International Journal on Artificial Intelligence Tools

Intelligent problem-solving technology is a typical application of artificial intelligence in the educational field. The purpose of intelligent problem-solving is to enable machine to solve problems like human beings and help people to find useful and accurate information in the test-questions. Correct understanding of test-questions is one of the key techniques of intelligent problem-solving. The existing methods are mainly to extract simple relations from text and graphs of test-questions. In the absence of text, deep understanding of graph elements and complex relationships are difficult, which is a challenge for machines. However, the information contained in motion test-question graphs is more abundant than that in the text, which is more beneficial for machine to solve problems. To solve this problem, based on the graph classification and characteristics analysis on test-questions, a novel automatic and weak text-dependent information extraction and thorough understanding method for test-question graph of junior high school physical mechanical motion is proposed (TQGIEU). It can extract information and understand regarding to test-question graphs based on several image preprocessing techniques and neural network topologies combined with the characteristics of test-question graphs, without relying on text processing. It will generate a readable mechanical motion solution at last. It fills the blank of automatic information extraction and thorough understanding for the test-question research of physical mechanical motion. The experimental results show that the accuracy of graph classification and line segmentation model are 99.60% and 90.87% respectively. The average accuracy of TQGIEU on test dataset is 83.08%, and the total F1 score of TQGIEU is 0.859, indicating that TQGIEU can provide a good information extraction and thorough understanding service for test-question graphs of junior high school physical mechanical motion with high performance.

Detecting intents and extracting necessary contextual information (aka named entities) in input utterances are two fundamental tasks in understanding what the users say in chatbot systems. While most work in this field has been dedicated to high-resource languages in popular domains like business and home automation, little research has been done for low-resource languages, especially in a less popular domain like education. To narrow this gap, this paper presents the first study on learning to detect student intents and to recognize named entities in the education domain targeted to the Vietnamese language. Specifically, we first introduce a complete corpus consisting of 3690 utterances of students. It was manually annotated with both named entities and intent information at two levels of granularity: the fine-grained and the coarse-grained levels. We then systematically investigate different approaches to deal with the two tasks using not only independent but also joint learning architectures. The experimental results show that the joint architectures based on pre-trained language models are superior in boosting the performance of both tasks. They outperformed the conventional independent learning architectures which looked at the two tasks separately. Moreover, to further enhance the final performance, this paper proposes a technique to enrich the models with more useful linguistic features. Compared to the standard approaches, we achieve considerably better results for two tasks in both architectures. Overall, for the named entity recognition task, the best model yielded an F₁ score of 88.61%. For the intent detection task, it yielded F₁ scores of 94.36% and 91.62% at the coarse-grained and fine-grained levels, respectively.

Clause Learning is one of the most important components of a conflict driven clause learning (CDCL) SAT solver that is effective on industrial SAT instances. Since the number of learned clauses is proved to be exponential in the worst case, it is necessary to identify the most relevant clauses to maintain and delete the irrelevant ones. As reported in the literature, several learned clauses deletion strategies have been proposed. However the diversity in both the number of clauses to be removed at each step of reduction and the results obtained with each strategy increase the difficulty to determine which criterion is better. Thus, the problem to select which learned clauses are to be removed during the search step remains very challenging. In this paper, we propose a novel approach to identify the most relevant learned clauses without favoring or excluding any of the proposed measures, but by adopting the notion of dominance relationship among those measures. Our approach bypasses the problem of difference in the results obtained with different measures and reaches a compromise between the measures assessments. Furthermore, the proposed approach also reduces the non-trivial problem of choosing the number of clauses to delete at each reduction of the learned clause database.

Diabetic retinopathy (DR), a common complication of diabetes, is one of the leading causes of visual loss in a growing population. Thus, it is essential to identify DR at an early stage in order to minimize the problem of vision loss. As a result, the Henry Gas SailFish Optimizer (HGSO) algorithm and a successful lesion segmentation method based on Entropy Weighted and Kernalized Power K-Means Clustering (EWKPC) are developed for the DR detection. However, the SailFish Optimizer (SFO) and Henry Gas Solubility Optimization (HGSO) were combined to form the novel approach known as HGSO. Here, noise from the image is removed during pre-processing using the Region of Interest (ROI) approach and median filtering. Another use of the suggested EWKPC approach is lesion segmentation. Based on the outcomes of the data augmentation phase, Deep Convolutional Neural Network (DCNN)-based DR detection is carried out using the created HGSO algorithm. Along with effective performance, the created HGSO-based DCNN also displayed higher accuracy (0.910), specificity (0.929), and sensitivity (0.909).

Forecasting solar radiation for a given region is an emerging field of study. It will help to identify the places for installing large-scale photovoltaic-systems, designing energy-efficient buildings and energy estimation. The different machine learning kernel-based approaches for prediction problems uses either a local or global kernel. These models can provide either strong training capability or good generalization performance. In this paper, a new hybrid kernel is proposed using the combination of a local and global kernel. A novel algorithm Hybrid Kernel-based Extreme Learning Machine is proposed for predictive modelling of Incident Solar Radiation (ISR) time series using new hybrid kernel. The proposed algorithm uses the surface, atmospheric, cloud properties obtained from the MODIS instrument and observed ISR at time t – 1 to predict ISR at time t. This study is conducted for 41 diverse sites of Australia for the period of 2012–2015. Further, the proposed model is experimented with other time series datasets to prove its efficacy. It is shown that the proposed methodology outperforms five other benchmarking methods in terms of MAE and Willmott’s Index (WI). Therefore, the suggested approach can be used for modelling solar energy at a national scale using remotely-sensed satellite footprints.

Collection and synthesis of evidence is a key task in the development of the simulation models required for health technology assessment (HTA). The implementation of some of these models, such as discrete event simulation (DES) models, presents technical difficulties and requires higher technical skills. This work presents a method to extract the knowledge stored in an ontology, Rare Disease Ontology for Simulation (RaDiOS), to generate a DES model. RaDiOS is a domain ontology focused on collecting evidence on rare diseases for simulation models. We reviewed and enhanced the ontology to increase its semantic expressiveness. Besides, we developed a transformation tool (RaDiOS-MTT) to automatically generate DES models from the knowledge stored in the ontology. We defined a set of “synthetic” diseases, with simple natural histories, represented them in RaDiOS, and compared the results of the automatically generated simulation models with their manually created counterparts. Afterwards, we used a case study on a real intervention (newborn screening for profound biotinidase deficiency) to validate our approach. The automatically generated models for the synthetic diseases mimicked their programmatic counterparts in structure and results. The same happened to the model for profound biotinidase deficiency.

Inverse frequent itemset mining can be successfully modelled as an instance of the Probabilistic Satisfiability problem. Given a transaction database we can perform a frequent itemset mining algorithm, like the Apriori algorithm, to obtain useful itemset collections such as frequent or closed itemsets. We then use these itemset collections as frequency constraints in order to reconstruct the original database by solving a linear programming problem. There are cases however, where the reconstructed database is not in direct agreement with the original one. In this study, we analyse the degree of similarity between the original database and the reconstructed one, when different variations of condensed itemset representations are used as the initial frequency constraints. We examine how much the initial database properties and itemset relations have been preserved when we emphasize on the frequent itemset border and assess database quality by measuring database distance metrics. As this solution framework presents increased computational cost, we also consider a heuristic approach that is based on the notion that a transaction can be also considered as an itemset and compare the strengths and weaknesses of each framework when the same conditions apply. We manage to improve the efficiency of existing heuristic based approaches to the problem by utilizing smaller initial itemset collections. Our work contributes in (a) privacy preserving data mining and (b) gain in transaction database storage memory savings.

Many problems in NLP such as language translation and sentiment analysis have shown a lot of improvement in recent years. As simpler language problems are solved or better understood, the focus shifts to more complex problems such as semantic analysis and understanding. Unfortunately, a lot of studies in the literature suffer from a too much specificity problem. The algorithms and datasets are too domain specific. In this study, we analyze and elaborate on this notion of generality. Instead of selecting a highly specialized data set for semantic analysis, we take a generic and possibly dry data set, and we study how a plain vanilla Transformer performs in learning higher level semantic patterns beyond what was obvious or expected. We tune our Transformer model on a classic language task to ensure correct performance. Once tuned, the goal is to select sentences with specific key words and study whether higher level semantic patterns may have been learned by our model. We believe that we obtained promising results. The average BLEU score for sentences less than 25 words is equal to 39.79. Our initial qualitative analysis of possible semantic content of interest shows a 17 percent rate in finding interesting semantic patterns. We provide discussion of data driven results of unexpectedness as a measure of semantic learning.

Traffic forecasting is an integral part of modern intelligent transportation systems. Although many techniques have been proposed in the literature to address the problem, most of them focus almost exclusively on forecasting accuracy and ignore other important aspects of the problem. In the paper at hand, a new method for both accurate and fast large-scale traffic forecasting, named “sparse feature regression”, is presented. Initially, a set of carefully selected features is extracted from the available traffic data. Then, some of the initial features are sparsified, namely they are transformed into sets of sparse features. Finally, a linear regression model is designed using the sparse feature set, which is trained by solving an optimization problem using a sparse approximate pseudoinverse as a preconditioner. We evaluated the proposed method by conducting experiments on two real-world traffic datasets, and the experimental results showed that the method presents the best balance between accuracy of predictions and time required for achieving them, in comparison with a set of benchmark models.

In this manuscript, a proficient control strategy-based improved Quasi-Z-Source Cascaded Multilevel Inverter (QZS-CMI) topology for interfacing photovoltaic (PV) system is proposed. The control strategy is joint execution of both recalling-enhanced recurrent neural network (RERNN) and Shell Game Optimization (SGO), therefore it is called RERNN-SGO technique. The major intention of proposed system determined the photovoltaic system efficiency and maximize the power extraction. The interface among load and PV dc source is to accomplish by the QZSI. At first, the objective function is determined depending on the control constraint and parameters, like current, voltage, modulation index, power. These parameters are utilized to propose RERNN-SGO method. The RERNN-SGO method is used to improve the power delivery, voltage profile and minimum power oscillation for power distribution with load. The maximal power delivered the load is to ensure the artificial intelligence (AI) strategy with the help of maximal power point tracking (MPPT). To regulate shoot through duty ratio and modulation load, the proposed AI method is used. The proposed system is inspired on MATLAB/Simulink site, and then efficiency is related with existing systems under various load conditions. The efficiency of proposed system in Case 1 and Case 2 is 87.363% and 85.3904%.

Illumination pattern recognition of face image has always been a hot research topic in the field of human-computer interaction, and has been widely used in lighting recovery, virtual scene construction and other multimedia fields. Most of the traditional methods achieve this task by analyzing the illumination components from the image texture and structure information, which is often considered to be indirect, complex and time-consuming. This paper introduces a Double-inputs Illumination Pattern Recognizing Model (DIPRM) with Automatic Shadow Detection Network (ASDN) based on Convolution Neural Networks (CNNs). In the proposed coherent system framework, we first annotate the shadow regions of face images with uneven illumination to obtain the face shadow detection dataset. Second, an ASDN which has the encoder-decoder structure is designed. The main architecture of the ASDN is based on the nested U-Net, and for this nesting, the attention mechanism is applied to fuse the output of each sublayer. Third, a double-inputs Facial Illumination Pattern Recognizing Network (FIPRN) following the ASDN is organized, which consists of AlexNet and the attention module. As the double-inputs, the binary image after the shadow segmentation from the ASDN and the original image are input into the FIPRN to make the whole network converge to a good collaboration state eventually. For shadow detection task, the ASDN was evaluated in comparisons with U-Net and UNet++. Experimental results demonstrated that the ASDN achieved an average IoU and Dice gains of 1.9 and 1.2 points over the base-line model with best results. Moreover, the FIPRN was tested in comparison with some baseline models in illumination pattern recognizing task, where the results demonstrated that it achieved an accuracy rate gain of 1.0 points than the AlexNet with a signal-input.

Deep learning has been successfully applied to video anomaly detection. However, the way that deep network learn spatio-temporal features autonomously will ignore the specificity of different pattern features. Therefore, this paper focuses on how to efficiently learn deep appearance feature, introduces the idea of learning appearance information by predicting future frame, and proposes dual stream conditional generative adversarial network fusion for video abnormal behavior detection. The video frame and its corresponding optical flow image are transferred to the conditional generative adversarial network to learn the motion feature representation. In addition, inputting the video frame and its corresponding future frame to the network to generate the appearance representation complementary to motion feature. The model is only trained with normal events, therefore it is not able to generate abnormal events accurately. During the test, for the foreground moving targets, the images generated by the model are compared with the ground truth to obtain a two-stream anomaly probability distribution model based on the mean square error used to achieve the purpose of region anomaly detection. Experiments on the public datasets show that the proposed method can effectively detect and locate abnormal behaviors in the video.

There is a wave of interest in using physics-informed neural networks for solving differential equations. Most of the existing methods are based on feed-forward networks, while recurrent neural networks solvers have not been extensively explored. We introduce a reservoir computing (RC) architecture, an echo-state recurrent neural network capable of discovering approximate solutions that satisfy ordinary differential equations (ODEs). We suggest an approach to calculate time derivatives of recurrent neural network outputs without using back-propagation. The internal weights of an RC are fixed, while only a linear output layer is trained, yielding efficient training. However, RC performance strongly depends on finding the optimal hyper-parameters, which is a computationally expensive process. We use Bayesian optimization to discover optimal sets in a high-dimensional hyper-parameter space efficiently and numerically show that one set is robust and can be transferred to solve an ODE for different initial conditions and time ranges. A closed-form formula for the optimal output weights is derived to solve first-order linear equations in a one-shot backpropagation-free learning process. We extend the RC approach by solving nonlinear systems of ODEs using a hybrid optimization method consisting of gradient descent and Bayesian optimization. Evaluation of linear and nonlinear systems of equations demonstrates the efficiency of the RC ODE solver.

In recent years, the development of deep learning has contributed to various areas of machine learning. However, deep learning requires a huge amount of data to train the model, and data collection techniques such as web crawling can easily generate incorrect labels. If a training dataset has noisy labels, the generalization performance of deep learning significantly decreases. Some recent works have successfully divided the dataset into samples with clean labels and ones with noisy labels. In light of these studies, we propose a novel data expansion framework to robustly train the models on noisy labels with the attention mechanisms. First, our method trains a deep learning model with the sample selection approach and saves the samples selected as clean at the end of training. The original noisy dataset is then extended with the selected samples and the model is trained on the dataset again. To prevent over-fitting and allow the model to learn different patterns of the selected samples, we leverage the attention mechanism of deep learning to modify the representation of the selected samples. We evaluated our method with synthetic noisy labels on CIFAR-10 and CUB-200-2011 and real-world dataset Clothing1M. Our method obtained comparable results to baseline CNNs and state-of-the-art methods.