Search results for: deep Boltzmann machines

Authors: Thiago Spilborghs Bueno Meyer, Plinio Thomaz Aquino Junior

Abstract:

Through speech, which privileges the functional and interactive nature of the text, it is possible to ascertain the spatiotemporal circumstances, the conditions of production and reception of the discourse, the explicit purposes such as informing, explaining, convincing, etc. These conditions allow bringing the interaction between humans closer to the human-robot interaction, making it natural and sensitive to information. However, it is not enough to understand what is said; it is necessary to recognize emotions for the desired interaction. The validity of the use of neural networks for feature selection and emotion recognition was verified. For this purpose, it is proposed the use of neural networks and comparison of models, such as recurrent neural networks and deep neural networks, in order to carry out the classification of emotions through speech signals to verify the quality of recognition. It is expected to enable the implementation of robots in a domestic environment, such as the HERA robot from the RoboFEI@Home team, which focuses on autonomous service robots for the domestic environment. Tests were performed using only the Mel-Frequency Cepstral Coefficients, as well as tests with several characteristics of Delta-MFCC, spectral contrast, and the Mel spectrogram. To carry out the training, validation and testing of the neural networks, the eNTERFACE’05 database was used, which has 42 speakers from 14 different nationalities speaking the English language. The data from the chosen database are videos that, for use in neural networks, were converted into audios. It was found as a result, a classification of 51,969% of correct answers when using the deep neural network, when the use of the recurrent neural network was verified, with the classification with accuracy equal to 44.09%. The results are more accurate when only the Mel-Frequency Cepstral Coefficients are used for the classification, using the classifier with the deep neural network, and in only one case, it is possible to observe a greater accuracy by the recurrent neural network, which occurs in the use of various features and setting 73 for batch size and 100 training epochs.

Keywords: emotion recognition, speech, deep learning, human-robot interaction, neural networks

Procedia PDF Downloads 146

Authors: A. Degale Desta, Cheng Jian

Abstract:

Deep learning application in computer vision is rapidly advancing, giving it the ability to monitor the public and quickly identify potentially anomalous behaviour from crowd scenes. Therefore, the purpose of the current work is to improve the performance of safety of people in crowd events from panic behaviour through introducing the innovative idea of Aggregation of Ensembles (AOE), which makes use of the pre-trained ConvNets and a pool of classifiers to find anomalies in video data with packed scenes. According to the theory of algorithms that applied K-means, KNN, CNN, SVD, and Faster-CNN, YOLOv5 architectures learn different levels of semantic representation from crowd videos; the proposed approach leverages an ensemble of various fine-tuned convolutional neural networks (CNN), allowing for the extraction of enriched feature sets. In addition to the above algorithms, a long short-term memory neural network to forecast future feature values and a handmade feature that takes into consideration the peculiarities of the crowd to understand human behavior. On well-known datasets of panic situations, experiments are run to assess the effectiveness and precision of the suggested method. Results reveal that, compared to state-of-the-art methodologies, the system produces better and more promising results in terms of accuracy and processing speed.

Keywords: action recognition, computer vision, crowd detecting and tracking, deep learning

Procedia PDF Downloads 143

Authors: Majid Bagherinia

Abstract:

Lime and cement are frequently used as binders in the Deep Mixing Method (DMM) to improve soft clay soils. The most significant disadvantages of these materials are carbon dioxide emissions and the consumption of natural resources. In this study, three different biopolymers, guar gum, locust bean gum, and sodium alginate, were investigated for the improvement of soft clay using DMM. In the experimental study, the effects of the additive ratio and curing time on the Unconfined Compressive Strength (UCS) of stabilized specimens were investigated. According to the results, the UCS values of the specimens increased as the additive ratio and curing time increased. The most effective additive was sodium alginate, and the highest strength was obtained after 28 days.

Keywords: deep mixing method, soft clays, ground improvement, biopolymers, unconfined compressive strength

Procedia PDF Downloads 57

Authors: Judith Mwakalonge, Geophrey Mbatta, Saidi Siuhi, Gurcan Comert, Cuthbert Ruseruka

Abstract:

Transportation agencies worldwide continuously monitor their roads' conditions to minimize road maintenance costs and maintain public safety and rideability quality. Existing methods for carrying out road condition surveys involve manual observations of roads using standard survey forms done by qualified road condition surveyors or engineers either on foot or by vehicle. Automated road condition survey vehicles exist; however, they are very expensive since they require special vehicles equipped with sensors for data collection together with data processing and computing devices. The manual methods are expensive, time-consuming, infrequent, and can hardly provide real-time information for road conditions. This study contributes to this arena by utilizing built-in vehicle technologies, drones, and deep learning to automate road condition surveys while using low-cost technology. A single model is trained to capture flexible pavement distresses (Potholes, Rutting, Cracking, and raveling), thereby providing a more cost-effective and efficient road condition monitoring approach that can also provide real-time road conditions. Additionally, data fusion is employed to enhance the road condition assessment with data from vehicles and drones.

Keywords: road conditions, built-in vehicle technology, deep learning, drones

Procedia PDF Downloads 97

Authors: Hongmei Wang, Ziyun Xiang, Ying liu, Li Yu, Dongsheng Yue

Abstract:

Intro: The COVID-19 (Corona Virus Disease 2019) has greatly changed the global economic, political and financial ecology. The mutation of the coronavirus in the UK in December 2020 has brought new panic to the world. Deep learning was performed on Chest Computed tomography (CT) of COVID-19 and Influenza and describes their characteristics. The predominant features of COVID-19 pneumonia was ground-glass opacification, followed by consolidation. Lesion density: most lesions appear as ground-glass shadows, and some lesions coexist with solid lesions. Lesion distribution: the focus is mainly on the dorsal side of the periphery of the lung, with the lower lobe of the lungs as the focus, and it is often close to the pleura. Other features it has are grid-like shadows in ground glass lesions, thickening signs of diseased vessels, air bronchi signs and halo signs. The severe disease involves whole bilateral lungs, showing white lung signs, air bronchograms can be seen, and there can be a small amount of pleural effusion in the bilateral chest cavity. At the same time, this year's flu season could be near its peak after surging throughout the United States for months. Chest CT for Influenza infection is characterized by focal ground glass shadows in the lungs, with or without patchy consolidation, and bronchiole air bronchograms are visible in the concentration. There are patchy ground-glass shadows, consolidation, air bronchus signs, mosaic lung perfusion, etc. The lesions are mostly fused, which is prominent near the hilar and two lungs. Grid-like shadows and small patchy ground-glass shadows are visible. Deep neural networks have great potential in image analysis and diagnosis that traditional machine learning algorithms do not. Method: Aiming at the two major infectious diseases COVID-19 and influenza, which are currently circulating in the world, the chest CT of patients with two infectious diseases is classified and diagnosed using deep learning algorithms. The residual network is proposed to solve the problem of network degradation when there are too many hidden layers in a deep neural network (DNN). The proposed deep residual system (ResNet) is a milestone in the history of the Convolutional neural network (CNN) images, which solves the problem of difficult training of deep CNN models. Many visual tasks can get excellent results through fine-tuning ResNet. The pre-trained convolutional neural network ResNet is introduced as a feature extractor, eliminating the need to design complex models and time-consuming training. Fastai is based on Pytorch, packaging best practices for in-depth learning strategies, and finding the best way to handle diagnoses issues. Based on the one-cycle approach of the Fastai algorithm, the classification diagnosis of lung CT for two infectious diseases is realized, and a higher recognition rate is obtained. Results: A deep learning model was developed to efficiently identify the differences between COVID-19 and influenza using chest CT.

Keywords: COVID-19, Fastai, influenza, transfer network

Procedia PDF Downloads 126

Authors: Aminu Yakubu Umar

Abstract:

X-ray is the major contributor to the effective dose of both the patient and the personnel. Because of the radiological risks involved, it is usually recommended that dose to patient from X-ray be kept as low as reasonably achievable (ALARA) with adequate image quality. The implementation of quality assurance in diagnostic radiology can help greatly in achieving that, as it is a technique designed to reduce X-ray doses to patients undergoing radiological examination. In this study, quality control was carried out in six hospitals, which involved KVp test, evaluation of total filtration, test for constancy of radiation output, and check for mA linearity. Equipment used include KVp meter, Rad-check meter, aluminum sheets (0.1–1.0 mm) etc. The results of this study indicate that, the age of the X-ray machines in the hospitals ranges from 3-13 years, GHI and GH2 being the oldest and FMC being the newest. In the evaluation of total filtration, the HVL of the X-ray machines in the hospitals varied, ranging from 2.3-5.2 mm. The HVL was found to be highest in AHC (5.2 mm), while it was lowest in GH3 (2.3 mm). All HVL measurements were done at 80 KVp. The variation in voltage accuracy in the hospitals ranges from 0.3%-127.5%. It was only in GH1 that the % variation was below the allowed limit. The test for constancy of radiation output showed that, the coefficient of variation ranges from 0.005–0.550. In GH3, FMC and AHC, the coefficient of linearity were less than the allowed limit, while in GH1, GH2 and GH4 the coefficient of linearity had exceeded the allowed limit. As regard to mA linearity, FMC and AHC had their coefficients of linearity as 0.12 and 0.10 respectively, which were within the accepted limit, while GH1, GH3 and GH4 had their coefficients as 0.16, 0.69 and 0.98 respectively, which exceeded the allowed limit.

Keywords: radiation, X-ray output, quality control, half-value layer, mA linearity, KVp variation

Procedia PDF Downloads 599

Authors: Mohammad R. Jafari, Francois G. Bernardeau

Abstract:

With its arid climate, Qatar experiences low annual rainfall, intense storms, and high evaporation rates. However, the fast-paced rate of infrastructure development in the capital city of Doha has led to recurring instances of surface water flooding as well as rising groundwater levels. Public Work Authority (PWA/ASHGHAL) has implemented an approach to collect and discharge the flood water into a) positive gravity systems; b) Emergency Flooding Area (EFA) – Evaporation, Infiltration or Storage off-site using tankers; and c) Discharge to deep injection wells. As part of the flood prevention scheme, 21 deep injection wells have been constructed to discharge the collected surface and groundwater table in Doha city. These injection wells function as an alternative in localities that do not possess either positive gravity systems or downstream networks that can accommodate additional loads. These injection wells are 400-m deep and are constructed in a complex karstic subsurface condition with large cavities. The injection well system will discharge collected groundwater and storm surface runoff into the permeable Umm Er Radhuma Formation, which is an aquifer present throughout the Persian Gulf Region. The Umm Er Radhuma formation contains saline water that is not being used for water supply. The injection zone is separated by an impervious gypsum formation which acts as a barrier between upper and lower aquifer. State of the art drilling, grouting, and geophysical techniques have been implemented in construction of the wells to assure that the shallow aquifer would not be contaminated and impacted by injected water. Injection and pumping tests were performed to evaluate injection well functionality (injectability). The results of these tests indicated that majority of the wells can accept injection rate of 200 to 300 m³ /h (56 to 83 l/s) under gravity with average value of 250 m³ /h (70 l/s) compared to design value of 50 l/s. This paper presents design and construction process and issues associated with these injection wells, performing injection/pumping tests to determine capacity and effectiveness of the injection wells, the detailed design of collection system and conveying system into the injection wells, and the operation and maintenance process. This system is completed now and is under operation, and therefore, construction of injection wells is an effective option for flood control.

Keywords: deep injection well, flood prevention scheme, geophysical tests, pumping and injection tests, wellhead assembly

Procedia PDF Downloads 104

Authors: Luojie Tang

Abstract:

Riverside space is an important public space and ecological corridor in urban areas, but mountainous urban rivers are often overlooked due to their deep valleys and poor accessibility. This article takes the Qing Shui Xi River in Chongqing as an example, and through long-term field inspections, measurements, interviews, and online surveys, summarizes the problems of poor accessibility, limited space for renovation, lack of waterfront facilities, excessive artificial intervention, low average runoff, severe river water pollution, and difficulty in integrated watershed management in riverside space. Based on the current situation and drawing on relevant experiences, this article summarizes the design methods for riverside space in deep valley rivers in mountainous urban areas. Regarding spatial design techniques, the article emphasizes the importance of integrating waterfront spaces into the urban public space system and vertical linkages. Furthermore, the article suggests different design methods and improvement strategies for the already developed areas and new development areas. Specifically, the article proposes a planning and design strategy of "protection" and "empowerment" for new development areas and an updating and transformation strategy of "improvement" and "revitalization" for already developed areas. In terms of ecological restoration methods, the article suggests three focus points: increasing the runoff of urban rivers, raising the landscape water level during dry seasons, and restoring vegetation and wetlands in the riverbank buffer zone while protecting the overall pattern of the watershed. Additionally, the article presents specific design details of the Qingshuixi River to illustrate the proposed design and restoration techniques.

Keywords: deep-cut river, design method, mountainous city, Qingshuixi river in Chongqing, waterfront space design

Procedia PDF Downloads 85

Authors: Ahmed Negm, George Aggidis, Xiandong Ma

Abstract:

With the diverse nature of urban cities, customer demand patterns, landscape topologies or even seasonal weather trends; managing our water distribution networks (WDNs) has proved a complex task. These unpredictable circumstances manifest as pipe failures, intermittent supply and burst events thus adding to water loss, energy waste and increased carbon emissions. Whilst these events are unavoidable, advanced pressure management has proved an effective tool to control and mitigate them. Henceforth, water utilities have struggled with developing a real-time control method that is resilient when confronting the challenges of water distribution. In this paper we use deep reinforcement learning (DRL) algorithms as a novel pressure control strategy to minimise pressure violations and leakage under both burst and background leakage conditions. Agents based on asynchronous actor critic (A2C) and recurrent proximal policy optimisation (Recurrent PPO) were trained and compared to benchmarked optimisation algorithms (differential evolution, particle swarm optimisation. A2C manages to minimise leakage by 32.48% under burst conditions and 67.17% under background conditions which was the highest performance in the DRL algorithms. A2C and Recurrent PPO performed well in comparison to the benchmarks with higher processing speed and lower computational effort.

Keywords: deep reinforcement learning, pressure management, water distribution networks, leakage management

Procedia PDF Downloads 64

Authors: R. Alias, A. Kasa, M. R. Taha

Abstract:

The effect of particle size on shear strength of granular materials are investigated using direct shear tests. Small direct shear test (60 mm by 60 mm by 24 mm deep) were conducted for particles passing the sieves with opening size of 2.36 mm. Meanwhile, particles passing the standard 20 mm sieves were tested using large direct shear test (300 mm by 300 mm by 200 mm deep). The large direct shear tests and the small direct shear tests carried out using the same shearing rate of 0.09 mm/min and similar normal stresses of 100, 200, and 300 kPa. The results show that the peak and residual shear strength decreases as particle size increases.

Keywords: particle size, shear strength, granular material, direct shear test

Procedia PDF Downloads 471

Authors: Endrick Barnacin, Jean-Luc Henry, Jimmy Nagau, Jack Molinie

Abstract:

Palynology is a field of interest in many disciplines due to its multiple applications: chronological dating, climatology, allergy treatment, and honey characterization. Unfortunately, the analysis of a pollen slide is a complicated and time consuming task that requires the intervention of experts in the field, which are becoming increasingly rare due to economic and social conditions. That is why the need for automation of this task is urgent. A lot of studies have investigated the subject using different standard image processing descriptors and sometimes hand-crafted ones.In this work, we make a comparative study between classical feature extraction methods (Shape, GLCM, LBP, and others) and Deep Learning (CNN, Autoencoders, Transfer Learning) to perform a recognition task over 80 regional pollen species. It has been found that the use of Transfer Learning seems to be more precise than the other approaches

Keywords: pollens identification, features extraction, pollens classification, automated palynology

Procedia PDF Downloads 120

Authors: Anshika Kankane, Dongshik Kang

Abstract:

Marine debris has a significant influence on coastal environments, damaging biodiversity, and causing loss and damage to marine and ocean sector. A functional cost-effective and automatic approach has been used to look up at this problem. Computer vision combined with a deep learning-based model is being proposed to identify and categorize marine debris of seven kinds on different beach locations of Japan. This research compares state-of-the-art deep learning models with a suggested model architecture that is utilized as a feature extractor for debris categorization. The model is being proposed to detect seven categories of litter using a manually constructed debris dataset, with the help of Mask R-CNN for instance segmentation and a shape matching network called HOGShape, which can then be cleaned on time by clean-up organizations using warning notifications of the system. The manually constructed dataset for this system is created by annotating the images taken by fixed KaKaXi camera using CVAT annotation tool with seven kinds of category labels. A pre-trained HOG feature extractor on LIBSVM is being used along with multiple templates matching on HOG maps of images and HOG maps of templates to improve the predicted masked images obtained via Mask R-CNN training. This system intends to timely alert the cleanup organizations with the warning notifications using live recorded beach debris data. The suggested network results in the improvement of misclassified debris masks of debris objects with different illuminations, shapes, viewpoints and litter with occlusions which have vague visibility.

Keywords: computer vision, debris, deep learning, fixed live camera images, histogram of gradients feature extractor, instance segmentation, manually annotated dataset, multiple template matching

Procedia PDF Downloads 89

Authors: Milagros Jaén-Vargas, Javier García Martínez, Karla Miriam Reyes Leiva, María Fernanda Trujillo-Guerrero, Francisco Fernandes, Sérgio Barroso Gonçalves, Miguel Tavares Silva, Daniel Simões Lopes, José Javier Serrano Olmedo

Abstract:

Human Activity Recognition (HAR) is having a growing impact in creating new applications and is responsible for emerging new technologies. Also, the use of wearable sensors is an important key to exploring the human body's behavior when performing activities. Hence, the use of these dispositive is less invasive and the person is more comfortable. In this study, a database that includes three activities is used. The activities were acquired from inertial measurement unit sensors (IMU) and motion capture systems (MOCAP). The main objective is differentiating the performance from four Deep Learning (DL) models: Deep Neural Network (DNN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) and hybrid model Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), when considering acceleration, velocity and position and evaluate if integrating the IMU acceleration to obtain velocity and position represent an increment in performance when it works as input to the DL models. Moreover, compared with the same type of data provided by the MOCAP system. Despite the acceleration data is cleaned when integrating, results show a minimal increase in accuracy for the integrated signals.

Keywords: HAR, IMU, MOCAP, acceleration, velocity, position, feature maps

Procedia PDF Downloads 83

Authors: Shady Adib, Vladimir Vinogradov, Peter Gosling

Abstract:

Digital Twin (DT) technology is a new technology that appeared in the early 21st century. The DT is defined as the digital representation of living and non-living physical assets. By connecting the physical and virtual assets, data are transmitted smoothly, allowing the virtual asset to fully represent the physical asset. Although there are lots of studies conducted on the DT concept, there is still limited information about the ability of the DT models for monitoring and detecting unexpected changes in structural behaviour in real time. This is due to the large computational efforts required for the analysis and an excessively large amount of data transferred from sensors. This paper aims to develop the DT concept to be able to detect the abnormal changes in structural behaviour in real time using advanced modelling techniques, deep learning algorithms, and data acquisition systems, taking into consideration model uncertainties. finite element (FE) models were first developed offline to be used with a reduced basis (RB) model order reduction technique for the construction of low-dimensional space to speed the analysis during the online stage. The RB model was validated against experimental test results for the establishment of a DT model of a two-dimensional truss. The established DT model and deep learning algorithms were used to identify the location of damage once it has appeared during the online stage. Finally, the RB model was used again to identify the damage severity. It was found that using the RB model, constructed offline, speeds the FE analysis during the online stage. The constructed RB model showed higher accuracy for predicting the damage severity, while deep learning algorithms were found to be useful for estimating the location of damage with small severity.

Keywords: data acquisition system, deep learning, digital twin, model uncertainties, reduced basis, reduced order model

Procedia PDF Downloads 82

Authors: Xiangchun Wang, Lihui Zheng, Maozong Gan, Peng Zhang, Tong Wu, An Chang

Abstract:

The deep coal measure tight gas reservoir in Sulige is usually reformed by fracturing, because the reservoir thickness is small, the water layers can be easily communicated during fracturing, which will lead to water production of gas wells and lower production of gas wells. Therefore, it is necessary to control water during fracturing in deep coal measure tight gas reservoir. Using fuzzy-ball fluid to control water fracturing can not only increase the output but also reduce the water output. The fuzzy-ball fluid was prepared indoors to carry out evaluation experiments. The fuzzy ball fluid was mixed in equal volume with the pre-fluid and formation water to test its compatibility. The core displacement device was used to test the gas and water breaking through the matrix and fractured cores blocked by fuzzy-ball fluid. The breakthrough pressure of the plunger tests its water blocking performance. The experimental results show that there is no precipitation after the fuzzy-ball fluid is mixed with the pad fluid and the formation water, respectively. The breakthrough pressure gradients of gas and water after the fuzzy-ball fluid plugged the cracks were 0.02MPa/cm and 0.04MPa/cm, respectively, and the breakthrough pressure gradients of gas and water after the matrix was plugged were 0.03MPa/cm and 0.2MPa/cm, respectively, which meet the requirements of field operation. Two wells A and B in the Sulige Gas Field were used on site to implement water control fracturing. After the pre-fluid was injected into the two wells, 50m3 of fuzzy-ball fluid was pumped to plug the water. The construction went smoothly. After water control and fracturing, the average daily output in 161 days was increased by 13.71% and 6.99% compared with that of adjacent wells in the same layer. The adjacent wells were bubbled for 3 times and 63 times respectively, while there was no effusion in A and B construction wells. The results show that fuzzy-ball fluid is a water plugging material suitable for water control fracturing in tight gas wells, and its water control mechanism can also provide a new idea for the development of water control fracturing materials.

Keywords: coal seam, deep layer, fracking, fuzzy-ball fluid, reservoir reconstruction

Procedia PDF Downloads 205

Authors: Rahma Abed, Sahbi Bahroun, Ezzeddine Zagrouba

Abstract:

Due to the huge amount of data in videos, extracting the relevant frames became a necessity and an essential step prior to performing face recognition. In this context, we propose a method for extracting keyframes from videos based on face quality and deep learning for a face recognition task. This method has two steps. We start by generating face quality scores for each face image based on the use of three face feature extractors, including Gabor, LBP, and HOG. The second step consists in training a Deep Convolutional Neural Network in a supervised manner in order to select the frames that have the best face quality. The obtained results show the effectiveness of the proposed method compared to the methods of the state of the art.

Keywords: keyframe extraction, face quality assessment, face in video recognition, convolution neural network

Procedia PDF Downloads 210

Authors: Ehsan Heidaryan

Abstract:

Gas hydrate sediments are ice like permafrost in deep see and oceans. Methane production in sequestration process and reducing atmospheric carbon dioxide, a main source of greenhouse gas, has been accentuated recently. One focus is capture, separation, and sequestration of industrial carbon dioxide. As a hydrate former, carbon dioxide forms hydrates at moderate temperatures and pressures. This phenomenon could be utilized to capture and separate carbon dioxide from flue gases, and also has the potential to sequester carbon dioxide in the deep seabeds. This research investigated the effect of synthetic surfactants on carbon dioxide hydrate formation, catalysis and consequently, methane production from hydrate permafrosts in sediments. It investigated the sequestration potential of carbon dioxide hydrates in ocean sediments. Also, the catalytic effect of biosurfactants in these processes was investigated.

Keywords: carbon dioxide, hydrate, sequestration, surfactant

Procedia PDF Downloads 418

Authors: Abdelhak Djoudi, Hachemi Chekireb, El Madjid Berkouk

Abstract:

The doubly fed induction generator (DFIG) received recently an important consideration in medium and high power wind energy conversion systems integration, due to its advantages compared to other generators types. The stator power sliding mode control (SPSMC) proves a great efficiency judge against other control laws and schemes. In the SPSMC laws elaborated by several authors, only the slide surface tracking conditions are elaborated using Lyapunov functions, and the boundedness of the DFIG states is never treated. Some works have validated theirs approaches by experiments results in the case of specified machines, but these verifications stay insufficient to generalize to other machines range. Adding to this argument, the DFIG states boundedness demonstration is widely suggested in goal to ensure that in the application of the SPSMC, the states evaluates within theirs tolerable bounds. Our objective in the present paper is to highlight the efficiency of the SPSMC by stability analysis. The boundedness of the DFIG states such as the stator current and rotor flux is discussed. Moreover, the states trajectories are finding using analytical proves taking into consideration the SPSMC gains.

Keywords: Doubly Fed Induction Generator (DFIG), Stator Powers Sliding Mode Control (SPSMC), lyapunov function, stability, states boundedness, trajectories mathematical proves

Procedia PDF Downloads 383

Authors: Martin K. Steiger, Lukas Heisler, Hans-Georg Brachtendorf

Abstract:

Deep neural networks and their variants form the backbone of many AI applications. Based on the so-called residual networks, a continuous formulation of such models as ordinary differential equations (ODEs) has proven advantageous since different techniques may be applied that significantly increase the learning speed and enable controlled trade-offs with the resulting error at the same time. For the evaluation of such models, high-performance numerical differential equation solvers are used, which also provide the gradients required for training. However, whether classical gradient-based methods are even applicable or which one yields the best results has not been discussed yet. This paper aims to redeem this situation by providing empirical results for different applications.

Keywords: deep neural networks, gradient-based learning, image processing, ordinary differential equation networks

Procedia PDF Downloads 144

Authors: Simone Leonardi, Luca Ardito

Abstract:

Micro-influencers are effective elements in the marketing strategies of companies and institutions because of their capability to create an hyper-engaged audience around a specific topic of interest. In recent years, many scientific approaches and commercial tools have handled the task of detecting this type of social media users. These strategies adopt solutions ranging from rule based machine learning models to deep neural networks and graph analysis on text, images, and account information. This work compares the existing solutions and proposes an ensemble method to generalize them with different input data and social media platforms. The deployed solution combines deep learning models on unstructured data with statistical machine learning models on structured data. We retrieve both social media accounts information and multimedia posts on Twitter and Instagram. These data are mapped into feature vectors for an eXtreme Gradient Boosting (XGBoost) classifier. Sixty different topics have been analyzed to build a rule based gold standard dataset and to compare the performances of our approach against baseline classifiers. We prove the effectiveness of our work by comparing the accuracy, precision, recall, and f1 score of our model with different configurations and architectures. We obtained an accuracy of 0.91 with our best performing model.

Keywords: deep learning, gradient boosting, image processing, micro-influencers, NLP, social media

Procedia PDF Downloads 159

Authors: Burak Oksuzoglu, Savas Yildirim, Ferhat Kutlu

Abstract:

Opinion mining (OM) is one of the natural language processing (NLP) problems to determine the polarity of opinions, mostly represented on a positive-neutral-negative axis. The data for OM is usually collected from various social media platforms. In an era where social media has considerable control over companies’ futures, it’s worth understanding social media and taking actions accordingly. OM comes to the fore here as the scale of the discussion about companies increases, and it becomes unfeasible to gauge opinion on individual levels. Thus, the companies opt to automize this process by applying machine learning (ML) approaches to their data. For the last two decades, OM or sentiment analysis (SA) has been mainly performed by applying ML classification algorithms such as support vector machines (SVM) and Naïve Bayes to a bag of n-gram representations of textual data. With the advent of deep learning and its apparent success in NLP, traditional methods have become obsolete. Transfer learning paradigm that has been commonly used in computer vision (CV) problems started to shape NLP approaches and language models (LM) lately. This gave a sudden rise to the usage of the pretrained language model (PTM), which contains language representations that are obtained by training it on the large datasets using self-supervised learning objectives. The PTMs are further fine-tuned by a specialized downstream task dataset to produce efficient models for various NLP tasks such as OM, NER (Named-Entity Recognition), Question Answering (QA), and so forth. In this study, the traditional and modern NLP approaches have been evaluated for OM by using a sizable corpus belonging to a large private company containing about 76,000 comments in Turkish: SVM with a bag of n-grams, and two chosen pre-trained models, multilingual universal sentence encoder (MUSE) and bidirectional encoder representations from transformers (BERT). The MUSE model is a multilingual model that supports 16 languages, including Turkish, and it is based on convolutional neural networks. The BERT is a monolingual model in our case and transformers-based neural networks. It uses a masked language model and next sentence prediction tasks that allow the bidirectional training of the transformers. During the training phase of the architecture, pre-processing operations such as morphological parsing, stemming, and spelling correction was not used since the experiments showed that their contribution to the model performance was found insignificant even though Turkish is a highly agglutinative and inflective language. The results show that usage of deep learning methods with pre-trained models and fine-tuning achieve about 11% improvement over SVM for OM. The BERT model achieved around 94% prediction accuracy while the MUSE model achieved around 88% and SVM did around 83%. The MUSE multilingual model shows better results than SVM, but it still performs worse than the monolingual BERT model.

Keywords: BERT, MUSE, opinion mining, pretrained language model, SVM, Turkish

Procedia PDF Downloads 123

Authors: Atish Bagchi, Siva Chandrasekaran

Abstract:

Background: According to World Bank national accounts data, the estimated global manufacturing value-added output in 2020 was 13.74 trillion USD. These manufacturing processes are monitored, modelled, and controlled by advanced, real-time, computer-based systems, e.g., Industrial IoT, PLC, SCADA, etc. These systems measure and manipulate a set of physical variables, e.g., temperature, pressure, etc. Despite the use of IoT, SCADA etc., in manufacturing, studies suggest that unplanned downtime leads to economic losses of approximately 864 billion USD each year. Therefore, real-time, accurate detection, classification and prediction of machine behaviour are needed to minimise financial losses. Although vast literature exists on time-series data processing using machine learning, the challenges faced by the industries that lead to unplanned downtimes are: The current algorithms do not efficiently handle the high-volume streaming data from industrial IoTsensors and were tested on static and simulated datasets. While the existing algorithms can detect significant 'point' outliers, most do not handle contextual outliers (e.g., values within normal range but happening at an unexpected time of day) or subtle changes in machine behaviour. Machines are revamped periodically as part of planned maintenance programmes, which change the assumptions on which original AI models were created and trained. Aim: This research study aims to deliver a Graph Neural Network(GNN)based hybrid forecasting model that interfaces with the real-time machine control systemand can detect, predict machine behaviour and behavioural changes (anomalies) in real-time. This research will help manufacturing industries and utilities, e.g., water, electricity etc., reduce unplanned downtimes and consequential financial losses. Method: The data stored within a process control system, e.g., Industrial-IoT, Data Historian, is generally sampled during data acquisition from the sensor (source) and whenpersistingin the Data Historian to optimise storage and query performance. The sampling may inadvertently discard values that might contain subtle aspects of behavioural changes in machines. This research proposed a hybrid forecasting and classification model which combines the expressive and extrapolation capability of GNN enhanced with the estimates of entropy and spectral changes in the sampled data and additional temporal contexts to reconstruct the likely temporal trajectory of machine behavioural changes. The proposed real-time model belongs to the Deep Learning category of machine learning and interfaces with the sensors directly or through 'Process Data Historian', SCADA etc., to perform forecasting and classification tasks. Results: The model was interfaced with a Data Historianholding time-series data from 4flow sensors within a water treatment plantfor45 days. The recorded sampling interval for a sensor varied from 10 sec to 30 min. Approximately 65% of the available data was used for training the model, 20% for validation, and the rest for testing. The model identified the anomalies within the water treatment plant and predicted the plant's performance. These results were compared with the data reported by the plant SCADA-Historian system and the official data reported by the plant authorities. The model's accuracy was much higher (20%) than that reported by the SCADA-Historian system and matched the validated results declared by the plant auditors. Conclusions: The research demonstrates that a hybrid GNN based approach enhanced with entropy calculation and spectral information can effectively detect and predict a machine's behavioural changes. The model can interface with a plant's 'process control system' in real-time to perform forecasting and classification tasks to aid the asset management engineers to operate their machines more efficiently and reduce unplanned downtimes. A series of trialsare planned for this model in the future in other manufacturing industries.

Keywords: GNN, Entropy, anomaly detection, industrial time-series, AI, IoT, Industry 4.0, Machine Learning

Procedia PDF Downloads 130

Authors: S. Mohamed, D. Gonzalez, C. Sayada, P. Halfon

Abstract:

Drug resistance mutations are routinely detected using standard Sanger sequencing, which does not detect minor variants with a frequency below 20%. The impact of detecting minor variants generated by ultra-deep sequencing (UDS) on HIV drug-resistance (DR) interpretations has not yet been studied. Fifty HIV-1 patients who experienced virological failure were included in this retrospective study. The HIV-1 UDS protocol allowed the detection and quantification of HIV-1 protease and reverse transcriptase variants related to genotypes A, B, C, E, F, and G. DeepChek®-HIV simplified DR interpretation software was used to compare Sanger sequencing and UDS. The total time required for the UDS protocol was found to be approximately three times longer than Sanger sequencing with equivalent reagent costs. UDS detected all of the mutations found by population sequencing and identified additional resistance variants in all patients. An analysis of DR revealed a total of 643 and 224 clinically relevant mutations by UDS and Sanger sequencing, respectively. Three resistance mutations with > 20% prevalence were detected solely by UDS: A98S (23%), E138A (21%) and V179I (25%). A significant difference in the DR interpretations for 19 antiretroviral drugs was observed between the UDS and Sanger sequencing methods. Y181C and T215Y were the most frequent mutations associated with interpretation differences. A combination of UDS and DeepChek® software for the interpretation of DR results would help clinicians provide suitable treatments. A cut-off of 1% allowed a better characterisation of the viral population by identifying additional resistance mutations and improving the DR interpretation.

Keywords: HIV-1, ultra-deep sequencing, Sanger sequencing, drug resistance

Procedia PDF Downloads 317

Authors: Jiaxin Shi, Kaifeng Hao, Qingfan An, Zeng Peng

Abstract:

Due to a lack of data sources and data processing techniques, it has always been difficult to quantify public space quality, which includes urban construction quality and how it is perceived by people, especially in large urban areas. This study proposes a quantitative research method based on the consideration of emotional health and physical health of the built environment. It highlights the low quality of public areas in Tianjin, China, where there are many negative elements. Deep learning technology is then used to measure how effectively people perceive urban areas. First, this work suggests a deep learning model that might simulate how people can perceive the quality of urban construction. Second, we perform semantic segmentation on street images to identify visual elements influencing scene perception. Finally, this study correlated the scene perception score with the proportion of visual elements to determine the surrounding environmental elements that influence scene perception. Using a small-scale labeled Tianjin street view data set based on transfer learning, this study trains five negative spatial discriminant models in order to explore the negative space distribution and quality improvement of urban streets. Then it uses all Tianjin street-level imagery to make predictions and calculate the proportion of negative space. Visualizing the spatial distribution of negative space along the Tianjin Inner Ring Road reveals that the negative elements are mainly found close to the five key districts. The map of Tianjin was combined with the experimental data to perform the visual analysis. Based on the emotional assessment, the distribution of negative materials, and the direction of street guidelines, we suggest guidance content and design strategy points of the negative phenomena in Tianjin street space in the two dimensions of perception and substance. This work demonstrates the utilization of deep learning techniques to understand how people appreciate high-quality urban construction, and it complements both theory and practice in urban planning. It illustrates the connection between human perception and the actual physical public space environment, allowing researchers to make urban interventions.

Keywords: human perception, public space quality, deep learning, negative elements, street images

Procedia PDF Downloads 95

Authors: Ali Almadan, Anoop Krishnan, Ajita Rattani

Abstract:

Facial recognition is a widely adopted technology in surveillance, border control, healthcare, banking services, and lately, in mobile user authentication with Apple introducing “Face ID” moniker with iPhone X. A lot of research has been conducted in the area of face recognition on datasets captured by surveillance cameras, DSLR, and mobile devices. Recently, face recognition technology has also been deployed on body-worn cameras to keep officers safe, enabling situational awareness and providing evidence for trial. However, limited academic research has been conducted on this topic so far, without the availability of any publicly available datasets with a sufficient sample size. This paper aims to advance research in the area of face recognition using body-worn cameras. To this aim, the contribution of this work is two-fold: (1) collection of a dataset consisting of a total of 136,939 facial images of 102 subjects captured using body-worn cameras in in-door and daylight conditions and (2) evaluation of various deep-learning architectures for face identification on the collected dataset. Experimental results suggest a maximum True Positive Rate(TPR) of 99.86% at False Positive Rate(FPR) of 0.000 obtained by SphereFace based deep learning architecture in daylight condition. The collected dataset and the baseline algorithms will promote further research and development. A downloadable link of the dataset and the algorithms is available by contacting the authors.

Keywords: face recognition, body-worn cameras, deep learning, person identification

Procedia PDF Downloads 150

Authors: Jamison Duckworth, Shankarachary Ragi

Abstract:

Detection and segmentation of wildfire smoke plumes from remote sensing imagery are being pursued as a solution for early fire detection and response. Smoke plume detection can be automated and made robust by the application of artificial intelligence methods. Specifically, in this study, the deep learning approach Mask Region-based Convolutional Neural Network (RCNN) is being proposed to learn smoke patterns across different spectral bands. This method is proposed to separate the smoke regions from the background and return masks placed over the smoke plumes. Multispectral data was acquired using NASA’s Earthdata and WorldView and services and satellite imagery. Due to the use of multispectral bands along with the three visual bands, we show that Mask R-CNN can be applied to distinguish smoke plumes from clouds and other landscape features that resemble smoke.

Keywords: deep learning, mask-RCNN, smoke plumes, spectral bands

Procedia PDF Downloads 107

Authors: Shri Krishna Mishra

Abstract:

In an educational context, technology can prompt interactive plight only when it is used in conjunction with interactive plight methods. This study, therefore, examines the relationships among components that contribute to higher levels of interactive plight and execution, such as interactive Plight methods, technology, intrinsic motivation and deep learning. 526 students participated in this study. With structural equation modelling, the authors test the conceptual model and identify satisfactory model fit. The results indicate that interactive Plight methods, technology and intrinsic motivation have significant relationship with interactive Plight; deep learning mediates the relationships of the other variables with Execution.

Keywords: searching the relationship among components, contribute to interactive plight, educational execution, intrinsic motivation

Procedia PDF Downloads 437

Authors: Kartik Thakore, Anna-Roza Tamas, Adam Cole

Abstract:

The world health organization (WHO) reports that by 2030 more than 23.7 million deaths annually will be caused by Cardiovascular Diseases (CVDs); with a 2008 economic impact of $3.76 T. Metabolic syndrome is a disorder of multiple metabolic risk factors strongly indicated in the development of cardiovascular diseases. Guided lifestyle intervention driven by live coaching has been shown to have a positive impact on metabolic risk factors. Individuals’ path to improved (decreased) metabolic risk factors are driven by personal motivation and personalized messages delivered by coaches and augmented by technology. Using interactions captured between 400 individuals and 3 coaches over a program period of 500 days, a preliminary model was designed. A novel real time event tracking system was created to track and classify clients based on their genetic profile, baseline questionnaires and usage of a mobile application with live coaching sessions. Classification of clients and coaches was done using a support vector machines application build on Apache Spark, Stanford Natural Language Processing Library (SNLPL) and decision-modeling.

Keywords: guided lifestyle intervention, metabolic risk factors, personal coaching, support vector machines application, Apache Spark, natural language processing

Procedia PDF Downloads 417

Authors: Prathiksha Kamath, Pratibha Jamkhandi, Prateek Ghanti, Priyanshu Gupta, M. Lakshmi Neelima

Abstract:

Combining Image Captioning and Visual Question Answering (VQA) tasks have emerged as a new and exciting research area. The image captioning task involves generating a textual description that summarizes the content of the image. VQA aims to answer a natural language question about the image. Both these tasks include computer vision and natural language processing (NLP) and require a deep understanding of the content of the image and semantic relationship within the image and the ability to generate a response in natural language. There has been remarkable growth in both these tasks with rapid advancement in deep learning. In this paper, we present a comprehensive review of recent progress in combining image captioning and visual question-answering (VQA) tasks. We first discuss both image captioning and VQA tasks individually and then the various ways in which both these tasks can be integrated. We also analyze the challenges associated with these tasks and ways to overcome them. We finally discuss the various datasets and evaluation metrics used in these tasks. This paper concludes with the need for generating captions based on the context and captions that are able to answer the most likely asked questions about the image so as to aid the VQA task. Overall, this review highlights the significant progress made in combining image captioning and VQA, as well as the ongoing challenges and opportunities for further research in this exciting and rapidly evolving field, which has the potential to improve the performance of real-world applications such as autonomous vehicles, robotics, and image search.

Keywords: image captioning, visual question answering, deep learning, natural language processing

Procedia PDF Downloads 59

Authors: Edward K. Boahen, Brunel E. Bouya-Moko, Changda Wang

Abstract:

The major threat to online social network (OSN) users is account compromisation. Spammers now spread malicious messages by exploiting the trust relationship established between account owners and their friends. The challenge in detecting a compromised account by service providers is validating the trusted relationship established between the account owners, their friends, and the spammers. Another challenge is the increase in required human interaction with the feature selection. Research available on supervised learning (machine learning) has limitations with the feature selection and accounts that cannot be profiled, like application programming interface (API). Therefore, this paper discusses the various behaviours of the OSN users and the current approaches in detecting a compromised OSN account, emphasizing its limitations and challenges. We propose a deep learning approach that addresses and resolve the constraints faced by the previous schemes. We detailed our proposed optimized nonsymmetric deep auto-encoder (OPT_NDAE) for unsupervised feature learning, which reduces the required human interaction levels in the selection and extraction of features. We evaluated our proposed classifier using the NSL-KDD and KDDCUP'99 datasets in a graphical user interface enabled Weka application. The results obtained indicate that our proposed approach outperformed most of the traditional schemes in OSN compromised account detection with an accuracy rate of 99.86%.

Keywords: computer security, network security, online social network, account compromisation

Procedia PDF Downloads 100