Search

There are a number of issues which are involved with organizing a conference. Among these issues, assigning conference-papers to reviewers is one of the most difficult tasks. Assigning conference-papers to reviewers is automatically the most crucial part. In this paper, we address this issue of paper-to-reviewer assignment, and we propose a method to model the reviewers, based on the matching degree between the reviewers and the papers by combining a preference-based approach and a topic-based approach. We explain the assignment algorithm and show the evaluation results in comparison with the Hungarian algorithm.

For centuries, the research paper have been the main vehicle for scientific progress. From the paper, readers in the scientific community are expected to extract all the relevant information necessary to reproduce and validate the results presented by the paper's authors. However, the increased use of computer software in science makes reproducing scientific results increasingly difficult. The research paper in its current state is no longer sufficient to fully reproduce, validate, or review a paper's experimental results and conclusions. This impedes scientific progress. To remedy these concerns, we introduce PaperMâche, a new system for creating dynamic, executable research papers. The key novelty of PaperMâche is its use of virtual machines, which lets readers and reviewers easily view and interact with a paper, and reproduce key experimental results. For authors, the Paper Mâche workbench provides an easy-touse interface to build an executable paper. By transforming the static research paper into a dynamic and interactive entity, Paper Mâche brings the presentation of scientific results into the 21st century. We believe that Paper Mâche will become indispensable to the scientific process, and increase the visibility of key findings among members and non-members of the scientific community.

The paper describes a system for executable papers for publishers enabling them to reuse content and to generate further advances of science and engineering. The executable algorithmic descriptions within a paper are presented in natural language sentences and basic code, thereby making long term compatibility absolute. Authors are required to use publicly numerical libraries on the Internet or references to publications with executable papers. As used by authors the system automatically creates a web of algorithmic knowledge on the Internet. Novelty of new algorithms in publications can be evaluated by automated tools available to authors, reviewers and readers of scientific papers published.

Image acquisition has great influence on the performance of any computer vision application. Different methods can be utilized to acquire the digital image of a paper, whilst scanning scheme is among the most attractive methods. This attractiveness is because of the fewer types of potential deformations and the low cost of the scanning devices, e.g. flatbed scanners. However, paper is commonly placed imperfectly on the scanner. This slight rotation is not usually based on a pivot around the paper's geometrical center (the well known regular rotation) but instead it is based on a pivot placed at the corner of the paper. Thus, the result is a digital image that is deformed with an “irregular rotation”. The characteristic of this deformation phenomenon is currently unknown to computer vision scientists. In this paper we provide an extensive investigation of this deformation. In addition, a new set of equations that sway and measure the transformation is proposed. Our investigation leads to the conclusion that the “irregular rotation” phenomenon produces a shear transformation. Furthermore, the experimental results confirm the theoretical findings.

What is the nature of knowledge representation needed for human-level artificial intelligence? This position paper contends that to achieve human-level AI, a system architecture for human-level knowledge representation would benefit from a neuro-symbolic approach combining deep neural networks with a ‘natural language of thought’ and would be greatly handicapped if relying only on formal logic systems.

Unimodal biometric systems are susceptible to a variety of problems such as noisy data, intra-class variations, limited degrees of freedom, non-universality, spoof attacks and unacceptable error rates. Some of these limitations can be addressed by deploy multimodal biometric systems that integrates the evidence presented by multiple sources of information The proposed system provides effective fusion scheme that combines information presented by the multiple domain experts based on the Rank level fusion integration method, thereby increasing the efficiency of the system which is not possible by the unimodal biometric system. The proposed multimodal biometric system has a number of unique qualities, including principal component analysis and fisher's linear discriminate methods for individual matchers authentication. The novel rank level fusion method is used in order to consolidate the results obtained from different biometric matchers.

Shredded document recovery is an important research sub-field of information security. The paper document is broken into a large number of similar shreds by the shredder. Usually these shreds imply some important information. If they are restored, it will have an important impact on judicial investigation, military command, and archaeological research. Therefore, it has a great value in research of shredded document recovery. In consideration of large differences between languages and great demands for Chinese document recovery, our study will focus on the shredded Chinese document. In this paper, a method of shredded document recovery based on character stroke feature is proposed. We use horizontal strokes to match Chinese characters. Finally the shredded document is restored. In the test of actual sample, the method proposed in this paper has achieved good recovery result. The average accuracy is 13.96% higher than the traditional algorithm.

In this paper, an approach to isolated sign language recognition with data provided by a depth camera is presented. In the introduced method, sequences of depth maps of dynamic sign language gestures are divided into smaller regions (cells). Then, statistical information is used to describe the cells. Since gesture executions have different lengths, the dynamic time warping (DTW) technique with the nearest neighbor (NN) rule is often employed for their comparison. However, due to time-consuming computations, the DTW limits the usability of the classifier. Therefore, in this paper, a selection of representative depth maps using statistics for cells is proposed. It is shown that such gesture representation can be successfully employed for isolated sign language recognition with the NN classifier using the city block distance. Furthermore, the NN rule with the DTW and the introduced statistics for cells provides superior gesture recognition performance.

The precise diagnostics of different diseases is very important for its treatment. It is particularly important to differentiate the disease on the early stages when the pathological alterations have not yet caused great harm to the whole organism, since this allows using greater number of therapies and increase the recovery probability. One of the methods of early diagnostics is based on examination of human biological liquids (blood, tears, cervical mucus, urine, etc,). A small drop of liquid is drawn on an object-plate and dried out slowly, thus a thin dry film (facies) remains. In the process of fluid crystallization there appear characteristic patterns (markers) in the facies. Each marker is a highly definite sign of some pathology even at an early stage of a disease development. When mass population health examination is carried out, it is necessary to analyze a large number of images This is the reason the problem of algorithm and software development for automated processing of images is rather urgent nowadays. Mathematical models of facies images are needed to formalize the tasks of image processing and synthesis of algorithms. The facies has a round shape, as it is obtained from a drop of liquid. Facies images have a radial-ring structure due to the peculiarities of the drying and crystallization of the contained fractions. These facies features should be taken into account in their imaging models. In this paper, autoregressive models of circular images are used to represent facies images. These models use grids defined on concentric circles or Archimedes’ spirals. The radial-ring structure of facies image correlations is naturally taken into account in these models. The paper contains examples of model identification, that is, estimation of the values ​​of its parameters at which the correlation properties of the model are closest to the characteristics of a given facies. It is shown that the classification of the facies type (corresponding to the general condition of the patient) is possible according to these estimates.

The past decade has witnessed an explosion in the generation of Arabic textual data, written in both modern standard Arabic (MSA) and in dialectal formats. Observing existing data reveals that verses from the Holy Quran are often quoted whether someone is giving a speech, conveying condolences, going through a hardship, and when expressing hope or joy among other emotions. Many applications can benefit from the automatic detection of those verses yet detecting Quranic Verses in text is not as easy as it may initially seem. This paper lists the challenges posed by this task, and proceeds to present an intelligent matching algorithm that addresses those. The paper also briefly describes a tool that has implemented the proposed algorithm, and which is currently publicly available as an open-source tool.

The development of the Internet of Things over the last decade has led to large amounts of data being generated at the network edge. This highlights the importance of local data processing and reasoning. Machine learning is most commonly used to automate tasks and perform complex data processing and reasoning. Collecting such data in a centralized location has become increasingly problematic in recent years due to network bandwidth and data privacy concerns. The easy-to-change behavior of edge infrastructure enabled by software-defined networking (SDN) allows IoT data to be gathered on edge servers and gateways, where federated learning (FL) can be performed: creating a centralized model without uploading data to the cloud. In this paper, we analyze the use of edge computing and federated learning, a decentralized machine learning methodology that increases the amount and variety of data used to train deep learning models. To the best of our knowledge, this paper reports the first use of federated learning to help the Microgrid Energy Management System (EMS) predict load and obtain promising results. Simulations were performed using TensorFlow Federated with data from a modified version of the Dataport site.

Nowadays, it is very important to quickly recognize the false information referred to as fake news. This is especially important in the case of news appearing on the Internet because of its wide and rapid spreading. It is equally important to be able to initially classify news as fake or true based on the title itself. In this paper, we propose an approach to classifying news based on the title without analyzing the other aspects. The obtained results will be compared with classification based on the whole news text. The goal of this work is to propose a method that balances between data analysis time and quality of classification in fake news prediction. We use natural language processing (NLP) to describe the title and text of the news. This is a complex process, requiring good analysis to be applied to classification. Therefore, the use of complex classifiers – in this case, classical ensemble methods – has been proposed in order to achieve a high quality of classification (measured by popular measure). In this paper, we present analyses of a real data set and results of news classification using the proposed model – including an ensemble of classifiers and single classifiers.

The human immune system is studied at different scales (molecular, cellular, tissue, organism, population) with varied granularity and spatiotemporal dimensions. Innate and adaptive immune responses to viral infections are researched in vitro, in vivo, and in silico using mathematical and computational methods. The construction of computational models is an iterative and interactive process which combines existing knowledge and hypotheses (knowledge-driven approach) with empirical datasets (data-driven approach). The resulting models should ideally provide reusability, composability, and interoperability to support system simulation. The main premises of this paper are (1) computational models can be used not only for representation of existing knowledge, but also as experimental tools to demonstrate, exemplify, create, and serve as test-beds for the analysis of empirical data;and (2) computational models should be evaluated in terms of their usability (defined as a measure of how well a specified group of users can achieve predefined goals effectively, efficiently, and satisfactorily in a specified context of use). The usability measure is specified by the context information: user characteristics, their goals, and additional constraints (e.g., data availability, temporal limitations). This paper focuses on the modelling process performed by multidisciplinary teams and the usability of models for the exploration of immune response to viral infections. (c) 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the Complex Adaptive Systems Conference, June 2021.

Conventionally, majority physicians used manual-based approach to determine heart sound parameters by manually printing heart sound signal waveform on the paper and trace heart sound parameters duration. Towards modern medical diagnostic technologies, this paper presents a modified framework of automated heart sound signal segmentation and parameters identification for deployment in computer aided auscultation. Envelope-based approach with reference information by abrupt changes and peak detection algorithm are proposed for PASCAL Classifying Heart Sounds Challenge database. Noted that the performance of segmentation process was measured by calculating the accuracy and F1-score while the heart sound parameters were identified by computing its mean and standard deviation. The findings yield F1-score and accuracy of proposed approach for segmentation and identification of normal heart sound signal at 95.29% and 91.0%, respectively. Heart sound parameters such as first sound, S1 duration, second sound, S2 duration, systole duration, diastole duration, heart cycle duration and ratio of systole and diastole were also determined. This proposed approach is suitable to be applied on two prominent heart sound signal peaks of S1 and S2, which continuously varies among samples due to the different auscultatory sites.

Anaphora resolution is one of the problems in natural language processing. It is the process of disambiguating the antecedent of a referring expression from the set of entities in a discourse. The correct interpretation of pronouns plays an important role in the construction of meaning. Thus, the resolution of pronominal anaphors remains a very important task for many natural language processing applications. Additionally, it plays an increasingly significant role in computational linguistics. However, a significant amount of work on anaphora resolution is focused on English; anaphora resolution for other languages, including Arabic, is still limited. In this paper, we present a new set of computational and linguistic features to resolve Arabic anaphors using a machine learning approach. In this paper, an in-depth study was conducted on a set of computational and linguistic features to exploit their effectiveness and investigate their effect on anaphora resolution. The aim was to efficiently integrate different feature sets and classification algorithms to synthesize a more accurate classification procedure. Four well-known machine learning algorithms k-nearest neighbor, maximum entropy, decision tree and meta-classifier, were employed as base-classifiers for each of the feature sets. A wide range of comparative experiments on Quran datasets was conducted, the discussion presented, and conclusions were drawn. The experimental results show that our approach gives satisfactory results.

In both developed and emerging countries, like Bangladesh, uterine cancer is one of the most common cancers for women. It is also the sixth most common cancer among females in the world and the fourteenth most common cancer in total. In recent years, the high prevalence of uterine cancer in women has risen dramatically. This covers several aspects that are linked to the signs and symptoms of this disease that are to be assessed and tested. Typically, these dimensions are expressed in quantitative and qualitative forms and also exist different kinds of uncertainty. Therefore, by using suitable methods to resolve the problem of uncertainty; otherwise, the detecting procedure will become unreliable. There are several systems proposed in this paper to tackle the problem. None of them, however, is worthy of solving the problem of ambiguity. This paper explains the development of a Belief Rule Based Expert System using the RIMER method, which by taking into account signs and symptoms, is able to detect the involvement of malignant tumors and risk factors for this cancer under uncertainty. Specifically, the system will expect a huge employment in diminishing the cost of lab assessments. The structure will support patients in making judicious strides well early. Moreover, this research has been done on MATLAB environment shown that the yields produced using the made system are more reliable than from manual structure, artificial neural network (ANN) and support vector machine (SVM).

Recently, two types of common sensors, LiDAR and Camera, show significant performance on all tasks in 3D vision. LiDAR provides accurate 3D geometry structure, while camera captures more scene context and semantic information. The fusion of two different sensor becomes a fundamental and common idea to achieve better performance. To give a thorough cognition of the complementary and boosting about two kind of sensors. This paper briefly reviews the fusion and enhancement systems between both two sensors in the field of depth completion, 3D object detection, 2D\3D semantic segmentation and 3D object tracking. Meanwhile, the state of art fusion algorithms is quantitatively demonstrated, in this paper, based on the in KITTI widely-used public dataset. Furthermore, the technical challenge and the future potential of LiDAR and camera fusion are also discussed.

The purpose of this paper is to investigate whether it is possible to predict text readability with ensemble-based classifiers. In this article, the authors calculated and analyzed the readability indices. In the next stage, they defined additional features for each text and determined the relationships between readability and features. Among the various tasks of machine learning, they chose the classification problem. The authors calculated and compared the accuracy of different machine learning models. After building the models, they proceeded to the Random decision forests model interpretation step using the SHAP method. The authors show that machine learning models based on only three features are capable of predicting text readability. Long sentences and a low percentage of stop words can cause low readability. The machine learning model shown in this paper allows to classify texts according to readability with a model accuracy of 0.9.

Researchers have extensively used machine learning techniques and data mining methods to build prediction models and classify data in various domains such as aviation, computer science, education, finance, marketing and particularly in medical field where those methods are applied as support systems for diagnosis and analysis in order to make better decisions. On this subject, our research paper attempts to assess the performance of Individual and Ensemble machine learning techniques based on the effectiveness and the efficiently, in terms of accuracy, specificity, sensitivity and precision to choose the most effective. The main object of our research paper is to define the best and effective machine learning approach for the Breast Cancer diagnosis and prediction. To achieve our objective, we applied individual based level machine learning algorithms Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Naive Bayes (NB), Decision tree (C4.5), Simple Logistic and well known ensembles methods like Majority Voting and Random Forest with 10 cross field technique on the Breast Cancer Diagnosis Dataset obtained from UCI Repository. The experimental results show that the Majority Voting Ensemble technique based on 3 top classifiers SVM, K-NN, Simple Logistic gives the highest accuracy 98.1% with the lowest error rate 0.01% and outperformed all other individual classifiers. This study demonstrates that our proposal approach based on Majority Voting Ensemble technique was the best classification machine learning model with the highest level of accuracy for breast cancer prediction and diagnosis. All experiments are effectuated within a simulation environment and realized in Weka data mining tool.

With the rapid development of deep learning and structures of neural networks, parameter settings of trained neural networks become new products from knowledge. Since supervised learning is used in 85% of machine learning to train the neural networks, well-labeled training data becomes new knowledge capital. In order to stimulate the knowledge sharing and protect the Intelligent Property (IP) right and privacy of knowledge capital at the same time, we need a framework to conduct training of neural networks with certain encrypted data and exchange training results with certain rewards. Federated learning provides a framework to train the networks without disclose the privacy and a block-chain based transaction system can organize a group of participants under certain consensus and reward mechanism. Their combination is an effective solution to manage knowledge capital, called knowledge capital bank in this paper. This paper defines the economic meaning of knowledge capital and all behaviors of related participants. Then we propose a block-chain based federated learning algorithm for secured knowledge sharing with IP and privacy protection. Finally, a complete design for a use case of this algorithm called Knowledge capital bank is demonstrated.

There are more than 80 million students in the Arab world. Students take tests and examinations as an intrinsic part of their educational journey to assess the quality of learning and understanding of the examined material. Exams require an immense amount of resources to be conducted, but even a bigger amount of resources to be scored. Scoring of exams and papers takes up enormous time and effort from teachers all over the Arab world. This time and effort can be better utilized in other teaching activities that would elevate the quality of learning in the region. For that reason, the work presented in this paper investigates multiple supervised learning approaches for the domain of Arabic short answer scoring. Moreover, due to the scarcity of Arabic datasets, and to our knowledge, there is only one publicly available dataset for this specific task(AR-ASAG). Thus, we introduce a new publicly available dataset AraScore-Dataset. The model proposed is evaluated on 3 different datasets: 1-AraScore-Dataset 2-AR-ASAG 3-Two translated answer sets taken from the Hewlett Foundation SAS dataset. Based on the investigation, the most effective approach for Arabic automated short answer scoring is proposed, and the performance of using our dataset is compared to other publicly available datasets. The aim is to scale efficient and unbiased content scoring applications across different Arabic educational domains. The results showed that our newly developed response-based system and AraScore-Dataset have achieved state-of-the-art performance on Arabic Automated Short Answer Scoring.

The purpose of current paper is to create a smart and effective tool for telemedicine to early detect and diagnose COVID-19 disease and therefore help to manage Pandemic Crisis (MCPC) in Sultanate of Oman, as a tool for future pandemic containment. In this paper, we used tools to create robust models in real-time to support Telemedicine, it is Machine Learning (ML), Deep Learning (DL), Convolutional Neural Networks using Tensorflow (CNN-TF), and CNN Deployment. These models will assist telemedicine, 1) developing Automated Medical Immediate Diagnosis service (AMID). 2) Analysis of Chest X-rays image (CXRs). 3) Simplifying Classification of confirmed cases according to its severity. 4) Overcoming the lack of experience, by improving the performance of medical diagnostics and providing recommendations to the medical staff. The results show that the best Regression among the five Regression models is Random Forest Regression. while the best classification among the eight classification models and Recurrent Neural Network using Tensorflow (RNNTF) is Random Forest classification, and the best Clustering model among two Clustering models is K-Means++. Furthermore, CNN-TF model was able to discriminate between those with positive cases Covid-19 and those with negative cases.

Predicting sports results is a well-known problem, and it applies to the more predictable individual sports as well as to the less predictable team sports – e.g., football. In the vast amount of information and parameters that are associated with team sports, there are few solutions related to the comprehensive approach of knowledge discovery in databases (KDD). In this paper, we propose a new approach to the whole process of discovering knowledge from data to predict the result of Bundesliga football matches. The main goal of the paper is to improve the prediction of football match results using heterogeneous ensembles of classifiers. To achieve the goal, it is necessary to prepare a training set and to develop a learning model – in this case, we analyze whether the use of limited information allows for the correct prediction of the results of football matches. The use of the developed model based on heterogeneous ensembles in the prediction of Bundesliga match results allows for improved prediction compared to classical ensembles models and single classifiers. Additionally, the possibilities arising from the use of heterogeneous ensembles of classifiers allow to distinguish some of the analyzed cases. The proposed approach was tested on real-world data, the results of Bundesliga football matches analyzed from the 2010/2011 season to the 2019/2020 season. Experiments were performed on methods selected based on an in-depth literature review. Conducted experiments, supported by the proper statistical analysis, confirm the usefulness of the proposed approach in predicting Bundesliga football results.

This paper describes the process of capturing an expert’s knowledge in the form of human understandable rules and then inserting them into a self-organizing map neural network. The Dynamic Cell Structure (DCS) is a form of self-organizing map used for many purposes, including classification and prediction. The DCS is a topology-preserving network that starts with no pre-structure, but assumes a structure once trained. This paper explores applying the DCS to analyse a social science data set that contains some of the many factors that may influence the mortality rate of a region.

In this paper, we propose an OCR system that recognizes the contents of manual writing in handouts. Since the target handouts contain a mixture of Japanese and mathematical formulas, we thought that one type of OCR systems alone would not provide sufficient accuracy. Therefore, in this proposed system, we will consider combining two types of OCR systems. The OCRs to be combined are Tesseract and Mathpix, which are good at recognizing Japanese and mathematical formulas, respectively. When the two OCRs are combined, the final plausible result must be selected, but we wondered if the OCR score could be used as an indicator of selection. The OCR score is a self-evaluation value that is output together with the recognition result. However, there is no guarantee that the OCR score and the actual recognition result show the same tendency. Therefore, in this paper, after classifying the written contents into three categories, words, sentences, and mathematical formulas, we verified whether the recognition results of each OCR can be appropriately selected based on the OCR score. At this point, the result is that further improvement is needed to make the selection based on the OCR score alone. However, if it becomes possible to classify handwritten contents with the proposal system by additional learning etc., it will be possible to analyze the handwritten contents efficiently, and it is expected that it will be possible to improve the handouts and provide feedback to the lesson plan.

Paper discusses modern methods for localization of mobile ground robot in outdoor environment. Special attention is payed to lidar-based methods which allow to do precise simultaneous localization and mapping (SLAM) independently to lighting conditions. Modular approach to lidar-based localization is proposed. It consists of three modules: map reconstructor, map merger and corrector, localization module. As map reconstructor we use 3D Lidar-based Graph SLAM, state-of-the-art method LOAM and its modification with point cloud segmentation techniques A-LeGO-LOAM. As map merger and corrector we apply the technique of interactive map correction for 3D Graph SLAM. For localization module we compare different real-time approaches: NDT scan matching and Localization with scan batch processing. They do not allow to localize the robot with an unknown initial position. To overcome this problem, a stage of global Monte Carlo localization is proposed, which allows to generate a switching condition and a rough initial pose. Matrix of localization convergence speed is introduced to study influence of possible inaccuracies of initial pose estimation to the convergence of real-time localization approach. An important emphasis in the paper is made to the evaluation of the localization quality with relative and absolute metrics. Experimental results are obtained for own data from mobile ground robot ClearPath Husky equipped with Velodyne HDL-32E lidar. Proposed approach to robot localization achieves real-time performance for different hardware platforms in particular single-board computer Nvidia Jetson Xavier. It demonstrates possibility for integration of studied modular localization approach to onboard control systems of autonomous vehicles.

Aiming at the requirements of effective assessment and accurate quantification of vehicle target apparent damage degree in war, natural disasters and other environments, this paper presents a damage assessment technique based on deep learning regression calculation. First, the image containing vehicle target is preprocessed by scale adjustment, segmentation and graying. Then, extracting and fusing the high-dimensional features of the preprocessed image through the deep convolution neural network. At last, obtaining the evaluation value of vehicle target damage degree through the fusion feature calculation of full connection regression network. In this paper, the automobile target is taken as the experimental object, and completing the relevant data collection, training and testing. The experimental results show the accuracy and effectiveness of this method for vehicle target apparent damage degree evaluation.

In this paper, we propose an ontology-based architecture for the classification of a scheduling problem. In fact, planning and scheduling is a major concern for the production enterprise as it contributes to its profit. Most of the time, researchers propose algorithms that relax some reel situations’ constraints; given that reel shop description are not taken directly from the floor supervisor, or given the complexity of the problem. The objective of ontology for the classification of scheduling problem is to build a bridge between practitioners and researchers in order to identify the reel classification of the problem. Thus, in this paper, we start by presenting a general model for the classification of scheduling problem. On this model, we build specific domain ontology. The specific domain ontology functions as a knowledge base and a data access point for the scheduling problem classification system.

Natural Language Processing (NLP) has witnessed considerable advances in textual understanding through statistical and rule-based techniques, which have now yielded to developments in the field of neural networks and deep learning. The paper surveys research relevant to the topic of Reading Comprehension and Question Answering (QA) implementations. The initial focus of the paper is on Attention and Transformer models. A brief description of the architectures is presented highlighting the essence of the ‘Attention is all you Need’ paper wherein the authors of Bidirectional Encoding Transformer (BERT) have elucidated the significant departure from the recurrence concept of Recurrent Neural Networks (RNN). Subsequently the trends in Open Domain Question Answering (ODQA) which mark the progression from the passage- based question answering is presented. Of particular interest is Haystack which is an end-to-end open-source framework for Question Answering & Neural search. This field seems a promising avenue for a more intelligent form of ‘search’. In a nutshell, the paper weaves through RNNs, Long Short-Term Memory (LSTM), and the currently trending Attention based Transformer models in NLP. Finally, we dwell on more contemporary pieces of research such as ODQA, Multi-Hop QA, evaluation using Adversarial Networks.

Imbalanced classification represents a challenge for supervised learning, as an unequal distribution of classes in the training data set is mainly connected to poor predictive performance for the minority class. However, usually the minority class is the most relevant one, from a practical perspective. But due to the imbalancement of the training data, the classification errors for the minority class are higher, as the classifiers are usually biased to predict the majority class. In this paper we investigate the use of autoencoders for improving the predictive performance for imbalanced binary classification problems. As an application domain we consider breast cancer detection, that is an imbalanced classification problem of great interest in the medical domain. According to the World Health Organisation, breast cancer represents the primary cause of cancer mortality in women. Nowadays there is an increasing interest in applying conventional machine learning and more recently deep learning techniques in the breast cancer detection field by helping medical experts in the early detection of the disease. One of the paper’s goal is to investigate the ability of deep autoencoders to learn patterns within the classes of benign and malignant instances. Secondly, we propose and compare two autoencoders-based classification models for breast cancer detection. The performances of the proposed models were empirically assessed on data sets previously used in the breast cancer detection literature. The results show that our best model compares favourably with the results of most of the classifiers used for comparison and that it is able to handle well the data imbalancement.

Iris and face are two most popular biometric traits. Unfortunately, their performance can be degraded when these systems encountered with non-ideal images. The primary motivation of this paper is to analyze the usability of a new feature rich biometric trait known as periocular region (nearby region around the eye) for two different non-ideal scenarios 1) Matching of images with different pose variation and 2) Matching of different side of periocular images (Left eye to Right eye & Right eye to Left eye). In this paper, seven different off-the-shelf deep learning based Convolutional Neural Networks (CNN) using transfer learning approach are implemented on UBIPr database. Experimental results demonstrated that for images with pose variation VGG 19 obtained maximum recognition accuracy while for matching of two different side of periocular region Res Net 18 outperform others. As an additional contribution, a novel approach for the use of mirror images to improve the matching accuracy of two different side of periocular region is also presented.

Sentiment Analysis deals with analysing the reviews stated by its consumers for any product. If such analysis is performed at a deeper level, it enables us to identify the consumer’s sentiment towards each feature of the product. The sentiment expressed may not be same towards each feature. The analysis of this sort is called Aspect Based Sentiment Analysis (ABSA) and it has been sub-divided into four subtasks. In this paper, the detailed study of the approaches used for the first subtask of ABSA, i.e. Aspect Term Extraction (ATE) is presented. This paper discusses how ATE can be performed for the reviews in a rich morphological language, like Hindi. The models proposed for ATE of Hindi Reviews are Conditional Random Field (CRF) and Bidirectional Long-Short-Term-Memory (Bi-LSTM) models with novel architecture. The CRF based approach with novel feature, ‘Cluster-id’ improved F-measure from 41.07% to 42.71%. However, with 5-fold cross-validation, the CRF model attained an F-measure of 44.54%. By using our proposed Bi-LSTM based model with PoS vector, the F-measure obtained is 44.49%.

With the rapid development of Unmanned Aerial Vehicle (UAV), stealth warship and stealth tank, the mission of radar has become more difficult. The modern radar system must detect and track target through strong clutter from many aspect, such as ground and sea. Radar detection for low slow small （LSS）targets has gradually become a new focus. LSS target has a small Radar Cross Section (RCS) because of its small size, making the SNR of the echo signal very low. At the same time, the Doppler frequency of slow target is very low, making it difficult for conventional signal processing methods to separate the Doppler information from the clutter spectrum. Thus, it does make sense to study how to detect the LSS target in strong clutter efficiently. In this paper, the distribution characteristics of radar clutter and coherent accumulation algorithm for LSS target detection in complex electromagnetic environment are studied. On the base of that, a double-threshold target detection method based on coherent accumulation is proposed. To check out the efficiency of the detection method, an experiment has been conducted, and the data collected was used to test the detection algorithm, and the result has been shown in this paper.

Assisting table tennis coaching using modern technologies is one of the most trending researches in the sports field. In this paper, we present a methodology to identify and recognize the wrong strokes executed by players to improve the training experience by the usage of an IR depth camera. The proposed system focuses mainly on the errors in table tennis player’s strokes and evaluating them efficiently and based on the analysis and classification of the data obtained from an IR depth camera using multiple algorithms. This paper is a continuation of our previous work [10], focusing more on identifying common wrong strokes in table tennis by utilizing IR depth camera classification algorithms. The classification of the mistakes that took place while playing can be classified based on each player dependently or independently for all players.

Visual tracking is an art of tracking a moving object over video frames using non-stationary cameras, for which feature descriptors of the target are computed and applied to a motion model. In this paper, SURF is used to compute the feature descriptors and Mean-Shift algorithm as a motion model. Mean-shift algorithm is fundamentally a logical approach to track the object on an image frame where the appearance is described by histograms. Mean-Shift tracking can directly be applied to SURF features but there is a big constraint of availability of an adequate number of feature keypoints for a given object. To get over this limitation, the re-projection technique is implemented in this paper for tracking an object in any video recorded from a mobile or stationary camera. Re-projection, online updates the histogram of object template for every frame by superimposing the feature keypoints from upcoming frames to the first frame. The object can be tracked using SURF descriptors without using any secondary information (color, texture, optical flow, gradient, etc.) regarding the object, which makes it computationally inexpensive to be used in real-time systems. Also, the SURF feature is calculated only for the target object that is to be tracked, this further reduces computational complexity. The simulation results prove the effectiveness of the work.

In today’s world of internet, a massive amount of data is getting generated every day and content-based classification of images is becoming an essential aspect for efficient retrieval of images and have attracted application in several fields and one of such field is sports. Sport is an integral part of everybody’s daily life and it is very important to play the sport with the right posture and environment otherwise it can lead to medical issues. This paper presents a robust framework for classifying the sport images based on the environment and related surroundings. In this paper, our approach is based on the use of the Inception V3 for the extraction of features and Neural Networks for the classification of various sport categories. Six categories rugby, tennis, cricket, basketball, volleyball, and badminton have been used for analysis and classification. To validate the effectiveness of the framework and Neural Networks, comparisons have been done with other classifiers like Random Forest, K-Nearest Neighbors (KNN) and Support Vector Machine (SVM). Our framework has successfully achieved an average accuracy of 96.64 % over six categories which demonstrate the effectiveness of the framework and can be used for the detection and classification of various sport activities in an efficient manner.

Analyzing sensor data from a production environment is quite challenging because of the high-dimensional nature of the data. In addition, the generated data is in the form of time-series, where the sequence of registrations may be of utmost significance. One of the main goals of the paper is to determine if the given time-series of feature combinations is normal or rare. This goal could successfully be achieved by combining multiple machine learning models. In this paper, a sliding window based ensemble method is proposed to detect outliers in a streaming fashion. The proposed method uses a combination of clustering algorithms to construct subgroups (clusters) representing different data structures. These structures are later used in a one-class classification algorithm to identfy the outliers. Thus, if a pattern does not belong to any of the common structures or clusters, it is an outlier. Further, based on the rare pattern classification, machine failures could be predicted in advance.

Image registration is the process of overlaying two or more multi sensor or multi temporal or multi resolution images of the same scene. Image registration is very crucial preprocessing step in any remote sensing image processing application. This paper presents a brief review of evolution of different image registration methodologies with challenges involved in it. Image registration basically aligns the two images (reference and sensed image) geometrically. In the first part of this paper, classical methods will be discussed which are mainly based on area based (Intensity based) approach and feature based approach. Ample amount of work has already been done in both methods in manual and automatic manner. In the second part all recent approaches will be discussed and compared with results, on the basis of registration accuracy, number of correct correspondence. In recent studies Convolutional Neural Network with deep learning algorithms are used in combination with traditional methods to boost the overall performance.

Nowadays, standard intake of healthy food is necessary for keeping a balanced diet to avoid obesity in the human body. In this paper, we present a novel system based on machine learning that automatically performs accurate classification of food images and estimates food attributes. This paper proposes a deep learning model consisting of a convolutional neural network that classifies food into specific categories in the training part of the prototype system. The main purpose of the proposed method is to improve the accuracy of the pre-training model. The paper designs a prototype system based on the client server model. The client sends an image detection request and processes it on the server side. The prototype system is designed with three main software components, including a pre-trained CNN model training module for classification purposes, a text data training module for attribute estimation models, and a server-side module. We experimented with a variety of food categories, each containing thousands of images, and through machine learning training to achieve higher classification accuracy.

This paper introduces a new research area called Explainable Robotics, which studies explainability in the context of human-robot interactions. The focus is on developing novel computational models, methods and algorithms for generating explanations that allow robots to operate at different levels of autonomy and communicate with humans in a trustworthy and human-friendly way. Individuals may need explanations during human-robot interactions for different reasons, which depend heavily on the context and human users involved. Therefore, the research challenge is identifying what needs to be explained at each level of autonomy and how these issues should be explained to different individuals. The paper presents the case for Explainable Robotics using a scenario involving the provision of medical health care to elderly patients with dementia with the help of technology. The paper highlights the main research challenges of Explainable Robotics. The first challenge is the need for new algorithms for generating explanations that use past experiences, analogies and real-time data to adapt to particular audiences and purposes. The second research challenge is developing novel computational models of situational and learned trust and new algorithms for the real-time sensing of trust. Finally, more research is needed to understand whether trust can be used as a control variable in Explainable Robotics.

This paper presents the use of a class of Deep Neural Networks for recognizing indoor home scenes so as to aid Intelligent Assistive Systems (IAS) in performing indoor services to assist elderly or infirm people. Identifying exact indoor location is important so that objects associated with particular tasks can be located speedily and efficiently irrespective of position or orientation. In this way, IAS developed for providing services may become more efficient in accomplishing designated tasks satisfactorily. There are many Convolutional Neural Networks (CNNs) which have been developed for outdoor scene classification and, also, for interior (not necessarily indoor home) scene classification. However, to date, there are no CNNs which are trained, validated and tested on indoor home scene datasets as there appears to be an absence of sufficiently large databases of home scenes. Nonetheless, it is important to train systems which are meant to operate within home environments with the correct relevant data. To counteract this problem, it is proposed that a different type of network is used, which is not very deep (i.e., a network which does not have too many layers) but which can attain sufficiently high classification accuracy using smaller training datasets. A type of neural network likely to help achieve this is a Capsule Neural Network (CapsNet). In this paper, 20,000 indoor home scenes were used for training the CapsNet, and 5000 images were used for testing it. The validation accuracy achieved is 71% and testing accuracy achieved is 70%.

This paper focuses on enhancing videos taken from a moving camera capable of capturing videos at any speed and in multiple directions. The second objective is to locate humans in the video to be able to conduct surveillance in case of disasters, without risking human lives. Most reported work for disaster management deals with satellite images. Indoor disaster management using image processing has been less explored. This work addresses applications involving disasters like a building on fire, wherein it will be useful to send a camera in to scan the rooms and locate people if present. This allows rescue teams/robots to make informed decisions and act accordingly. In this paper, the working of a 3600 camera is emulated using multiple ordinary cameras and the location of people is estimated using image processing techniques without Internet connectivity. The challenge is to train the model with a small dataset and accurately determine the location without any hardware setup (like sensors). Combination of Blind Deconvolution followed by Lucy-Richardson method gave optimal results. The frames were enhanced by de-noising using median filter, sharpening using unsharp mask and contrast enhancement using histogram equalization in order to make the object detection more accurate. Humans in the video were identified using either body or facial detection. Their locations were estimated with respect to a reference point and were quantified by predicting distance, angle and direction. In order to increase the accuracy, a fine-tuning mechanism has been proposed for location estimation using a non-linear curve, which gave acceptable accuracy within a range of 0.01m.

Epigraphy refers to the study of Epigraphs, or inscriptions. Estampage or stamping, is a term that is commonly used in epigraphy in order to get an exact copy of an inscription. An estampage is typically derived by applying wet paper onto the rock face, over which any ink material is wiped. This process also results in the appearance of unwanted noise which arises due to the texture of the rock, human errors or a variety of different factors. In this paper, we propose CCD: Connected Component Denoising, a novel approach for denoising and character segmentation via connected component analysis and multi scale template matching. A dynamic threshold is obtained via histogram density projections to effectively remove noisy areas. A comparative study is then performed between the different approaches (MSTM, CC, Hybrid). Inspection of the results of the hybrid approach by the HVS(Human Visual System) shows that it outperforms all the previous techniques.

Synthetic aperture radar (SAR) which acquires planar images at any time of the day and under any weather conditions is an important surveillance sensor in modern hi-tech war. For airborne SAR the image formation algorithms are ideally analyzed with the presumption that the aircraft moves in a straight line and at constant speed. However, the aircraft is affected by kinds of factors such as airstreams and its own instability, so the motion error between real aviation and ideal aviation exists, which can seriously blur the SAR image. So the motion compensation must be executed. In this paper a novel method for resolving turbulent aircraft motion is proposed based on SAR raw data. This method is very valid to conquer the turbulent aircraft motion, also the SAR image can be focused. The effectiveness of the proposed method is demonstrated by measured data imaging experiment. Meanwhile the method introduced in this paper causes only low computational load.

This paper mainly deals with Bio Inspired which means biological structures. Bio Inspired easily handle failure which determines the real-world complicated engineering issues. It continually finds the optimum resolution to unravel its drawback maintaining good balance among its parts. Widen the range and growth of Bio Inspired Algorithms analyze new fields of application. A disaster is a sudden, disastrous event that seriously disrupts the functioning of a community or society and causes human, material, and economic or environmental losses that exceed the community’s or society’s ability to cope using its own resources. This paper presents comparative study between ant colony optimization algorithm and fireflies algorithm to find shortest route during disaster. ACO is inspired by behavior of ants. The specialty of ACO algorithm is to find finest route with respect to each road’s length and damage degree. ACO provides reliable paths and also the decision is based on pheromone level released by ants. One of the recent developed swarm intelligence algorithm is firefly algorithm introduced by Yang. This algorithm inspired by behavior of insects. It works based on three rule: They are unisex, attractiveness and brightness. So these two algorithms helps in finding the shortest path between origin point and destination point.