Browsing by Subject "Recurrent neural networks"

Now showing 1 - 12 of 12

Open Access
Control and system identification of legged locomotion with recurrent neural networks
(2022-06) Çatalbaş, Bahadır
In recent years, robotic systems have gained massive popularity in the industry, military, and daily use for various purposes, thanks to advancements in artificial intelligence and control theory. As an exciting sub-branch of robotics with their differences and opportunities, legged robots have the potential to diversify and spread the use of robotic systems to new fields. Especially, legged locomotion is a desirable ability for mechanical systems where agile mobility and a wide range of motions are required to fulfill the designated task. On the other hand, unlike wheeled robots, legged robot platforms have a hybrid dynamical structure consisting of the flight and contact phases of the legs. Since the hybrid dynamical structure and nonlinear dynamics in the robot model make it challenging to apply control and perform system identification for them, various methods are proposed to solve these problems in the literature. This thesis focuses on developing new neural network-based techniques to apply control and system identification to legged locomotion so that robotic platforms can be designed to move efficiently as animal counterparts do in nature. In the first part of this thesis, we present our works on neural network-based controller development and evaluation studies for bipedal locomotion. In detail, neural controllers, in which long short-term memory (LSTM) type of neuron models are employed at recurrent layers, are utilized in the feedback and feedforward paths. Supervised learning data sets are produced using a biped robot platform controlled by a central pattern generator to train these neural networks. Then, the ability of the neural networks to perform stable gait by controlling the robot platform is assessed under various ground conditions in the simulation environment. After that, the stable walking generation capacity of the neural networks and the central pattern generators are compared with each other. It is shown that the proposed neural networks are more successful gait controllers than the central pattern generator, which is employed to generate data sets used in training. In the second part, we present our studies on the end-to-end usage of neural networks in system identification for bipedal locomotion. To this end, supervised learning data sets are produced using a biped robot model controlled by a central pattern generator. After that, neural networks are trained under series-parallel and parallel system identification schemes to approximate the input-output relations of the biped robot model. In detail, different neural models and neural network architectures are trained and tested in an end-to-end manner. Among neuron models, LeakyReLU and LSTM are found as the most suitable feedforward and recurrent neuron types for system identification, respectively. Moreover, neural network architecture consisting of recurrent and feedforward layers is found to be efficient in terms of learnable parameter numbers for system identification of the biped robot model. The last part discusses the results obtained in the control and system identification studies using neural networks. In the light of acquired results, neural networks with recurrent layers can apply control and systems identification in an end-to-end manner. Finally, the thesis is completed by discussing possible future research directions with the obtained results.
Open Access
Deep fractional Fourier networks
(2024-08) Koç, Emirhan
This thesis introduces the integration of the fractional Fourier Transform (FrFT) into the deep learning domain, with the aim of opening new avenues for incorporating signal processing into deep neural networks (DNNs). This work starts by introducing FrFT into recurrent neural networks (RNNs) for time series prediction, leveraging its ability and flexibility to perform infinitely many continuous transformations and offering an alternative to the traditional Fourier Transform (FT). Despite the initial success, a significant challenge identified is the manual tuning of the fraction order parameter a, which can be cumbersome and limits broader applicability. To overcome this limitation, we introduce a novel approach where the fraction order a is treated as a learnable parameter within deep learning models. First, a theoretical foundation is established to support the learnability of this parameter, followed by extensive experimentation in image classification and time series prediction tasks. The results demonstrate that incorporating a learnable fraction order significantly improves model performance, particularly when integrated with well-known architectures such as ResNet and VGG models. Furthermore, the thesis proposes fractional Fourier Pooling (FrFP), a pooling technique that replaces traditional Global Average Pooling (GAP) layers in Convolutional Neural Networks (CNNs). FrFP enhances feature representation by processing intermediate signal regions, leading to better model performance and offering a new perspective on integrating signal transformations within deep learning frameworks. Overall, this thesis contributes to the growing body of research exploring advanced signal processing techniques in deep learning, highlighting the potential of FrFT as a powerful tool for improving model accuracy and efficiency across various applications.
Open Access
Deep learning based decoders for concatenated codes over insertion and deletion channels
(2025-01) Kargı, Eksal Uras
Channels with synchronization errors, including insertion/deletion channels, are of significant importance, as they are encountered in various systems, such as communication networks and various storage technologies, including DNA data storage. Serially concatenated codes where the outer code is a powerful channel code, such as a low-density parity-check (LDPC) or convolutional code, and the inner code is a watermark or marker code, are shown to be effective solutions over such channels. In particular, the use of marker codes, referring to insertion of preselected sequences in the transmitted data stream periodically, are shown to work well in regaining synchronization at the receiver and achieving improved error rate performance compared to other alternatives. In the current literature, maximum a posteriori (MAP) detector realized by the well-known forward-backward algorithm is commonly employed to decode the inner marker code and estimate the log-likelihood ratios (LLRs) of the bits encoded by the outer code, and the resulting log-likelihood estimates are fed to the outer decoder to estimate the transmitted data. Alternative to the MAP detector, this thesis proposes deep learning-based solutions to estimate the LLRs of the coded bits in the paradigm of concatenated codes, exploiting the marker information and addressing some limitations of conventional methods. Bit-level deep learning-based detectors offer good alternatives when the channel statistics are not perfectly available at the decoder, degrading of the performance of the MAP detector. They can also be employed for one-shot decoding when the outer code is a convolutional code. Also developed are symbol-level deep learning-based detectors to exploit the correlations among adjacent bits at the detector output. Contrary to the existing symbol-level decoders for insertion/deletion channels, the newly proposed approaches can go beyond the case of combining three bits, offering further enhancements in performance while keeping the complexity tolerable. As a final contribution, deep learning-based detectors are developed for insertion and deletion channels that are further exacerbated by inter-symbol interference, e.g., modeling bit-patterned media recording channels, and their performance is studied via numerical examples.
Open Access
Deep-learning for communication systems: new channel estimation, equalization, and secure transmission solutions
(2023-08) Gümüş, Mücahit
Traditional communication system design takes a model-based approach that aims to optimize relevant performance metrics using somewhat simple and tractable channel and signal models. For instance, channel codes are designed for simple additive white Gaussian or fading channel models, channel equalization algorithms are based on mathematical models for inter-symbol interference (ISI), and channel estimation techniques are developed with the underlying channel statistics and characterizations in mind. Through utilizing superior mathematical models and expert knowledge in signal processing and information theory, the model-based approach has been highly successful and has enabled development of many communication systems until now. On the other hand, beyond 5G wireless communication systems will further exploit the massive number of antennas, higher bandwidths, and more advanced multiple access technologies. As communication systems become more and more complicated, it is becoming increasingly important to go beyond the limits of the model-based approach. Noting that there have been tremendous advancements in learning from data over the past decades, a major research question is whether machine learning based approaches can be used to develop new communication technologies. With the above motivation, this thesis deals with the development of deep neural network (DNN) solutions to address various challenges in wireless communications. We ﬁrst consider orthogonal frequency division multiplexing (OFDM) over rapidly time-varying multipath channels, for which the performance of standard channel estimation and equalization techniques degrades dramatically due to inter-carrier in-terference (ICI). We focus on improving the overall system performance by designing DNN architectures for both channel estimation and data demodulation. In addition, we study OFDM over frequency-selective channels without cyclic preﬁx insertion in an eﬀort to improve the overall throughputs. Speciﬁcally, we design a recurrent neu-ral network to mitigate the eﬀects of ISI and ICI for improved symbol detection. Furthermore, we explore secure transmission over multi-input multi-output multi-antenna eavesdropper wiretap channels with ﬁnite alphabet inputs. We use a linear precoder to maximize the secrecy rate, which beneﬁts from the generalized singular value decomposition to obtain independent streams and exploits function approximation abilities of DNNs for solving the required power allocation problem. We also propose a DNN technique to jointly optimize the data precoder and the power allocation for artiﬁcial noise. We use extensive numerical examples and computational complexity analyses to demonstrate the eﬀectiveness of the proposed solutions.
Open Access
Efficient online training algorithms for recurrent neural networks
(2020-12) Vural, Nuri Mert
Recurrent Neural Networks (RNNs) are widely used for online regression due to their ability to learn nonlinear temporal dependencies. As an RNN model, Long-Short-Term-Memory Networks (LSTMs) are commonly preferred in prac-tice, since these networks are capable of learning long-term dependencies while avoiding the exploding gradient problem. On the other hand, the performance improvement of LSTMs usually comes with the price of their large parameter size, which makes their training signiﬁcantly demanding in terms of computational and data requirements. In this thesis, we address the computational challenges of LSTM training. We introduce two training algorithms, designed for obtaining the online regression performance of LSTMs with less computational requirements than the state-of-the-art. The introduced algorithms are truly online, i.e., they do not assume any underlying data generating process and future information, except that the dataset is bounded. We discuss theoretical guarantees of the introduced algo-rithms, along with their asymptotic convergence behavior. Finally, we demon-strate their performance through extensive numerical studies on real and synthetic datasets, and show that they achieve the regression performance of LSTMs with signiﬁcantly shorter training times.
Open Access
Fractional fourier transform in time series prediction
(IEEE, 2022-12-09) Koç, Emirhan; Koç, Aykut
Several signal processing tools are integrated into machine learning models for performance and computational cost improvements. Fourier transform (FT) and its variants, which are powerful tools for spectral analysis, are employed in the prediction of univariate time series by converting them to sequences in the spectral domain to be processed further by recurrent neural networks (RNNs). This approach increases the prediction performance and reduces training time compared to conventional methods. In this letter, we introduce fractional Fourier transform (FrFT) to time series prediction by RNNs. As a parametric transformation, FrFT allows us to seek and select better-performing transformation domains by providing access to a continuum of domains between time and frequency. This flexibility yields significant improvements in the prediction power of the underlying models without sacrificing computational efficiency. We evaluated our FrFT-based time series prediction approach on synthetic and real-world datasets. Our results show that FrFT gives rise to performance improvements over ordinary FT.
Open Access
Identification and adaptive control of bipedal robot motion with artificial neural networks
(2024-07) Çatalbaş, Burak
Artificial Neural Networks (ANNs) is one of the most popular fields of machine learning thanks to the critical improvements in the last decade, including their applications in the field of robotics and control. The important usage of neural networks in robotics makes it possible for robots to act and interact similar to humans. In this manner, legged robots are important platforms to mimick human locomotion. However, there are significant difficulties to apply system identification and control schemes for these hybrid dynamical structures. With this purpose, this thesis focuses on using artificial neural network-based novel techniques on these problems, for reaching to an efficient walking ability for bipedal robot systems like their counterparts in the nature. In this thesis, our work to find and apply our novel techniques is mainly divided into two parts. In the first part, inspired by a class of activation functions frequently used in deep learning literature, we propose a novel activation function and investigate its performance in various segmentation and classification tasks by using different well-known datasets. In the second part of the thesis, biped robot locomotion is chosen as the main topic. Separate datasets are created for three experiment configurations. For 2D and 3D simulations, locomotion control, system identification and adaptive control are applied with neural networks for successful periodical walking with low errors, having approximations of robot models and preparing for the adaptive learning using both control and identification blocks, respectively. For 2D physical robot system, system identification is completed for a walking dataset generated with varying speed levels. For all cases, proposed novel activation function DELU (ExtendeD Exponential Linear Unit) and its tuned functions are tried together with other activation functions in comparison, to reach better performances.
Open Access
Non-uniformly sampled sequential data processing
(2019-09) Şahin, Safa Onur
We study classification and regression for variable length sequential data, which is either non-uniformly sampled or contains missing samples. In most sequential data processing studies, one considers data sequence is uniformly sampled and complete, i.e., does not contain missing input values. However, non-uniformly sampled sequences and the missing data problem appear in a wide range of fields such as medical imaging and financial data. To resolve these problems, certain preprocessing techniques, statistical assumptions and imputation methods are usually employed. However, these approaches suffer since the statistical assumptions do not hold in general and the imputation of artificially generated and unrelated inputs deteriorate the model. To mitigate these problems, in chapter 2, we introduce a novel Long Short-Term Memory (LSTM) architecture. In particular, we extend the classical LSTM network with additional time gates, which incorporate the time information as a nonlinear scaling factor on the conventional gates. We also provide forward pass and backward pass update equations for the proposed LSTM architecture. We show that our approach is superior to the classical LSTM architecture, when there is correlation between time samples. In chapter 3, we investigate regression for variable length sequential data containing missing samples and introduce a novel tree architecture based on the Long Short-Term Memory (LSTM) networks. In our architecture, we employ a variable number of LSTM networks, which use only the existing inputs in the sequence, in a tree-like architecture without any statistical assumptions or imputations on the missing data. In particular, we incorporate the missingness information by selecting a subset of these LSTM networks based on presence-pattern of a certain number of previous inputs.
Open Access
Nonstationary time series prediction with Markovian switching recurrent neural networks
(2021-07) İlhan, Fatih
We investigate nonlinear prediction for nonstationary time series. In most real-life scenarios such as ﬁnance, retail, energy and economy applications, time se-ries data exhibits nonstationarity due to the temporally varying dynamics of the underlying system. This situation makes the time series prediction challenging in nonstationary environments. We introduce a novel recurrent neural network (RNN) architecture, which adaptively switches between internal regimes in a Markovian way to model the nonstationary nature of the given data. Our model, Markovian RNN employs a hidden Markov model (HMM) for regime transitions, where each regime controls hidden state transitions of the recurrent cell inde-pendently. We jointly optimize the whole network in an end-to-end fashion. We demonstrate the signiﬁcant performance gains compared to conventional methods such as Markov Switching ARIMA, RNN variants and recent statistical and deep learning-based methods through an extensive set of experiments with synthetic and real-life datasets. We also interpret the inferred parameters and regime belief values to analyze the underlying dynamics of the given sequences.
Open Access
Spatio-temporal forecasting over graphs with deep learning
(2020-12) Ceyani, Emir
We study spatiotemporal forecasting of high-dimensional rectangular grid graph structured data, which exhibits both complex spatial and temporal dependencies. In most high-dimensional spatiotemporal forecasting scenarios, deep learningbased methods are widely used. However, deep learning algorithms are overconfident in their predictions, and this overconfidence causes problems in the human-in-the-loop domains such as medical diagnosis and many applications of 5 th generation wireless networks. We propose spatiotemporal extensions to variational autoencoders for regularization, robustness against out-of data distribution, and incorporating uncertainty in predictions to resolve overconfident predictions. However, variational inference methods are prone to biased posterior approximations due to using explicit exponential family densities and mean-field assumption in their posterior factorizations. To mitigate these problems, we utilize variational inference & learning with semi-implicit distributions and apply this inference scheme into convolutional long-short term memory networks(ConvLSTM) for the first time in the literature. In chapter 3, we propose variational autoencoders with convolutional long-short term memory networks, called VarConvLSTM. In chapter 4, we improve our algorithm via semi-implicit & doubly semi-implicit variational inference to model multi-modalities in the data distribution . In chapter 5, we demonstrate that proposed algorithms are applicable for spatiotemporal forecasting tasks, including space-time mobile traffic forecasting over Turkcell base station networks.
Open Access
Two-legged robot motion control with recurrent neural networks
(Springer, 2022-04) Çatalbaş, Bahadır; Morgül, Ömer
Legged locomotion is a desirable ability for robotic systems thanks to its agile mobility and wide range of motions that it provides. In this paper, the use of neural network-based nonlinear controller structures which consist of recurrent and feedforward layers have been examined in the dynamically stable walking problem of two-legged robots. In detail, hybrid neural controllers, in which long short-term memory type of neuron models employed at recurrent layers, are utilized in the feedback and feedforward paths. To train these neural networks, supervised learning data sets are created by using a biped robot platform which is controlled by a central pattern generator. Then, the ability of the neural networks to perform stable gait by controlling the robot platform is examined under various ground conditions in the simulation environment. After that, the stable walking generation capacity of the neural networks and the central pattern generators are compared with each other. It is shown that the inclusion of recurrent layer provides smooth transition and control between stance and flight motion phases and L2 regularization is beneficial for walking performance. Finally, the proposed hybrid neural network models are found to be more successful gait controllers than the central pattern generator, which is employed to generate data sets used in training. © 2022, The Author(s), under exclusive licence to Springer Nature B.V.
Open Access
UKSB sinir ağları ile eksik veri kümesi üzerinde sıralı bağlanım
(IEEE, 2019-04) Şahin, Safa Onur
Bu bildiride, içerisinde eksik bilgi bulunan veri kümesinin Uzun Kısa-Soluklu Bellek (UKSB) sinir ağları ile sıralı bağlanımı çalışılmıştır. UKSB sinir ağını kullanan sıralı bağlanım uygulamalarında veri kümesi genellikle eksiksiz olarak olarak kabul edilir. Ancak, eksik veri problemi sıralı veri içeren gerçek hayat uygulamalarında sıklıkla karşılaşılan bir sorundur. Bu probleme çözüm amacıyla sunulan yöntemlerde eksik veri, sıralı verideki örüntüyü yakalayacak derecede modellenememekte ve bu yüzden yüksek performans artışları görünmemektedir. Bu bildiride, eksik veri, bağlanımı gerçekleştiren UKSB ağı tabanlı sinir ağının kendisi tarafından modellenmekte ve bağlanım sırasında üretilen bu veri kullanılmaktadır. Gerçek hayat uygulamalarından elde edilmiş sırlı veri kümeleriyle yapılan deneylerde, önerilen algoritmanın geleneksel metotlar karşısında üstün performans artışına sahip olduğu gözlemlenmiştir.