Help Of Deep Neural Network Research Articles

Unsupervised random quantum networks for PDEs

Classical Physics-informed neural networks (PINNs) approximate solutions to PDEs with the help of deep neural networks trained to satisfy the differential operator and the relevant boundary conditions. We revisit this idea in the quantum computing realm, using parameterised random quantum circuits as trial solutions. We further adapt recent PINN-based techniques to our quantum setting, in particular Gaussian smoothing. Our analysis concentrates on the Poisson, the Heat and the Hamilton–Jacobi–Bellman equations, which are ubiquitous in most areas of science. On the theoretical side, we develop a complexity analysis of this approach, and show numerically that random quantum networks can outperform more traditional quantum networks as well as random classical networks.

Research on Physical Education Management System in Higher Education Institutions in the Context of Deep Learning

Abstract This paper focuses on the management of physical movement sports disciplines and combines relevant algorithms to design an artificial intelligence-based physical education management system framework. In terms of related algorithms, the first is for the research of character recognition algorithms. Considering the intra-class differences between individual movements, the LR-based feature extraction algorithm is proposed. Secondly, for the research of the action evaluation algorithm, the statistical value, DTW parameter action time series difference and correlation coefficient based on the statistical value and DTW parameter are used as the features, combined with SVM-based feature extraction algorithm for perceiving the action differences. Finally, with the help of deep neural networks for movement analysis, we can improve students’ movement techniques during training. In the course schedule supported by the system, the proportion of moderate-intensity exercise time is high, basically, more than 40%, which meets the WHO requirements for high-intensity exercise time in physical education classes. The post-course RPE feedback from students was spread out between 12 and 15, and the highest post-course RPE value was 14.82<unk>1.38, which was evaluated within a reasonable range of intervals. The teaching management system in this paper can meet the requirements of physical training and teaching, and provide quantifiable basis and management tools for the informatization and precise teaching of physical movement subjects.

Technology of using deep learning algorithms

Deep learning is a branch of machine learning (machine learning-ML). Deep learning methods utilize high-level model abstraction of nonlinear transformations in large databases. In other areas, the implementation of deep learning architectures has contributed significantly to the development of artificial intelligence. This paper presents recent research on newly applied deep learning algorithms. Convolutional Neural Networks are used in deep learning. Database Management System PostgreSQL object-relational database. The implementation resulted in achieving the set goals and objectives. The method of analyzing the input data is described, the differences between machine learning and deep learning are explained, and an example of classifying an image representing a sign language image using logistic regression, one of the deep learning algorithms, is presented. Deep neural networks can work with the full set of available data better than alternative approaches. During the learning process, the neural network itself determines which features in the data are important and which are not. Artificial neural networks can predict symptoms that humans cannot. Thus, with the help of deep neural networks, we can solve problems that traditional machine learning algorithms cannot perform.

SRVIO: Super Robust Visual Inertial Odometry for Dynamic Environments and Challenging Loop-Closure Conditions

There has been extensive research on visual localization and odometry for autonomous robots and virtual reality during the past decades. Traditionally, this problem has been solved with the help of expensive sensors, such as light detection and ranging (LiDAR). Nowadays, the focus of the leading research in this field is on robust localization using more economic sensors, such as cameras and inertial measurement units. Consequently, geometric visual localization methods have become more accurate over time. However, these methods still suffer from significant loss and divergence in challenging environments, such as a room full of moving people. Scientists started using deep neural networks (DNNs) to mitigate this problem. The main idea behind using DNNs is to better understand challenging aspects of the data and overcome complex conditions such as the movement of a dynamic object in front of the camera that covers the full view of the camera, extreme lighting conditions, and the high speed of the camera. Prior end-to-end DNN methods did overcome some of these challenges. However, no general and robust framework is available to overcome all challenges together. In this article, we have combined geometric and DNN-based methods to have the generality and speed of geometric SLAM frameworks and overcome most of these challenging conditions with the help of DNNs and deliver the most robust framework so far. To do so, we have designed a framework based on VINS-Mono and shown that it can achieve state-of-the-art results on TUM-Dynamic, TUM-VI, ADVIO, and EuRoC datasets compared to geometric and end-to-end DNN-based simultaneous localization and mappings. Our proposed framework can also achieve outstanding results on extreme simulated cases resembling the aforementioned challenges.

Implementation of Simple and Efficient Picture Caption Generator

Image captioning or picture captioning has become one of the most widely used technologies in applications that generate and provide captions for specific photographs. All these things are done with the help of deep neural networks. It identifies the specific objects in an image and their attributes and relationships. The purpose of this research is to find different things in a photograph, figure out their relationships, and write captions. The proposed system is implemented on dataset Flickr8k along with python. The input images are pre-processed and then features from images are extracted using CNN. To translate the features and objects extracted by CNN to a natural sentence in English LSTM is utilized in the implementation. Different types of images are tested with the proposed system. The results are presented with the generated image captions. The results presented shows the accuracy of the system. The presented method has potentials for such applications where image captioning is essential.

Human Recognition Using Deep Neural Networks and Spatial Patterns of SSVEP Signals

Brain biometrics have received increasing attention from the scientific community due to their unique properties compared to traditional biometric methods. Many studies have shown that EEG features are distinct across individuals. In this study, we propose a novel approach by considering spatial patterns of the brain's responses due to visual stimulation at specific frequencies. More specifically, we propose, for the identification of the individuals, to combine common spatial patterns with specialized deep-learning neural networks. The adoption of common spatial patterns gives us the ability to design personalized spatial filters. In addition, with the help of deep neural networks, the spatial patterns are mapped into new (deep) representations where the discrimination between individuals is performed with a high correct recognition rate. We conducted a comprehensive comparison between the performance of the proposed method and several classical methods on two steady-state visual evoked potential datasets consisting of thirty-five and eleven subjects, respectively. Furthermore, our analysis includes a large number of flickering frequencies in the steady-state visual evoked potential experiment. Experiments on these two steady-state visual evoked potential datasets showed the usefulness of our approach in terms of person identification and usability. The proposed method achieved an averaged correct recognition rate of 99% over a large number of frequencies for the visual stimulus.

SEGMENTATION OF ELECTRICAL SUBSTATIONS USING DEEP CONVOLUTIONAL NEURAL NETWORK

Abstract. The location of electrical substations is one of the factors affecting the improvement of electrical energy distribution, as well as the management and control of this energy source. Less cost and manpower will be spent through automating the process of detection and segmentation of these features with the help of deep neural networks and the potential of existing high spatial resolution satellite images. In this study, a deep encoder-decoder neural network was used. This network is one of the most updated deep learning methods in image processing and segmentation. This network has been trained in three RGB bands with the help of high-resolution satellite images (∼1m) and eventually segmented the areas related to electrical substations with relatively high accuracy. As the results of this convolutional neural network, the IOU and Precision parameters were obtained, and their values were 88.2 and 93.7%, respectively, indicating the efficiency of the proposed deep learning method in the segmentation of existing satellite images.

Machine Learning Assisted Ultra-Wideband Fiber-Optics Mode Selective Coupler Design

We report a machine learning assisted design strategy for ultra-wideband fiber-optics mode selective coupler (MSC). With the help of deep neural network (DNN), we are able to efficiently obtain the complex mapping relationship between structure parameters of the bridge fiber and effective refractive index (RI) of LP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">01</sub> mode over the S+C+L band. For high-order modes (HOMs) arising in the arbitrarily chosen few-mode fiber (FMF), the trained DNN can rapidly optimize the bridge fiber parameters. After optimization, the ultra-wideband phase matching between the effective RI of LP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">01</sub> mode arising in the bridge fiber and the effective RI of specific mode arising in the FMF over the S+C+L band can be realized. As an example, ultra-wideband fiber-optics 4-LP MSC with an average mode conversion efficiency (MCE) of more than 93% and an average mode extinction ratio (MER) of more than 17 dB can be secured. In comparison with the traditional genetic algorithm (GA), the DNN-assisted fiber-optics MSC design can significantly reduce the optimization time from 4 hours to 45 seconds, together with the maintenance of excellent ultra-wideband performance. Meanwhile, the ultra-wideband MSC for orbital angular momentum (OAM) mode conversion is demonstrated. Both OAM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">+1</sub> and OAM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">+2</sub> modes with an average mode purity of more than 97% and an average MCE of more than 89% over the S+C+L band can be secured. We envision the DNN-assisted fiber-optics device enables potential applications for future mode division multiplexing and wavelength division multiplexing hybrid optical transmission system.

Multi-Sacle Dynamic Coding Improved Spiking Actor Network for Reinforcement Learning

With the help of deep neural networks (DNNs), deep reinforcement learning (DRL) has achieved great success on many complex tasks, from games to robotic control. Compared to DNNs with partial brain-inspired structures and functions, spiking neural networks (SNNs) consider more biological features, including spiking neurons with complex dynamics and learning paradigms with biologically plausible plasticity principles. Inspired by the efficient computation of cell assembly in the biological brain, whereby memory-based coding is much more complex than readout, we propose a multiscale dynamic coding improved spiking actor network (MDC-SAN) for reinforcement learning to achieve effective decision-making. The population coding at the network scale is integrated with the dynamic neurons coding (containing 2nd-order neuronal dynamics) at the neuron scale towards a powerful spatial-temporal state representation. Extensive experimental results show that our MDC-SAN performs better than its counterpart deep actor network (based on DNNs) on four continuous control tasks from OpenAI gym. We think this is a significant attempt to improve SNNs from the perspective of efficient coding towards effective decision-making, just like that in biological networks.

Pile-up mitigation using attention

Particle production from secondary proton-proton collisions, commonly referred to as pile-up, impair the sensitivity of both new physics searches and precision measurements at large hadron collider (LHC) experiments. We propose a novel algorithm, Puma, for modeling pile-up with the help of deep neural networks based on sparse transformers. These attention mechanisms were developed for natural language processing but have become popular in other applications. In a realistic detector simulation, our method outperforms classical benchmark algorithms for pile-up mitigation in key observables. It provides a perspective for mitigating the effects of pile-up in the high luminosity era of the LHC, where up to 200 proton-proton collisions are expected to occur simultaneously.

Day-Ahead Optimal Dispatch Strategy for Active Distribution Network Based on Improved Deep Reinforcement Learning

To effectively deal with the challenge of optimal dispatch caused by uncertainties such as renewable energy in active distribution network, a day-ahead optimal dispatch strategy for active distribution network based on improved deep reinforcement learning is proposed. First, the day-ahead optimal dispatch problem for the active distribution network is modeled as a multistage stochastic programming model. Multistage models attempt to capture the dynamics of unfolding uncertainties over time, so that the interaction between decision making and uncertainty unfolding is represented more accurately and realistically, and the decision can be adjusted dynamically and adaptively. Second, with the help of deep neural network to approximate the power flow model of the active distribution network in a data-driven manner, an improved deep reinforcement learning algorithm is proposed to solve the corresponding multistage stochastic programming problem. The proposed improved deep reinforcement learning algorithm effectively solves the problem that trial-and-error information acquisition methods are difficult to apply to high-cost power systems. Finally, the effectiveness of the proposed day-ahead optimization dispatch strategy for active distribution network based on improved deep reinforcement learning is verified by a modified IEEE33 case.

Prediction of drilling induced delamination and circularity deviation in GFRP nanocomposites using deep neural network

Drilling of Glass Fiber Reinforced Polymer (GFRP) nanocomposites is most prevailing topic to understand the composite behaviour under different cutting conditions. The present study is mainly focused on prediction of drilling output responses such as delamination factor and circularity error randomly with the help of deep neural network (DNN) model. L9 orthogonal array is used for experimentation. Drilling operation is performed on 0.3 wt% multi-walled carbon nano tubes reinforced GFRPs with solid carbide, TiCN and TiAlN coated (6 mm- diameter) twist drills. Based on experimental results, two different deep neural network models are prepared with single and double hidden layers by varying node numbers such as 8, 16, 32, 64, and 128. Thrust force, Acoustic Emission RMS voltage, and drill type (coating) are considered as input to the neural network and delamination factor at exit, circularity error are treated as predicted output responses for the given network model. The revealed predicted results recommended that two hidden layers with 32 nodes network model give the lowest absolute error of 0.08% and 3.13% in delamination factor and circularity errors respectively. Similarly, the highest absolute error is identified as 4.19% in delamination factor and 13.14% in circularity error by single hidden layer with 128 nodes. Therefore, it is urged that DNN is the most suitable modelling technique for prediction of drilling responses on GFRP nano composites.

Real Time Facial Gesture Recognition using Deep Learning

The human faces have vibrant frequency of characteristics, which makes it difficult to analyze the facial expression. Automated real time emotions recognition with the help of facial expressions is a work in computer vision. This environment is an important and interesting tool between the humans and computers. In this investigation an environment is created which is capable of analyzing the person’s emotions using the real time facial gestures with the help of Deep Neural Network. It can detect the facial expression from any image either real or animated after facial extraction (muscle position, eye expression and lips position). This system is setup to classify images of human faces into seven discrete emotion categories using Convolutional Neural Networks (CNNs). This type of environment is important for social interaction.

Video Super-Resolution Based on Generative Adversarial Network and Edge Enhancement

With the help of deep neural networks, video super-resolution (VSR) has made a huge breakthrough. However, these deep learning-based methods are rarely used in specific situations. In addition, training sets may not be suitable because many methods only assume that under ideal circumstances, low-resolution (LR) datasets are downgraded from high-resolution (HR) datasets in a fixed manner. In this paper, we proposed a model based on Generative Adversarial Network (GAN) and edge enhancement to perform super-resolution (SR) reconstruction for LR and blur videos, such as closed-circuit television (CCTV). The adversarial loss allows discriminators to be trained to distinguish between SR frames and ground truth (GT) frames, which is helpful to produce realistic and highly detailed results. The edge enhancement function uses the Laplacian edge module to perform edge enhancement on the intermediate result, which helps further improve the final results. In addition, we add the perceptual loss to the loss function to obtain a higher visual experience. At the same time, we also tried training network on different datasets. A large number of experiments show that our method has advantages in the Vid4 dataset and other LR videos.

Voice Pathology Detection Based on Deep Neural Network Approach

The advancement of technology offers solution to the complex problems faced by the society and brings the wellbeing of the individuals. Smart healthcare is prominent nowadays for diagnosis, treatment and constant monitoring which reduces visitation of hospital, transport cost and waiting time. Voice pathology is a decease which affects the person vocal cord so that one who facing difficult in speech. If the decease not identified in time, it leads to permanent loss of voice for an individual. Traditionally, the decease is identified through oral examination or manual procedures. Due to the advent of smart phone, one can record the voice and send it to the cloud server for processing. Our system classifies the voice data and provides the decision to the user. This greatly reduces the transport cost, and waiting time for oral examination at medical center. The mobile phone recorded the patient voice data and it will be stored into the cloud. The voice data is synthesized to signals and with the help of deep Neural network the voice pathology can be identified. The system has been tested with the data set and the test result shows promising.

COV-VGX: An automated COVID-19 detection system using X-ray images and transfer learning

Coronavirus (COVID-19) has been one of the most dangerous and acute deadly diseases across the world recently. Researchers are trying to develop automated and feasible COVID-19 detection systems with the help of deep neural networks, machine learning techniques, etc. In this paper, a deep learning-based COVID-19 detection system called COV-VGX is proposed that contributes to detecting coronavirus disease automatically using chest X-ray images. The system introduces two types of classifiers, namely, a multiclass classifier that automatically predicts coronavirus, pneumonia, and normal classes and a binary classifier that predicts coronavirus and pneumonia classes. Using transfer learning, a deep CNN model is proposed to extract distinct and high-level features from X-ray images in collaboration with the pretrained model VGG-16. Despite the limitation of the COVID-19 dataset, the model is evaluated with sufficient COVID-19 images. Extensive experiments for multiclass classifier have achieved 98.91% accuracy, 97.31% precision, 99.50% recall, 98.39% F1-score, while 99.37% accuracy, 98.76% precision, 100% recall, 99.38% F1-score for binary classifier. The proposed system can contribute a lot in diagnosing COVID-19 effectively in the medical field.

SIES: A Novel Implementation of Spiking Convolutional Neural Network Inference Engine on Field-Programmable Gate Array

Neuromorphic computing is considered to be the future of machine learning, and it provides a new way of cognitive computing. Inspired by the excellent performance of spiking neural networks (SNNs) on the fields of low-power consumption and parallel computing, many groups tried to simulate the SNN with the hardware platform. However, the efficiency of training SNNs with neuromorphic algorithms is not ideal enough. Facing this, Michael et al. proposed a method which can solve the problem with the help of DNN (deep neural network). With this method, we can easily convert a well-trained DNN into an SCNN (spiking convolutional neural network). So far, there is a little of work focusing on the hardware accelerating of SCNN. The motivation of this paper is to design an SNN processor to accelerate SNN inference for SNNs obtained by this DNN-to-SNN method. We propose SIES (Spiking Neural Network Inference Engine for SCNN Accelerating). It uses a systolic array to accomplish the task of membrane potential increments computation. It integrates an optional hardware module of max-pooling to reduce additional data moving between the host and the SIES. We also design a hardware data setup mechanism for the convolutional layer on the SIES with which we can minimize the time of input spikes preparing. We implement the SIES on FPGA XCVU440. The number of neurons it supports is up to 4 000 while the synapses are 256 000. The SIES can run with the working frequency of 200 MHz, and its peak performance is 1.562 5 TOPS.

Mining Mobile Intelligence for Wireless Systems: A Deep Neural Network Approach

Wireless big data contain valuable information on users' behaviors and preferences, which can drive the design and optimization for wireless systems. The fundamental issue is how to mine mobile intelligence and further incorporate them into wireless systems. To this end, this article discusses two challenges on big data based wireless system design and optimization, and proposes a unified framework to tackle them with the help of Deep Neural Networks (DNNs) and online learning techniques. In particular, we propose a DNN architecture by incorporating an embedding layer to project different types of raw data to a latent space and utilize a regression or classification function to predict the mobile access pattern. It outperforms the best traditional machine learning algorithm (76% vs. 63%) significantly. Moreover, combining the proposed DNN architecture with online learning techniques, we show two cases on how to apply the mobile intelligence for wireless video applications, including video adaption and video pre-fetching. In the former case, we utilize the proposed DNN method to predict the dynamics of user count within the coverage of base stations, and adaptively adjust the bitrate for video streaming to improve the video watching experience. In the latter one, we utilize the proposed method to predict the user trajectory, i.e., the associated base stations, and conduct content prefetching to reduce the access latency. Evaluating the performance with a real wireless dataset, we show that the perceived video QoE and cache hit ratio are greatly improved (0.7db and 25% respectively).

Bio-medical analysis of breast cancer risk detection based on deep neural network

Breast tumour remains a most important reason of cancer fatality among women globally and most of them pass away due to delayed diagnosis. But premature recognition and anticipation can significantly diminish the chances of death. Risk detection of breast cancer is one of the major research areas in bioinformatics. Various experiments have been conceded to examine the fundamental grounds of breast tumour. Alternatively, it has already been verified that early diagnosis of tumour can give the longer survival chance to a patient. This paper aims at finding an efficient set of features for breast tumour prediction using deep learning approaches called restricted Boltzmann machine (RBM). The proposed framework diagnoses and analyses breast tumour patient's data with the help of deep neural network (DNN) classifier using the Wisconsin dataset of UCI machine learning repository and, thereafter assesses their performance in terms of measures like accuracy, precision, recall, F-measure, etc.

Discovering Hate Sentiment within Twitter Data through Aspect-Based Sentiment Analysis

Aspect-based sentiment analysis is a vital issue in fine-grained sentiment evaluation, which intends to provide an automatic prediction of the sentiment polarity, given a particular aspect in its context. This paper presents an aspect-based sentiment analysis to find hate sentiment inside twitter data. Word embeddings have had prevalent utilisation in Natural Language Processing (NLP) applications because their vector representations have the ability to capture useful linguistic relationships and semantic properties between words with the help of deep neural networks. Word embeddings have often been used in machine learning models as feature input, which allows for the contextualisation of raw text data in machine learning techniques. The model has the ability to represent the relationship between the word embedding features and the aspects as feature representation within the suggested model. To assess the efficacy of the proposed method, extensive experiments were performed on the dataset of the researcher, as well as on widely utilised datasets. It was demonstrated by the experimental results that the proposed method was able to obtain impressive results among the three datasets.