Sequential learning navigation method and general correction model for Autonomous Underwater Vehicle

Achieving accurate navigation and localization is crucial for Autonomous Underwater Vehicle (AUV). Traditional navigation algorithms, such as Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF), require the system model and measurement model for state estimation to obtain the AUV position. However, this may introduce modeling errors and state estimation errors which will affect the final precision of AUV navigation system to a certain extent. To avoid these problems, in this paper, we proposed a deep framework - NavNet - by taking AUV navigation as a deep sequential learning problem. Firstly, the proposed NavNet can take raw sensor data at different frequencies as input, which benefits from the sequential learning capability of Recurrent Neural Network (RNN). Secondly, NavNet takes advantage of a simplified attention mechanism and Fully Connected (FC) layers to output AUV displacements per unit time, which accomplishes low-frequency AUV navigation by accumulation of it. More importantly, there is no need for the model building and state estimation with NavNet, which avoids the import of relevant errors. We compare the performance of NavNet to EKF and UKF using collected data by running Sailfish in the sea. Experimental results show that NavNet has an excellent performance in terms of both the navigation accuracy and fault tolerance. In addition, a reliable fusion strategy of NavNet and conventional method is applied to achieve high-frequency AUV navigation. The experimental results show that the proposed architecture can be a reliable supplement to limit the error growth of conventional algorithms.

A sequential subspace learning method and its application to dynamic texture analysis

This study proposes an approach to improving emotion classification performance by introducing a Sequential Emotion Learning (SEL) method. Conventional learning methods often struggle with fine-grained emotion categories. To address this, the SEL approach first trains the model on seven basic emotions, which are relatively easier to classify due to their clear distinctions. The model is then fine-tuned using 24 more nuanced emotion labels, enhancing its ability to tackle complex emotion classification tasks. Experimental results suggest that the SEL method performs better than the baseline, achieving higher accuracy from the early stages of training. The SEL model also reaches its peak performance relatively quickly and shows improved classification capabilities on unseen, general sentences, indicating its robustness across different text scenarios. These results suggest that the SEL method can effectively improve emotion classification, particularly in tasks that require distinguishing complex emotions. This sequential learning approach offers a potential advantage over traditional methods and may be applied to other domains that involve intricate classification tasks. Future research can explore the generalizability of this method to other classification problems to further enhance its utility.

Sequential and Reinforcement Learning for Neurocontrol

Recurrent neural network (RNN) has been widely used for sequential learning which has achieved a great success in different tasks. The temporal convolutional network (TCN), a variant of one-dimensional convolutional neural network (CNN), was also developed for sequential learning in presence of sequence data. RNN and TCN typically captures long-term and short-term features in temporal or spatial domain, respectively. This paper presents a new sequential learning, called the convolutional recurrent network (CRN), which fulfills TCN as an encoder and RNN as a decoder so that the global semantics as well as the local dependencies are simultaneously characterized from sequence data. To facilitate the interpretation and robustness in neural models, we further develop the stochastic modeling for CRN based on variational inference. The merits of CNN and RNN are then incorporated in inference of latent space which sufficiently produces a generative model for sequential prediction. Experiments on language model shows the effectiveness of stochastic CRN when compared with the other sequential machines.

Multi-sensor real-time monitoring of dam behavior using self-adaptive online sequential learning

Structural-optimized sequential deep learning methods for surface soil moisture forecasting, case study Quebec, Canada

The recent diffusion of online education (both MOOCs and e-courses) has led to an increased economic and scientific interest in e-learning environments. As widely documented, online students have a much higher chance of dropping out than those attending conventional classrooms. It is of paramount interest for institutions, students, and faculty members to find more efficient methodologies to mitigate withdrawals. Following the rise of attention on the Student Dropout Prediction (SDP) problem, the literature has witnessed a significant increase in contributions to this subject. In this survey, we present an in-depth analysis of the state-of-the-art literature in the field of SDP, under the central perspective, but not exclusive, of machine learning predictive algorithms. Our main contributions are the following: (i) we propose a comprehensive hierarchical classification of existing literature that follows the workflow of design choices in the SDP; (ii) to facilitate the comparative analysis, we introduce a formal notation to describe in a uniform way the alternative dropout models investigated by the researchers in the field; (iii) we analyse some other relevant aspects to which the literature has given less attention, such as evaluation metrics, gathered data, and privacy concerns; (iv) we pay specific attention to deep sequential machine learning methods—recently proposed by some contributors—which represent one of the most effective solutions in this area. Overall, our survey provides novice readers who address these topics with practical guidance on design choices, as well as directs researchers to the most promising approaches, highlighting current limitations and open challenges in the field.

Neural network models, which were initially used to model the processes of recognition of graphic images, are now widely used in the field of recognition of multi-parameter objects and in adjusting the parameters of complex systems. Such an application is effective in solving problems of information security management. With the help of neural networks, the tasks of classifying threats, choosing the parameters of protection measures, and adjusting the operating modes of information systems are successfully solved. In the scientific literature, there is a wide variety of approaches and methods for creating artificial neural networks. At the initial stage of designing information security systems, when it is necessary to determine the main parameters of the software being created, it poses a problem for the designer. Miscalculations made at this stage of the design can lead to the unsuccessful completion of the entire project. The article provides a brief comparative analysis of artificial neural networks that differ in the methods of setting (training) the network and in the form of its structure. The applicability, advantages, and disadvantages of such methods for tuning neural networks are characterized. These are the method of backpropagation of an error, genetic algorithm, iterative Widrow-Hoff algorithm with variable step, modified least squares method, and sequential learning method. The differences in the structure for networks designed to solve the problems of adjusting information processes and recognizing multi-parameter objects are shown. To regulate the processes, the structure of the neural network is studied on the example of the fuzzy neural network ANFIS. For the classification problems, the structure of a multilayer perceptron is given, in which the structure of the inner layers reflects the ontological network of the subject area under consideration. The presented results can be used to justify the type of neural network used for a specific task in the field of information security.

This paper investigates sequential learning method with new form of weight update applied on a decomposed form of training algorithms using Radial Basis Function (RBF) network. Adding each basis function to the hidden layer during the course of training facilitate the weight update to be decomposed on neuron by neuron basis. A new form weight update is introduced where the weight update is based on minimal displacement of the current input elements to the elements of the nearest centre of the Gaussian neuron.

This chapter addresses the problem of time-varying nonlinear modeling and monitoring of a continuously stirred tank reactor (CSTR) process using state estimation techniques. These techniques include the extended Kalman filter (EKF), particle filter (PF), and the more recently the variational Bayesian filter (VBF). The objectives of this chapter are threefold. The first objective is to use the variational Bayesian filter with better proposal distribution for nonlinear states and parameters estimation. The second objective is to extend the state and parameter estimation techniques to better handle nonlinear and non-Gaussian processes without a priori state information, by utilizing a time-varying assumption of statistical parameters. The third objective is to apply the state estimation techniques EKF, PF and VBF for time-varying nonlinear modeling and monitoring of CSTR process. The estimation performance is evaluated on a synthetic example in terms of estimation accuracy, root mean square error and execution times.

Objective: State estimation is used as a necessary tool for online monitoring, analysis and control of power systems. The main objective of this paper is to deal with static state estimation using the snap shot data and dynamic state estimation which accounts for the time varying behaviour of power system states in to consideration. Methods/Analysis: Effects of inclusion of PMU measurements along with the available metered data have been explored using the weighted least square state estimation technique in this paper. A comparative analysis of static state estimation with and without bad data has been carried out and the bad data has been identified and eliminated by using largest normalized residue test. Findings: To investigate the time varying nature of the system states linear state estimator with second order approximation and kalman filter techniques has been proposed in this paper. Case studies are conducted on IEEE 14 bus test system and the test results obtained from non linear, linear first order, second order and kalman filter techniques have been compared. Application/Improvements: Correction of state variables are obtained using linear state estimation with first order and second order approximation and kalman filter techniques has been compared to get the better state estimation algorithm.

Flat panel displays, such as the thin film transistor liquid crystal display, the organic light-emitting diode, and the polymer light-emitting diode, have been widely applied in many fields in recent decades. To ensure the quality of these displays, defect inspection is crucial. Mura defects, which are phenomena of uneven screen displays, are the most challenging visual defects to detect. This paper presents an online sequential classifier and transfer learning (OSC-TL) method for the online training and classification of Mura defects. OSC-TL is a new method that combines a deep convolutional feature extractor and a sequential extreme learning machine classifier. It makes online sequential training in a production line possible. To demonstrate the performance of the OSC-TL method, several experiments are performed to compare the results of this method with those of other popular classification algorithms. The experimental results show that the computational resources and time consumed by OSC-TL are well below those of other common methods because of the feature transfer and the online sequential classification strategies. Consequently, the OSC-TL method has been implemented in our automated optical inspection equipment to perform online Mura defect classification. It is able to learn and recognize a Mura defect image within 1.5 milliseconds.

System state monitoring is crucial in order to maintain the system security, reliability and quality of a power system. Considering the limitation of monitoring devices in distribution network, state estimation technique could be employed. State estimation is an action of appraising an unknown system variables based on limited number of real-time measurements. The main objective of this paper is to find the suitable state estimation that able to estimate the state of the islanded distribution network. In this paper, state estimation for a distribution network is developed and enhanced by incorporating composite load model. By incorporating this model, the calculated state variables of the network will be more accurate (to obtain the most practical value of the system state variables). Moreover, the proposed state estimation results are compared with backward-forward sweep load flow technique to justify the system state variables.

When an aircraft experiences component failures or external damage, system parameter identification is instrumental to the determination of the remaining flight envelope. On-line parameter estimation methods are either constantly adapting, by using a forgetting algorithm to prevent saturation of the covariance matrix, or use a covariance reset after a failure occurs. Previous research showed that methods with reset were more stable but the detection of the failure to trigger the reset was prone to false alarms. Traditionally these failure detection methods were based on changes in the statistical data of the residue. We previously developed three improved reset methods based on the Sequential Probability Ratio Test (SPRT) and applied them to a simple linear longitudinal aircraft model. In this paper the new detection methods are evaluated for practical usability. A non-linear Cessna Citation model and a non linear Piper Seneca model were subjected to control failures, engine failures and asymmetries in both smooth air and turbulent conditions and supplied the data to test the new detection methods. Adaptations to new detection methods were developed to enable to work under these realistic conditions. The performance of the new methods, expressed in detection time, false alarms and missed detections was compared with the traditional residue based methods. With some modification the new methods were capable of detecting failures under diverse circumstances. From the three new methods, the Sequential Learning (SL) method had the best performance and was superior to traditional systems based on residue. Furthermore, compared to constantly adapting method, using a forgetting algorithm, the new method showed greater stability in the calculation of the minimum lateral control speed.