Multi-Ontology Mapping Generative Adversarial Network in Internet of Things for Ontology Alignment
نگاشت چند هستی شناسی شبکه متخاصم مولد در اینترنت اشیا برای تراز هستی شناسی-2022
On the Semantic web, ontologies are thought to be the remedy to data heterogeneity, and correlating ontologies is a highly effective technique. Although the use of representation learning approaches to a variety of applications has showed significant promise, they have had little effect on the issue of ontology matching and classification. In order to establish alignments between two ontologies, this research presents the Multi-Ontology Mapping Generative Adversarial Network in Internet of Things (MOMGANI). For the instance of ontology mapping, we suggest using a two-system representation learning network consisting of a Generator and Discriminator. The Generator applies a probabilistic softmax classifier to the different Name, Label, Comments, Properties, Instance descriptions, concept characteristics, and the neighbourhood concepts for each of the ontologys properties. In order to support the assertions that the Generator has generated, the Discriminator network employs a novel Bidirectional Long Short-Term Memory (Bi-LSTM network) with an Ontology Attention mechanism enhanced by the concept’s descriptions. As a result, both systems are in a feedback mechanism where they can learn from one another. The system will produce a set of triples that list all the associated concepts from various ontologies as its final product. Domain experts will review these triples outside of the band to ensure that only true concepts and triples are chosen for the alignment. In comparison to using the ontologies separately, the aligned ontology enables extended querying and inference across related ontologies and domains. Considering metrics like recall, precision, and F-measure, the experimental evaluation was performed utilizing the datasets for classes alignment, property alignment, and instances alignment. The proposed architecture provides a recall, precision, and F-measure of 0.92, 0.99, and 0.83 respectively which reveals that this model outperforms the traditional methods.
Keywords: Generative adversarial network | Ontology alignment | IoT and OntoGenerator and OntoLSTM
Person-identification using familiar-name auditory evoked potentials from frontal EEG electrodes
شناسایی فرد با استفاده از پتانسیل نام-آشنا شنوایی الکترودهای EEG جلو برانگیخته-2021
Electroencephalograph (EEG) based biometric identification has recently gained increased attention of re- searchers. However, state-of-the-art EEG-based biometric identification techniques use large number of EEG electrodes, which poses user inconvenience and consumes longer preparation time for practical applications. This work proposes a novel EEG-based biometric identification technique using auditory evoked potentials (AEPs) acquired from two EEG electrodes. The proposed method employs single-trial familiar-name AEPs extracted from the frontal electrodes Fp1 and F7, which facilitates faster and user-convenient data acquisition. The EEG signals recorded from twenty healthy individuals during four experiment trials are used in this study. Different com- binations of well-known neural network architectures are used for feature extraction and classification. The cascaded combinations of 1D-convolutional neural networks (1D-CNN) with long short-term memory (LSTM) and with gated recurrent unit (GRU) networks gave the person identification accuracies above 99 %. 1D-convolutional, LSTM network achieves the highest person identification accuracy of 99.53 % and a half total error rate (HTER) of 0.24 % using AEP signals from the two frontal electrodes. With the AEP signals from the single electrode Fp1, the same network achieves a person identification accuracy of 96.93 %. The use of familiar-name AEPs from frontal EEG electrodes that facilitates user convenient data acquisition with shorter preparation time is the novelty of this work.
Keywords: Auditory evoked potential | Biometrics | Deep learning | Electroencephalogram | Familiar-name | Person identification
Vision-assisted recognition of stereotype behaviors for early diagnosis of Autism Spectrum Disorders
تشخیص رفتارهای کلیشه ای برای تشخیص زودهنگام اختلالات طیف اوتیسم با کمک بینایی ماشین-2021
Medical diagnosis supported by computer-assisted technologies is getting more popularity and acceptance among medical society. In this paper, we propose a non-intrusive vision-assisted method based on human action recognition to facilitate the diagnosis of Autism Spectrum Disorder (ASD). We collected a novel and comprehensive video dataset f the most distinctive Stereotype actions of this disorder with the assistance of professional clinicians. Several frameworks as a function of different input modalities were developed and used to produce extensive baseline results. Various local descriptors, which are commonly used within the Bag-of-Visual-Words approach, were tested with Multi-layer Perceptron (MLP), Gaussian Naive Bayes (GNB), and Support Vector Machines (SVM) classifiers for recognizing ASD associated behaviors. Additionally, we developed a framework that first receives articulated pose-based skeleton sequences as input and follows an LSTM network to learn the temporal evolution of the poses. Finally, obtained results were compared with two fine-tuned deep neural networks: ConvLSTM and 3DCNN. The results revealed that the Histogram of Optical Flow (HOF) descriptor achieves the best results when used with MLP classifier. The promising baseline results also confirmed that an action-recognition-based system can be potentially used to assist clinicians to provide a reliable, accurate, and timely diagnosis of ASD disorder.© 2021 Elsevier B.V. All rights reserved.
Keywords: Action recognition | Autism Spectrum Disorder | Patient monitoring | Bag-of-visual-words | Convolutional neural networks
Optokinetic response for mobile device biometric liveness assessment
پاسخ اپتوکینتیک برای ارزیابی زنده بودن بیومتریک دستگاه تلفن همراه-2021
As a practical pursuit of quantiﬁed uniqueness, biometrics explores the parameters that make us who we are and provides the tools we need to secure the integrity of that identity. In our culture of constant connectivity, an in- creasing reliance on biometrically secured mobile devices is transforming them into a target for bad actors. While no system will ever prevent all forms of intrusion, even state of the art biometric methods remain vulnerable to spoof attacks. As these attacks become more sophisticated, liveness based attack detection methods provide a po- tential deterrent. We present a novel optokinetc nystagmus (OKN) based liveness assessment system for mobile applications which leverages phase-locked temporal features of a unique reﬂexive behavioral response. In this paper we provide proof of concept for eliciting, collecting and extracting the OKN response motion signature on a mobile device. Results of our most successful experimental machine learning classiﬁer are reported for a multi-layer LSTM based model demonstrating a 98.4% single stimulus detection performance for simulated video based attacks.© 2021 Elsevier B.V. All rights reserved.
Keywords: Biometrics | Eye movement | Ocular biometrics | Ocular kinetics | Digital identity | Mobile device security | Liveness | Behavioral biometrics
An integrated approach using CNN-RNN-LSTM for classification of fruit images
یک رویکرد یکپارچه با استفاده از CNN-RNN-LSTM برای طبقه بندی تصاویر میوه-2021
With the advancement in technology, Computer and machine vision system is getting involved in the agriculture sector for the last few years. Deep Learning is a recent advancement in the Artificial Intelligence field. In the present era, many researchers have used deep learning applications for the classification of images, and is found to be one of the emerging areas in computer vision. In the classification of fruit images, the main goal is to improve the accuracy of the classification system. The accuracy of the classifier depends on various factors like the nature of acquired images, the number of features, types of features, selection of optimal features from extracted features, and type of classifiers used. In the pro- posed article, integration of CNN, RNN, and LSTM for the classification of fruit images are defined. In this approach, CNN and RNN are employed for the development of discriminative characteristics and sequential-labels respectively. LSTM presents an explanation by integrating a memory cell to encode learning at each interval of classification. Key parameters: accuracy, F-measure, sensitivity, and specificity are applied to assess the achievement of the proposed scheme. From empirical results, it has been declared that the offered classification method provides efficient results.© 2021 Elsevier Ltd. All rights reserved. Selection and peer-review under responsibility of the scientific committee of the 1st International Conference on Computations in Materials and Applied Engineering – 2021.
Keywords: CNN | RNN | LSTM | Integrated Approach | Fruit classification
A multimodal-Siamese Neural Network (mSNN) for person verification using signatures and EEG
شبکه عصبی چند حالته سیامی (mSNN) برای تأیید شخص با استفاده از امضا و EEG-2021
Signatures have long been considered to be one of the most accepted and practical means of user verification, despite being vulnerable to skilled forgers. In contrast, EEG signals have more recently been shown to be more difficult to replicate, and to provide better biometric information in response to known a stimulus. In this paper, we propose combining these two biometric traits using a multimodal Siamese Neural Network (mSNN) for improved user verification. The proposed mSNN network learns discriminative temporal and spatial features from the EEG signals using an EEG encoder and from the offline signatures using an image encoder. Features of the two encoders are fused into a common feature space for further processing. A Siamese network then employs a distance metric based on the similarity and dissimilarity of the input features to produce the verification results. The proposed model is evaluated on a dataset of 70 users, comprised of 1400 unique samples. The novel mSNN model achieves a 98.57% classification accuracy with a 99.29% True Positive Rate (TPR) and False Acceptance Rate (FAR) of 2.14%, outperforming the current state-of-the-art by 12.86% (in absolute terms). This proposed network architecture may also be applicable to the fusion of other neurological data sources to build robust biometric verification or diagnostic systems with limited data size.
Keywords: User verification | Multimodal | EEG | Siamese Neural Network | LSTM | CNN
AI-based Reference Ankle Joint Torque Trajectory Generation for Robotic Gait Assistance: First Steps
تولید مسیر حرکت گشتاور مفصل مچ پا مبتنی بر هوش مصنوعی برای کمک به راه رفتن رباتیک: اولین قدم ها-2020
Robotic-based gait rehabilitation and assistance have been growing to augment and to recover motor function in subjects with lower limb impairments. There is interest in developing user-oriented control strategies to provide personalized assistance. However, it is still needed to set the healthy user-oriented reference joint trajectories, namely, reference ankle joint torque, that would be desired under healthy conditions. Considering the potential of Artificial Intelligence (AI) algorithms to model nonlinear relationships of the walking motion, this study implements and compares two offline AI-based regression models (Multilayer Perceptron and Long-Short Term Memory-LSTM) to generate healthy reference ankle joint torques oriented to subjects with a body height ranging from 1.51 to 1.83 m, body mass from 52.0 to 83.7 kg and walking in a flat surface with a walking speed from 1.0 to 4.0 km/h. The best results were achieved for the LSTM, reaching a Goodness of Fit and a Normalized Root Mean Square Error of 79.6 % and 4.31 %, respectively. The findings showed that the implemented LSTM has the potential to be integrated into control architectures of robotic assistive devices to accurately estimate healthy useroriented reference ankle joint torque trajectories, which are needed in personalized and Assist-As-Needed conditions. Future challenges involve the exploration of other regression models and the reference torque prediction for remaining lower limb joints, considering a wider range of body masses, heights, walking speeds, and locomotion modes.
Keywords: Ankle Joint Torque Prediction | Artificial Intelligence | Control Strategies | Regression Models | Robotic Gait Rehabilitation
Looking in the Right Place for Anomalies: Explainable Ai Through Automatic Location Learning
جستجوی مکان مناسب برای ناهنجاری ها: هوش مصنوعی قابل توضیح از طریق یادگیری خودکار مکان-2020
Deep learning has now become the de facto approach to the recognition of anomalies in medical imaging. Their ’black box’ way of classifying medical images into anomaly labels poses problems for their acceptance, particularly with clinicians. Current explainable AI methods offer justifications through visualizations such as heat maps but cannot guarantee that the network is focusing on the relevant image region fully containing the anomaly. In this paper we develop an approach to explainable AI in which the anomaly is assured to be overlapping the expected location when present. This is made possible by automatically extracting location-specific labels from textual reports and learning the association of expected locations to labels using a hybrid combination of Bi-Directional Long Short-Term Memory Recurrent Neural Networks (Bi- LSTM) and DenseNet-121. Use of this expected location to bias the subsequent attention-guided inference network based on ResNet101 results in the isolation of the anomaly at the expected location when present. The method is evaluated on a large chest X-ray dataset.
Accelerating deep reinforcement learning model for game strategy
تسریع در مدل یادگیری تقویتی عمیق برای استراتژی بازی-2020
In recent years, deep reinforcement learning has achieved impressing accuracies in games compared with traditional methods. Prior schemes utilized Convolutional Neural Networks (CNNs) or Long Short- Term Memory networks (LSTMs) to improve the performances of the agents. In this paper, we consider the issue from a different perspective when the training and inference of deep reinforcement learning are required to be performed with limited computing resources. Mainly, we propose two efficient neu- ral network architectures of deep reinforcement learning: Light-Q-Network (LQN) and Binary-Q-Network (BQN). In LQN, The depth-wise separable CNNs are utilized in memory and computation saving. While, in BQN, the weights of convolutional layers are binary that help in shortening the training time and reduce memory consumption. We evaluate our approach on Atari 2600 domain and StarCraft II mini-games. The results demonstratethe efficiency of the proposed architectures. Though performances of agents in most games are still super-human, the proposed methods advance the agent from sub to super-human performance in particular games. Also, we empirically find that non-standard convolution and non-full- precision networks do not affect agent learning game strategy.
Keywords: Deep reinforcement learning | Convolutional neural network | Depthwise separable convolution | Binary weight network
A composite learning method for multi-ship collision avoidance based on reinforcement learning and inverse control
یک روش یادگیری ترکیبی برای جلوگیری از برخورد چند حرکتی مبتنی بر یادگیری تقویتی و کنترل معکوس-2020
Model-free reinforcement learning methods have potentials in ship collision avoidance under unknown environments. To defect the low efficiency problem of the model-free reinforcement learning, a composite learning method is proposed based on an asynchronous advantage actor-critic (A3C) algorithm, a long short-term memory neural network (LSTM) and Q-learning. The proposed method uses Q-learning for adaptive decisions between a LSTM inverse model-based controller and the model-free A3C policy. Multi-ship collision avoidance simulations are conducted to verify the effectiveness of the model-free A3C method, the proposed inverse model-based method and the composite learning method. The simulation results indicate that the proposed composite learning based ship collision avoidance method outperforms the A3C learning method and a traditional optimization-based method
Keywords: Ship collision avoidance | Asynchronous advantage actor-critic | Long short-term memory neural network | Inverse control