble layer to address the limitations of BLS. We present its mathematical derivation, provide comprehensive experiments with 11 datasets that demonstrate the adaptability of our model, including a comparison of our model with BLS, and provide parameter and robustness analysis on several drifting streams, showing that it statistically significantly outperforms seven state-of-the-art baselines. We show that our proposed method improves on average 44% compared to BLS, and 29% compared to other competitive baselines.
Open Access
Beyond Bouma's window: how to explain global aspects of crowding?
(Public Library of Science, 2019-05) Doerig, A.; Bornet, A.; Rosenholtz, R.; Francis, G.; Clarke, Aaron M.; Herzog, M. H.
In crowding, perception of an object deteriorates in the presence of nearby elements. Although crowding is a ubiquitous phenomenon, since elements are rarely seen in isolation, to date there exists no consensus on how to model it. Previous experiments showed that the global configuration of the entire stimulus must be taken into account. These findings rule out simple pooling or substitution models and favor models sensitive to global spatial aspects. In order to investigate how to incorporate global aspects into models, we tested a large number of models with a database of forty stimuli tailored for the global aspects of crowding. Our results show that incorporating grouping like components strongly improves model performance. Author summary Visual crowding highlights interactions between elements in the visual field. For example, an object is more difficult to recognize if it is presented in clutter. Crowding is one of the most fundamental aspects of vision, playing crucial roles in object recognition, reading and visual perception in general, and is therefore an essential tool to understand how the visual system encodes information based on its retinal input. Hence, classic models of crowding have focused only on local interactions between neighboring visual elements. However, abundant experimental evidence argues against local processing, suggesting that the global configuration of visual elements strongly modulates crowding. Here, we tested all available models of crowding that are able to capture global processing across the entire visual field. We tested 12 models including the Texture Tiling Model, a Deep Convolutional Neural Network and the LAMINART neural network with large scale computer simulations. We found that models incorporating a grouping component are best suited to explain the data. Our results suggest that in order to understand vision in general, mid-level, contextual processing is inevitable.
Open Access
Çarpmasız yapay sinir ağı
(IEEE, 2015-05) Akbaş, Cem Emre; Bozkurt, Alican; Çetin, A. Enis; Çetin-Atalay, R.; Üner, A.
Bu bildiride çarpma işlemi kullanmadan oluşturulan bir Yapay Sinir Ağı (YSA) sunulmaktadır. Girdi vektörleri ve YSA katsayılarının iç çarpımları çarpmasız bir vektör işlemiyle hesaplanmıştır. Yapay sinir ağının eğitimi sign-LMS algoritması ile yapılmıştır. Önerilen YSA sistemi, hesap gücü kısıtlı olan veya düşük enerji tüketimine ihtiyaç duyulan mikroişlemcilerde kullanılabilir.
Open Access
Circuit partitioning using mean field annealing
(Elsevier, 1995) Bultan, T.; Aykanat, Cevdet
Mean field annealing (MFA) algorithm, proposed for solving combinatorial optimization problems, combines the characteristics of neural networks and simulated annealing. Previous works on MFA resulted with successful mapping of the algorithm to some classic optimization problems such as traveling salesperson problem, scheduling problem, knapsack problem and graph partitioning problem. In this paper, MFA is formulated for the circuit partitioning problem using the so called net-cut model. Hence, the deficiencies of using the graph representation for electrical circuits are avoided. An efficient implementation scheme, which decreases the complexity of the proposed algorithm by asymptotical factors is also developed. Comparative performance analysis of the proposed algorithm with two wellknown heuristics, simulated annealing and Kernighan-Lin, indicates that MFA is a successful alternative heuristic for the circuit partitioning problem. © 1995.
Open Access
Classification of leg motions by processing gyroscope signals
(IEEE, 2009) Tunçel, Orkun; Altun, Kerem; Barshan, Billur
In this study, eight different leg motions are classified using two single-axis gyroscopes mounted on the right leg of a subject with the help of several pattern recognition techniques. The methods of least squares, Bayesian decision, k-nearest neighbor, dynamic time warping, artificial neural networks and support vector machines are used for classification and their performances are compared. This study comprises the preliminary work for our future studies on motion recognition with a much wider scope.
Open Access
Classifying human leg motions with uniaxial piezoelectric gyroscopes
(2009) Tunçel O.; Altun, K.; Barshan, B.
This paper provides a comparative study on the different techniques of classifying human leg motions that are performed using two low-cost uniaxial piezoelectric gyroscopes worn on the leg. A number of feature sets, extracted from the raw inertial sensor data in different ways, are used in the classification process. The classification techniques implemented and compared in this study are: Bayesian decision making (BDM), a rule-based algorithm (RBA) or decision tree, least-squares method (LSM), k-nearest neighbor algorithm (k-NN), dynamic time warping (DTW), support vector machines (SVM), and artificial neural networks (ANN). A performance comparison of these classification techniques is provided in terms of their correct differentiation rates, confusion matrices, computational cost, and training and storage requirements. Three different cross-validation techniques are employed to validate the classifiers. The results indicate that BDM, in general, results in the highest correct classification rate with relatively small computational cost. © 2009 by the authors.
Open Access
Comparative analysis of different approaches to target classification and localization with sonar
(IEEE, 2001-08) Ayrulu, Birsel; Barshan, Billur
The comparison of different classification and fusion techniques was done for target classification and localization with sonar. Target localization performance of artificial neural networks (ANN) was found to be better than the target differentiation algorithm (TDA) and fusion techniques. The target classification performance of non-parametric approaches was better than that of parameterized density estimator (PDE) using homoscedastic and heteroscedastic NM for statistical pattern recognition techniques.
Open Access
Comparative analysis of different approaches to target differentiation and localization with sonar
(Elsevier, 2003) Barshan, B.; Ayrulu, B.
This study compares the performances of different methods for the differentiation and localization of commonly encountered features in indoor environments. Differentiation of such features is of interest for intelligent systems in a variety of applications such as system control based on acoustic signal detection and identification, map building, navigation, obstacle avoidance, and target tracking. Different representations of amplitude and time-of-2ight measurement patterns experimentally acquired from a real sonar system are processed. The approaches compared in this study include the target differentiation algorithm, Dempster-Shafer evidential reasoning, different kinds of voting schemes, statistical pattern recognition techniques (k-nearest neighbor classifier, kernel estimator, parameterized density estimator, linear discriminant analysis, and fuzzy c-means clustering algorithm), and artificial neural networks. The neural networks are trained with different input signal representations obtained usingpre-processing techniques such as discrete ordinary and fractional Fourier, Hartley and wavelet transforms, and Kohonen's self-organizing feature map. The use of neural networks trained with the back-propagation algorithm, usually with fractional Fourier transform or wavelet pre-processing results in near perfect differentiation, around 85% correct range estimation and around 95% correct azimuth estimation, which would be satisfactory in a wide range of applications. © 2002 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.
Open Access
Control of subsonic cavity flows by neural networks-analytical models and experimental validation
(American Institute of Aeronautics and Astronautics, 2005) Efe, M. Ö.; Debiasi, M.; Yan, P.; Özbay, Hitay; Samimy, M.
Flow control is attracting an increasing attention of researchers from a wide spectrum of specialties because of its interdisciplinary nature and the associated challenges. One of the main goals of The Collaborative Center of Control Science at The Ohio State University is to bring together researchers from different disciplines to advance the science and technology of flow control. This paper approaches the control of subsonic cavity flow, a study case we have selected, from a computational intelligence point of view, and offers a solution that displays an interconnected neural architecture. The structures of identification and control, together with the experimental implementation are discussed. The model and the controller have very simple structural configurations indicating that a significant saving on computation is possible. Experimental testing of a neural emulator and of a directly-synthesized neurocontroller indicates that the emulator can accurately reproduce a reference signal measured in the cavity floor under different operating conditions. Based on preliminary results, the neurocontroller appears to be marginally effective and produces spectral peak reductions analogous to those previously observed by the authors using linearcontrol techniques. The current research will continue to improve the capability of the neural emulator and of the neurocontroller.
Open Access
Deep learning based channel equalization for MIMO ISI channels
(2022-09) Eren, Berke
Future wireless communications is expected to bring significant changes along with a number of emerging technologies such as 5G, virtual reality, edge computing, and IoT. These developments pose unprecedented demands in terms of capacity, coverage, latency, efficiency, flexibility, compatibility, and quality of experience on wireless communication systems. Machine Learning (ML) techniques are considered as a promising tool to tackle this challenge due to their ability to manage big data, powerful nonlinear mapping, and distributed computing capabilities. There have been many research results addressing different aspects of ML algorithms and their connections to wireless communications; however, there are still various challenges that need to be addressed. In particular, their use for communication systems with memory, is not fully investigated. With this motivation, this thesis considers an application of ML, in particular, deep learning (DL), techniques for communications over intersymbol interference (ISI) channels. In this thesis, we propose DL-based channel equalization algorithms for channels with ISI. We introduce three different DL-based ISI detectors, namely sliding bidirectional long short term memory (Sli-BiLSTM), sliding multi layer perceptron (Sli-MLP), and sliding iterative (Sli-Iterative), and demonstrate that they are computationally efficient and capable of performing equalization under a variety of channel conditions with the knowledge of the channel state information. We also employ sliding bidirectional gated recurrent unit (Sli-BiGRU) and Sli- MLP, which are more suitable for use with fixed ISI channels. As an extension, we also examine DL-based equalization techniques for multiple-input multipleoutput (MIMO) ISI channels. Numerical results show that proposed models are well suited for equalization of ISI channels with perfect as well as noisy CSI for a broad range of signal-to-noise ratio (SNR) levels as long as the ISI length is very close to the optimal solution, namely, the maximum likelihood sequence estimation, implemented through the Viterbi Algorithm while having considerably less complexity, and they have superior performance compared to MMSE-based channel equalization.