Deep neural networks (DNN) have remarkably progressed in applications involving large and complex datasets but have been criticized as a black-box. This downside has recently become a motivation for the research community to pursue the ideas of hybrid approaches, resulting in novel hybrid systems classified as deep neuro-fuzzy systems (DNFS). Studies regarding the implementation of DNFS have rapidly increased in the domains of computing, healthcare, transportation, and finance with high interpretability and reasonable accuracy. However, relatively few survey studies have been found in the literature to provide a comprehensive insight into this domain. Therefore, this study aims to perform a systematic review to evaluate the current progress, trends, arising issues, research gaps, challenges, and future scope related to DNFS studies. A study mapping process was prepared to guide a systematic search for publications related to DNFS published between 2015 and 2020 using five established scientific directories. As a result, a total of 105 studies were identified and critically analyzed to address research questions with the objectives: (i) to understand the concept of DNFS; (ii) to find out DNFS optimization methods; (iii) to visualize the intensity of work carried out in DNFS domain; and (iv) to highlight DNFS application subjects and domains. We believe that this study provides up-to-date guidance for future research in the DNFS domain, allowing for more effective advancement in techniques and processes. The analysis made in this review proves that DNFS-based research is actively growing with a substantial implementation and application scope in the future. 

/pdf/deep-neuro-fuzzy-system-application-trends-challenges-and-fd1u9097.pdf

Deep Neuro-Fuzzy System application trends, challenges, and future perspectives: a systematic survey

A comprehensive review of deep neuro-fuzzy system architectures and their optimization methods

Predicting airborne pollutant concentrations and events in a commercial building using low-cost pollutant sensors and machine learning: A case study

In this paper, a numerical approach is proposed based on least squares support vector regression for solving Volterra integral equations of the first and second kind. The proposed method is based on using a hybrid of support vector regression with an orthogonal kernel and Galerkin and collocation spectral methods. An optimization problem is derived and transformed to solving a system of algebraic equations. The resulting system is discussed in terms of the structure of the involving matrices and the error propagation. Numerical results are presented to show the sparsity of resulting system as well as the efficiency of the method.

Least squares support vector regression for solving Volterra integral equations

Machine learning may enable the automated generation of test oracles. We have characterized emerging research in this area through a systematic literature review examining oracle types, researcher goals, the ML techniques applied, how the generation process was assessed, and the open research challenges in this emerging field. Based on a sample of 22 relevant studies, we observed that ML algorithms generated test verdict, metamorphic relation, and---most commonly---expected output oracles. Almost all studies employ a supervised or semi-supervised approach, trained on labeled system executions or code metadata---including neural networks, support vector machines, adaptive boosting, and decision trees. Oracles are evaluated using the mutation score, correct classifications, accuracy, and ROC. Work-to-date show great promise, but there are significant open challenges regarding the requirements imposed on training data, the complexity of modeled functions, the ML algorithms employed---and how they are applied---the benchmarks used by researchers, and replicability of the studies. We hope that our findings will serve as a roadmap and inspiration for researchers in this field.

Using machine learning to generate test oracles: a systematic literature review

One of the challenging issues in software designing is testing the product in different condition. Various software Oracles had suggested in the literature, and the aim of all of them is minimizing the time and cost of the testing process. Software test Oracles have designed to do this job automatically with as less as possible human contribution. In this work, a novel Oracle based on deep learning and fuzzy inference system introduced. For this purpose, by the utility of Takagi-Sugeno-Kang fuzzy inference, the output of software mapped to the fuzzy space, and the deep neural network has trained by this data. Finally, data has remapped to the primary form and used as the competitor stage input. To validate the performance of the Oracle, four different models have chosen to assess the Oracle enforcement, and after training the Oracle by the correct output of applications, source codes have changed manually, and the efficiency of the Oracle monitored. Several measures have been applied to evaluate the accuracy of the test Oracle, and it is observed that in most cases Oracle correctly could detect the correct and false results. Finally, designing Oracles requires several preliminaries and in this work we only focus on the architecture of the system.

Performing Software Test Oracle Based on Deep Neural Network with Fuzzy Inference System

Similar to any service or product, industrialization of On-Orbit Servicing (OOS) demands performance enhancement through introducing relevant autonomy elements in planning and executing single and multiple servicing missions. This paper proposes an overall mission architecture for performing multiple on-orbit servicing missions by a fleet of servicers in the form of free-flying single-arm space manipulators. The architecture targets to improve the two key industrialization criteria of resource and service. In the far-range rendezvous with target satellites, the servicers burn most of their fuel. Furthermore, the time that servicers spend in transfer orbits determines the approximate duration of a servicing mission. Hence, as part of resource management, the presented architecture first identifies the main contributors to the fuel consumption and mission duration in far-range rendezvous phase of OOS missions being: (i) the location of the parking orbit, (ii) the type of transfer trajectories, and (iii) the dispatch scheduling. As the result, separate optimization loops are considered for minimizing the mission costs, across the OOS industry. Servicers are suggested to form an equally phased constellation in a parking orbit close to Sun-synchronous orbits in the Low Earth Orbital (LEO) region, where 57.5% of operational LEO satellites reside. A satellite in the parking orbit constellation is named “Administrator”, whose sole purpose is to plan and manage servicing missions. The Administrator determines the optimal number and sequence of servicing missions that must be performed by the available servicers, and the optimal transfer trajectories servicers shall follow to reach the targets. Upon completion of their missions, each servicer returns to the parking orbit and occupies the available position that requires the lowest fuel consumption to enter. In almost 90% of servicers' lifetime, they are in an idle state in the parking orbit awaiting dispatch or in transfer orbits. To enhance quality of the provided service, the proposed architecture suggests effective use of this time to task servicers with performing machine learning that helps improve the functionality of their guidance, navigation and control systems in upcoming missions. The task involves trajectory learning for a servicer's manipulator system in free-floating regime to reach a simulated moving target while avoiding (virtual) obstacles and compensating for environmental disturbances. Both supervised and unsupervised machine learning techniques are considered, and based on a qualitative analysis, the unsupervised DDPG algorithm is deemed most applicable in the free-floating trajectory learning task.

A Mission Architecture for On-Orbit Servicing Industrialization

Deep learning as a valuable intelligence tool to deal with complicated problems plays a crucial role in the 21st century. The utility of deep learning in solving partial differential equations (PDEs) is an interesting application of AI, which has been considered in recent years. However, supervision of learning procedure needs to have considerable labeled data to train the network, and this method could not be a beneficial technique to deal with unknown PDEs which we do not have any labeled data. To tackle this issue, in this paper a new method will be presented to solve PDEs only by using boundary and initial condition. Weakly supervision as an efficient method can provide an ideal bed to tackle boundary and initial value problems. To have better judgment about this method we chose Reaction-Diffusion equation as a versatile equation in engineering and science to be solved as a case study. By using the weakly supervised method and the finite difference method reaction-diffusion equation have solved, and the results of these methods have been compared. It has been shown that the results of deep learning have high consistency with finite difference results, and weakly supervised learning can be introduced as an efficient method to solve different types of differential equations.

Weakly Supervised Learning Technique for Solving Partial Differential Equations; Case Study of 1-D Reaction-Diffusion Equation

Nonlinear Model Predictive Control of J2-perturbed impulsive transfer trajectories in long-range rendezvous missions

Deep learning is a crucial point of valuable intelligence resources to deal with complicated mathematical problems. The effectiveness of deep learning in solving differential equations has been considered over the past few years. Supervision of the learning process requires significant information to be marked in order to train the network. Nevertheless, this approach could not be a helpful strategy in case of unknown differential equations that we have no identified data. In order to address this problem, a new method for solving differential equations will be introduced in this paper using only the boundary and initial conditions. As an efficient method, inadequate monitoring can provide an ideal bed to fix boundary and initial value issues. For verification of the proposed method, a reaction–diffusion equation was performed. This equation has a variety of applications in engineering and science.

/pdf/solving-partial-differential-equations-by-a-supervised-3ktw87597s.pdf

Sanaz Samsam

Papers

Performing Software Test Oracle Based on Deep Neural Network with Fuzzy Inference System

A Mission Architecture for On-Orbit Servicing Industrialization

Weakly Supervised Learning Technique for Solving Partial Differential Equations; Case Study of 1-D Reaction-Diffusion Equation

Nonlinear Model Predictive Control of J2-perturbed impulsive transfer trajectories in long-range rendezvous missions

Solving partial differential equations by a supervised learning technique, applied for the reaction–diffusion equation