Corporate Social Responsibility Reporting, State Ownership, and Corporate Performance in China: Proof from Longitudinal Data of Publicly Traded Enterprises from 2006 to 2020

This paper offered the primary methodical proof on how Corporate Social Responsibility (CSR) reporting related to enterprise earnings in listed firms in China in light of most evidence focusing on cross-sectional data or data in a short span of time. Using full economic and business panel data on China’s publicly listed enterprises from 2006 to 2020 over two decades in the China Stock Market & Accounting Research database, we found initial evidence of significant direct relations between CSR reporting and firm corporate performance in both state-owned and privately-owned firms over this period, supporting the stakeholder theory. Results also revealed that state-owned enterprises performed as well as private enterprises in the current period. But private enterprises performed better than state-owned enterprises in the subsequent years. Moreover, the release of social responsibility reports had the more significant impact on the financial performance of state-owned and private enterprises in the current period than in the subsequent periods. Specifically, CSR release was not significantly associated to the financial performance of state-owned enterprises on the lag of the first, second, and third periods. But it had an impact on the lag of the first, second, and third periods among private enterprises. Such findings suggested that CSR reporting helped improve the corporate financial performance of state-owned and private enterprises in the current period, but this kind of effect was more significant among private enterprises in the lag periods.

WebAppShield: An Approach Exploiting Machine Learning to Detect SQLi Attacks in an Application Layer in Run-Time

In recent years, SQL injection attacks have been identified as being prevalent against web applications. They affect network security and user data, which leads to a considerable loss of money and data every year. This paper presents the use of classification algorithms in machine learning using a method to classify the login data filtering inputs into "SQLi" or "Non-SQLi,” thus increasing the reliability and accuracy of results in terms of deciding whether an operation is an attack or a valid operation. A method as a Web-App is developed for auto-generated data replication to provide a twin of the targeted data structure. Shielding against SQLi attacks (WebAppShield) that verifies all users and prevents attackers (SQLi attacks) from entering and or accessing the database, which the machine learning module predicts as "Non-SQLi", has been developed. A special login form has been developed with a special instance of the data validation; this verification process secures the web application from its early stages. The system has been tested and validated, and up to 99% of SQLi attacks have been prevented.

Controlled Vocabularies and Information Retrieval: 1918 Pandemic’s Scientific Literature as an Example

The role of controlled vocabularies in information retrieval is broadly recognized as a relevant feature. Besides, there is a standing demand that editors and databases should consider the effective introduction of controlled vocabularies in their procedures to index scientific literature. That is especially important because information retrieval is pointed out as a significant point to drive systematic literature review. Hence, a first question emerges: Are the controlled vocabularies at this moment considered? On the other hand, subject searching in the catalogs is complex mainly due to the dichotomy between keywords from authors versus keywords based on controlled vocabularies. Finally, there is some demand to unify the terminology related to health to make easier the medical history exploitation and research. Considering these features, this paper focuses on controlled vocabularies related to the health field and their role for storing, classifying, and retrieving relevant literature. The objective is knowing which role plays the controlled vocabularies related to the health field to index and retrieve research literature in data bases such as Web of Science (WoS) and Scopus. So, this exploratory research is grounded over two research questions: 1) Which are the terms considered in specific controlled vocabularies of the health field; and 2) How papers are indexed in relevant databases to be easily retrieved, considering keywords vs specific health’ controlled vocabularies? This research takes as fieldwork the controlled vocabularies related to health and the scientific interest for 1918 flu pandemic, also known equivocally as ‘Spanish flu’. This interest has been fostered by the emergence in the early 21st of epidemics of pneumonic diseases caused by virus. Searches about and with controlled vocabularies on WoS and Scopus databases are conducted. First results of this work in progress are surprising. There are different controlled vocabularies for the health field, into which the terms collected and preferred related to ‘1918 pandemic’ are identified. To summarize, ‘Spanish influenza epidemic’ or ‘Spanish flu’ are collected as not preferred terms. The preferred terms are: ‘influenza’ or ‘influenza pandemic, 1918-1919’. Although the controlled vocabularies are clear in their election, most of the literature about ‘1918 pandemic’ is retrievable either by ‘Spanish’ or by ‘1918’ disjunct, and the dominant word to retrieve literature is ‘Spanish’ rather than ‘1918’. This is surprising considering the existence of suitable controlled vocabularies related to health topics, and the modern guidelines of World Health Organization concerning naming of diseases that point out to other preferred terms. A first conclusion is the failure of using controlled vocabularies for a field such as health, and in consequence for WoS and Scopus. This research opens further research questions about which is the role that controlled vocabularies play in the instructions to authors that journals deliver to documents’ authors.

Director Compensation, CEO Duality, State Ownership, and Firm Performance in China: Proof from Panel Data of Publicly Listed Enterprises from 1999 to 2020

This paper offered the primary methodical proof on how director remuneration related to enterprise earnings in listed firms in China in light of most evidence focusing on cross-sectional data or data in a short span of time. Using full economic and business panel data on China’s publicly listed enterprise from 1999 to 2020 over two decades in the China Stock Market & Accounting Research database, we found statistically significant positive associations between director pay and firm performance in privately owned firms over this period, supporting the agency theory. In contrast, among the state-owned enterprises, there was a reverse relation between director compensation and firm financial performance, contributing to the existing literature. But the results also revealed that state-owned enterprises financially performed as well as private enterprises. Such findings suggested that state ownership might line up officials’ career incentives with party prime concern rather than pecuniary incentives. Also, CEO duality enhanced firm performance. As such, allegiance to the party and possible advancement to an upper-level political position would motivate company directors in state-owned enterprises. On the other hand, directors in privately owned enterprises might be motivated by monetary incentives. In addition, a statistical regression model was proposed and tested to get the results of the performance of state-owned enterprises. Finally, some suggestions were made about how to improve the institutional management of government-owned corporations in China.

An Empirical Assessment of Sustainability of an Urban Water Supply Service Delivery

Urban population is rapidly increasing in Ilorin, (the capital of Kwara State of Nigeria) along with related increased water demand. The inadequacies of water supply services have forced the populace to depend on dug wells, boreholes, water tankers, street vendors etc. for their water needs. People spend hours daily carrying jerry can all around to collect and queue for water at the public water tap with high opportunity cost both in time and economic wastage. This situation motivated this study to assess the sustainability of an urban water supply services to unravel the factors undermining the effective delivery of services. Contingent Valuation Method was used to place value on water supply services using the Double Bounded Dichotomous Choice format for willingness to pay elicitation. A database was created with Microsoft Excel and Stata 12 Software to model and evaluate the variables that affect household willingness to pay. The results of the study reveal that about 92% of the total households surveyed were connected to the Government water supply out of which 87% reported that they were not satisfied with the existing services. The results furthered revealed that respondents are willing to pay ₦2500 monthly to enjoy sustainable water supply service delivery.

Clustering for Detection of Population Groups at Risk from Anticholinergic Medication

Anticholinergic medication has been associated with events such as falls, delirium, and cognitive impairment in older patients. To further assess this, anticholinergic burden scores have been developed to quantify risk. A risk model based on clustering was deployed in a healthcare management system to cluster patients into multiple risk groups according to anticholinergic burden scores of multiple medicines prescribed to patients to facilitate clinical decision-making. To do so, anticholinergic burden scores of drugs were extracted from the literature which categorizes the risk on a scale of 1 to 3. Given the patients’ prescription data on the healthcare database, a weighted anticholinergic risk score was derived per patient based on the prescription of multiple anticholinergic drugs. This study was conducted on 300,000 records of patients currently registered with a major regional UK-based healthcare provider. The weighted risk scores were used as inputs to an unsupervised learning algorithm (mean-shift clustering) that groups patients into clusters that represent different levels of anticholinergic risk. This work evaluates the association between the average risk score and measures of socioeconomic status (index of multiple deprivation) and health (index of health and disability). The clustering identifies a group of 15 patients at the highest risk from multiple anticholinergic medication. Our findings show that this group of patients is located within more deprived areas of London compared to the population of other risk groups. Furthermore, the prescription of anticholinergic medicines is more skewed to female than male patients, suggesting that females are more at risk from this kind of multiple medication. The risk may be monitored and controlled in a healthcare management system that is well-equipped with tools implementing appropriate techniques of artificial intelligence.

Random Subspace Neural Classifier for Meteor Recognition in the Night Sky

This article describes the Random Subspace Neural Classifier (RSC) for the recognition of meteors in the night sky. We used images of meteors entering the atmosphere at night between 8:00 p.m.-5: 00 a.m. The objective of this project is to classify meteor and star images (with stars as the image background). The monitoring of the sky and the classification of meteors are made for future applications by scientists. The image database was collected from different websites. We worked with RGB-type images with dimensions of 220x220 pixels stored in the BitMap Protocol (BMP) format. Subsequent window scanning and processing were carried out for each image. The scan window where the characteristics were extracted had the size of 20x20 pixels with a scanning step size of 10 pixels. Brightness, contrast and contour orientation histograms were used as inputs for the RSC. The RSC worked with two classes and classified into: 1) with meteors and 2) without meteors. Different tests were carried out by varying the number of training cycles and the number of images for training and recognition. The percentage error for the neural classifier was calculated. The results show a good RSC classifier response with 89% correct recognition. The results of these experiments are presented and discussed.

Data Collection in Hospital Emergencies: A Questionnaire Survey

Many methods are used to collect data like questionnaires, surveys, focus group interviews. Or the collection of poor-quality data resulting, for example, from poorly designed questionnaires, the absence of good translators or interpreters, and the incorrect recording of data allow conclusions to be drawn that are not supported by the data or to focus only on the average effect of the program or policy. There are several solutions to avoid or minimize the most frequent errors, including obtaining expert advice on the design or adaptation of data collection instruments; or use technologies allowing better "anonymity" in the responses. In this context, and to overcome the aforementioned problems, we suggest in this paper an approach to achieve the collection of relevant data, by carrying out a large-scale questionnaire-based survey. We have been able to collect good quality, consistent and practical data on hospital emergencies to improve emergency services in hospitals, especially in the case of epidemics or pandemics.

A Generic Middleware to Instantly Sync Intensive Writes of Heterogeneous Massive Data via Internet

Industry data centers often need to sync data changes reliably and instantly from a large-scale of heterogeneous autonomous relational databases accessed via the not-so-reliable Internet, for which a practical generic sync middleware of low maintenance and operation costs is most wanted. To this demand, this paper presented a generic sync middleware system (GSMS), which has been developed, applied and optimized since 2006, holding the principles or advantages that it must be SyncML-compliant and transparent to data application layer logic without referring to implementation details of databases synced, does not rely on host computer operating systems deployed, and its construction is light weighted and hence of low cost. Regarding these hard commitments of developing GSMS, in this paper we stressed the significant optimization breakthrough of GSMS sync delay being well below a fraction of millisecond per record sync. A series of ultimate tests with GSMS sync performance were conducted for a persuasive example, in which the source relational database underwent a broad range of write loads (from one thousand to one million intensive writes within a few minutes). All these tests showed that the performance of GSMS is competent and smooth even under ultimate write loads.

Developing a Coronavirus Academic Paper Sorting Application

The COVID-19 Literature Summary App, now live on the university website, was created for the primary purpose of enabling academicians and clinicians to quickly sort through the vast array of recent coronavirus publications by topics of interest. Multiple methods of summarizing and sorting the manuscripts were created. A summary page introduces the application function and capabilities, while an interactive map provides daily updates on infection, death, and recovery rates. A page with a pivot table allows publication sorting by topic, with an interactive data table that allows sorting topics by columns, as wells as the capability to view abstracts. Additionally, publications may be sorted by the medical topics they cover. We used the CORD-19 database to compile lists of publications. The data table can sort binary variables, allowing the user to pick desired publication topics, such as papers that describe COVID-19 symptoms. The application is primarily designed for use by researchers but can be used by anybody who wants a faster and more efficient means of locating papers of interest. 

Empirical Analysis of Velocity Behavior for Collaborative Robots in Transient Contact Cases

In this paper, a suitable measurement setup is presented to conduct force and pressure measurements for transient contact cases at the example of lathe machine tending. Empirical measurements were executed on a selected collaborative robot’s behavior regarding allowable operating speeds under consideration of sensor- and workpiece-specific factors. Comparisons between the theoretic calculations proposed in ISO/TS 15066 and the practical measurement results reveal a basis for future research. With the created database, preliminary risk assessment and economic assessment procedures of collaborative machine tending cells can be facilitated.

A Commercial Building Plug Load Management System That Uses Internet of Things Technology to Automatically Identify Plugged-In Devices and Their Locations

Plug and process loads (PPLs) account for a large portion of U.S. commercial building energy use. There is a huge potential to reduce whole building consumption by targeting PPLs for energy savings measures or implementing some form of plug load management (PLM). Despite this potential, there has yet to be a widely adopted commercial PLM technology. This paper describes the Automatic Type and Location Identification System (ATLIS), a PLM system framework with automatic and dynamic load detection (ADLD). ADLD gives PLM systems the ability to automatically identify devices as they are plugged into the outlets of a building. The ATLIS framework takes advantage of smart, connected devices to identify device locations in a building, meter and control their power, and communicate this information to a central database. ATLIS includes five primary capabilities: location identification, communication, control, energy metering, and data storage. A laboratory proof of concept (PoC) demonstrated all but the energy metering capability, and these capabilities were validated using a series of system tests. The PoC was able to identify when a device was plugged into an outlet and the location of the device in the building. When a device was moved, the PoC’s dashboard and database were automatically updated with the new location. The PoC implemented controls to devices from the system dashboard so that devices maintained correct schedules regardless of where they were plugged in within the building. ATLIS’s primary technology application is improved PLM, but other applications include asset management, energy audits, and interoperability for grid-interactive efficient buildings. An ATLIS-based system could also be used to direct power to critical devices, such as ventilators, during a brownout or blackout. Such a framework is an opportunity to make PLM more widespread and reduce the amount of energy consumed by PPLs in current and future commercial buildings.

Scholar Index for Research Performance Evaluation Using Multiple Criteria Decision Making Analysis

This paper aims to present an objective quantitative methodology on how to evaluate individual’s scholarly research output using multiple criteria decision analysis. A multiple criteria decision making analysis (MCDMA) methodological process is adopted to build a multiple criteria evaluation model. With the introduction of the scholar index, which gives significant information about a researcher's productivity and the scholarly impact of his or her publications in a single number (s is the number of publications with at least s citations); cumulative research citation index; the scholar index is included in the citation databases to cover the multidimensional complexity of scholarly research performance and to undertake objective evaluations with scholar index. The scholar index, one of publication activity indexes, is analyzed by considering it to be the most appropriate sciencemetric indicator which allows to smooth over many drawbacks of scholarly output assessment by mere calculation of the number of publications (quantity) and citations (quality). Hence, this study includes a set of indicators-based scholar index to be used for evaluating scholarly researchers. Google Scholar open science database was used to assess and discuss scholarly productivity and impact of researchers. Based on the experiment of computing the scholar index, and its derivative indexes for a set of researchers on open research database platform, quantitative methods of assessing scholarly research output were successfully considered to rank researchers. The proposed methodology considers the ranking, and the selection of data on which a scholarly research performance evaluation was based, the analysis of the data, and the presentation of the multiple criteria analysis results.

IntelligentLogger: A Heavy-Duty Vehicles Fleet Management System Based on IoT and Smart Prediction Techniques

Both daily and long-term management of a heavy-duty vehicles and construction machinery fleet is an extremely complicated and hard to solve issue. This is mainly due to the diversity of the fleet vehicles – machinery, which concerns not only the vehicle types, but also their age/efficiency, as well as the fleet volume, which is often of the order of hundreds or even thousands of vehicles/machineries. In the present paper we present “InteligentLogger”, a holistic heavy-duty fleet management system covering a wide range of diverse fleet vehicles. This is based on specifically designed hardware and software for the automated vehicle health status and operational cost monitoring, for smart maintenance. InteligentLogger is characterized by high adaptability that permits to be tailored to practically any heavy-duty vehicle/machinery (of different technologies -modern or legacy- and of dissimilar uses). Contrary to conventional logistic systems, which are characterized by raised operational costs and often errors, InteligentLogger provides a cost-effective and reliable integrated solution for the e-management and e-maintenance of the fleet members. The InteligentLogger system offers the following unique features that guarantee successful heavy-duty vehicles/machineries fleet management: (a) Recording and storage of operating data of motorized construction machinery, in a reliable way and in real time, using specifically designed Internet of Things (IoT) sensor nodes that communicate through the available network infrastructures, e.g., 3G/LTE; (b) Use on any machine, regardless of its age, in a universal way; (c) Flexibility and complete customization both in terms of data collection, integration with 3rd party systems, as well as in terms of processing and drawing conclusions; (d) Validation, error reporting & correction, as well as update of the system’s database; (e) Artificial intelligence (AI) software, for processing information in real time, identifying out-of-normal behavior and generating alerts; (f) A MicroStrategy based enterprise BI, for modeling information and producing reports, dashboards, and alerts focusing on vehicles– machinery optimal usage, as well as maintenance and scraping policies; (g) Modular structure that allows low implementation costs in the basic fully functional version, but offers scalability without requiring a complete system upgrade.

Research Design for Developing and Validating Ice-Hockey Team Diagnostics Scale

In the modern world, ice-hockey (and in a broader sense, team sports) is becoming an increasingly popular field of entertainment. Although the main element is most likely perceived as the show itself, winning is an inevitable part of the successful operation of any sports team. In this paper, the author creates a research design allowing to develop and validate an ice-hockey team-focused diagnostics scale, which enables researchers and practitioners to identify the problems associated with underperforming teams. The construction of the scale starts with personal interviews with experts of the field, carefully chosen from Hungarian ice-hockey sector. Based on the interviews, the author is shown to be in the position to create the categories and the relevant items for the scale. When constructed, the next step is the validation process on a Hungarian sample. Data for validation are acquired through reaching the licensed database of the Hungarian Ice-Hockey Federation involving Hungarian ice-hockey coaches and players. The Ice-Hockey Team Diagnostics Scale is to be created to orientate practitioners in understanding both effective and underperforming team work.

Dimensionality Reduction in Modal Analysis for Structural Health Monitoring

Autonomous structural health monitoring (SHM) of many structures and bridges became a topic of paramount importance for maintenance purposes and safety reasons. This paper proposes a set of machine learning (ML) tools to perform automatic feature selection and detection of anomalies in a bridge from vibrational data and compare different feature extraction schemes to increase the accuracy and reduce the amount of data collected. As a case study, the Z-24 bridge is considered because of the extensive database of accelerometric data in both standard and damaged conditions. The proposed framework starts from the first four fundamental frequencies extracted through operational modal analysis (OMA) and clustering, followed by time-domain filtering (tracking). The fundamental frequencies extracted are then fed to a dimensionality reduction block implemented through two different approaches: feature selection (intelligent multiplexer) that tries to estimate the most reliable frequencies based on the evaluation of some statistical features (i.e., entropy, variance, kurtosis), and feature extraction (auto-associative neural network (ANN)) that combine the fundamental frequencies to extract new damage sensitive features in a low dimensional feature space. Finally, one-class classification (OCC) algorithms perform anomaly detection, trained with standard condition points, and tested with normal and anomaly ones. In particular, principal component analysis (PCA), kernel principal component analysis (KPCA), and autoassociative neural network (ANN) are presented and their performance are compared. It is also shown that, by evaluating the correct features, the anomaly can be detected with accuracy and an F1 score greater than 95%.

Rolling Element Bearing Diagnosis by Improved Envelope Spectrum: Optimal Frequency Band Selection

The Rolling Element Bearing (REB) vibration diagnosis is worth of special interest by the variety of REB and the wide necessity of those elements in industrial applications. The presence of a localized fault in a REB gives rise to a vibrational response, characterized by the modulation of a carrier signal. Frequency content of carrier signal (Spectral Frequency –f) is mainly related to resonance frequencies of the REB. This carrier signal is modulated by another signal, governed by the periodicity of the fault impact (Cyclic Frequency –α). In this sense, REB fault vibration response gives rise to a second-order cyclostationary signal. Second order cyclostationary signals could be represented in a bi-spectral map, where Spectral Coherence –SCoh are plotted against f and α. The Improved Envelope Spectrum –IES, is a useful approach to execute REB fault diagnosis. IES could be applied by the integration of SCoh over a predefined bandwidth on the f axis. Approaches to select f-bandwidth have been recently exposed by the definition of a metric which intends to evaluate the magnitude of the IES at the fault characteristics frequencies. This metric is represented in a 1/3-binary tree as a function of the frequency bandwidth and centre. Based on this binary tree the optimal frequency band is selected. However, some advantages have been seen if the metric is changed, which in fact tends to dictate different optimal f-bandwidth and so improve the IES representation. This paper evaluates the behaviour of the IES from a different metric optimization. This metric is based on the sample correlation coefficient, detecting high peaks in the selected frequencies while penalizing high peaks in the neighbours of the selected frequencies. Prior results indicate an improvement on the signal-noise ratio (SNR) on around 86% of samples analysed, which belong to IMS database.

Multi-Temporal Mapping of Built-up Areas Using Daytime and Nighttime Satellite Images Based on Google Earth Engine Platform

The built-up area is a significant proxy to measure regional economic growth and reflects the Gross Provincial Product (GPP). However, an up-to-date and reliable database of built-up areas is not always available, especially in developing countries. The cloud-based geospatial analysis platform such as Google Earth Engine (GEE) provides an opportunity with accessibility and computational power for those countries to generate the built-up data. Therefore, this study aims to extract the built-up areas in Eastern Economic Corridor (EEC), Thailand using day and nighttime satellite imagery based on GEE facilities. The normalized indices were generated from Landsat 8 surface reflectance dataset, including Normalized Difference Built-up Index (NDBI), Built-up Index (BUI), and Modified Built-up Index (MBUI). These indices were applied to identify built-up areas in EEC. The result shows that MBUI performs better than BUI and NDBI, with the highest accuracy of 0.85 and Kappa of 0.82. Moreover, the overall accuracy of classification was improved from 79% to 90%, and error of total built-up area was decreased from 29% to 0.7%, after night-time light data from the Visible and Infrared Imaging Suite (VIIRS) Day Night Band (DNB). The results suggest that MBUI with night-time light imagery is appropriate for built-up area extraction and be utilize for further study of socioeconomic impacts of regional development policy over the EEC region.

Real-Time Land Use and Land Information System in Homagama Divisional Secretariat Division

Lands are valuable & limited resource which constantly changes with the growth of the population. An efficient and good land management system is essential to avoid conflicts associated with lands. This paper aims to design the prototype model of a Mobile GIS Land use and Land Information System in real-time. Homagama Divisional Secretariat Division situated in the western province of Sri Lanka was selected as the study area. The prototype model was developed after reviewing related literature. The methodology was consisted of designing and modeling the prototype model into an application running on a mobile platform. The system architecture mainly consists of a Google mapping app for real-time updates with firebase support tools. Thereby, the method of implementation consists of front-end and back-end components. Software tools used in designing applications are Android Studio with JAVA based on GeoJSON File structure. Android Studio with JAVA in GeoJSON File Synchronize to Firebase was found to be the perfect mobile solution for continuously updating Land use and Land Information System (LIS) in real-time in the present scenario. The mobile-based land use and LIS developed in this study are multiple user applications catering to different hierarchy levels such as basic users, supervisory managers, and database administrators. The benefits of this mobile mapping application will help public sector field officers with non-GIS expertise to overcome the land use planning challenges with land use updated in real-time.

Loss Function Optimization for CNN-Based Fingerprint Anti-Spoofing

As biometric systems become widely deployed, the security of identification systems can be easily attacked by various spoof materials. This paper contributes to finding a reliable and practical anti-spoofing method using Convolutional Neural Networks (CNNs) based on the types of loss functions and optimizers. The types of CNNs used in this paper include AlexNet, VGGNet, and ResNet. By using various loss functions including Cross-Entropy, Center Loss, Cosine Proximity, and Hinge Loss, and various loss optimizers which include Adam, SGD, RMSProp, Adadelta, Adagrad, and Nadam, we obtained significant performance changes. We realize that choosing the correct loss function for each model is crucial since different loss functions lead to different errors on the same evaluation. By using a subset of the Livdet 2017 database, we validate our approach to compare the generalization power. It is important to note that we use a subset of LiveDet and the database is the same across all training and testing for each model. This way, we can compare the performance, in terms of generalization, for the unseen data across all different models. The best CNN (AlexNet) with the appropriate loss function and optimizers result in more than 3% of performance gain over the other CNN models with the default loss function and optimizer. In addition to the highest generalization performance, this paper also contains the models with high accuracy associated with parameters and mean average error rates to find the model that consumes the least memory and computation time for training and testing. Although AlexNet has less complexity over other CNN models, it is proven to be very efficient. For practical anti-spoofing systems, the deployed version should use a small amount of memory and should run very fast with high anti-spoofing performance. For our deployed version on smartphones, additional processing steps, such as quantization and pruning algorithms, have been applied in our final model.