Influenced by fast fashion, the OEM model of large variety in small quantities has become the development trend of the textile industry "Fast fashion" features fast, cheap and fashionable Taiwan has been affected by the rise of fast fashion in recent years The OEM model of "wide variety in small quantities" has become the development trend of the textile industry The primary goal of the textile industry is to understand how to receive purchase orders under this fashion trend Customer inquiries for new patterns can only be searched manually, which is time-consuming and inefficient Chairman Chen of a leading domestic textile company took over as the chairman of the "Taiwan Underwear Innovation Alliance" in 2018 He has engaged in the design and development of embroidery patterns for more than 40 years and has developed more than 30,000 embroidery patterns Whenever international corporate customers request a price quotation for a new embroidery pattern, it takes about 25 hours of "manual search" to find 1 to 2 similar patterns for quotation Therefore, the main bottleneck is how to quickly identify "embroidery patterns" Cleaning and organizing raw data takes a lot of time To build an AI model that can quickly identify and find similar embroidery patterns, a large amount of embroidery pattern data needs to be used for learning during the model development stage Each embroidery pattern requires pre-processing, including watermark removal, border removal, and pattern standardization It will take one full-time employee six months to complete image pre-processing The textile company provided a total of 30,125 embroidery patterns for AI machine learning and identification The data were annotated and divided into seven categories of patterns Improved AI accuracy through pattern recognition and learning When a customer requests a price quotation for a new embroidery pattern, sales personnel can first upload the image to the system and check which important elements need to be identified, such as style, shape, category, pattern, and size, and then select several satisfactory options from the many options recommended by AI The results are sorted and stored according to "satisfaction," and recognition results and the user's score are stored in a cloud database By recording the standards and key points of AI pattern recognition training, we can verify whether any images were left out and the reason why certain images were not selected In addition to finding similar patterns, another challenge of "embroidery pattern recognition" is "psychological level" cognition of human beings, including "different users' preferences" and "users' consideration of customers' preferences," both of which will affect selection results The user's selection results, "satisfaction" scores, and "the operator's psychological level" preferences make the AI model more accurate The development of an "AI pattern recognition system" to assist manual work allows similar patterns and solutions to be found within 1 minute, significantly improving work efficiency by 50 and improving order-taking efficiency to cater to the fast fashion industry Schematic diagram of embroidery pattern AI recognition management system Schematic diagram of embroidery pattern AI recognition results Establish the "Taiwan Textile Industry AI Pattern Recognition Service Center and Platform" This "AI Embroidery Pattern Recognition System" project will work with more textile companies and resources in the future to establish a business model for the "Taiwan Textile Industry AI Pattern Recognition Service" Introducing this AI recognition system to the upstream and downstream of the industry chain will jointly improve the technological level, operational efficiency and international competitiveness of Taiwan's textile industry

Empower addresses with spatial coordinates to help drive the 'Open Data' policy In recent years, the government has been promoting 'Open Data' hoping that the openness of data will facilitate inter-agency data flow, enhance administrative efficiency, meet public needs, and strengthen public oversight of the government Among them, transportation data is closely related to daily life, often reported by the public with the incidents specifying obvious local landmarks or addresses there have also been public feedback about the traffic reports on police radio that lacked actual coordinates Introducing these addresses, which were originally without spatial attributes, into the geographical coordinate system is one step toward 'Smart Spatial Decision Making' However, unstructured addresses, without manual intervention to improve the inconsistency of address formats, do not yield high location accuracy, necessitating an improvement in data quality and usability to unlock the potential applications of open data This further aids in policy promotion and widespread application to different sectors including tourism, employment, birth and adoption Unregulated and diverse writing styles of addresses lead to low location accuracy Address Locator is jointly developed by SongXu Information Co, Ltd and YanDing Intelligent Co, Ltd GOLiFE as a 'stand-alone address locating software' providing single or batch address location services To imbue address data with spatial attributes, the core technology of Address Locator involves 'Address Parsing' and 'Address Location' in two stages Initially, 'Address Parsing' distributes the addresses aimed for positioning according to administrative region hierarchy keywords provincecity, townshipdistrict, village, roadstreet, alley, lane, number subsequently, 'Address Location' matches the split addresses with the parent address to obtain the location level and corresponding coordinates However, in the actual business integration process, since address sources are maintained separately by different authorities, a lack of consistent standards remains a common issue Problems include special characters at address examples in specific regions, omitted administrative units, repetitive administrative hierarchical keywords, special street-alley segments, mismatch in Chinese numericals vs Arabic numerals, and non-current addresses leading to complex address formats that are difficult to accurately split Establishing an address tokenization model, achieving precise location alignment To effectively handle various messy address formats and alleviate the difficulties in location alignment for the existing Address Locator, AI and Natural Language Processing technologies are implemented for 'Address Normalization' and a 'Chinese Tokenization Tool' to optimize existing address location capability 'Address Normalization' addresses the issues of missing keywords, variant character forms, and missing administrative areas whereas 'Chinese Tokenization Tool' helps resolve 'split errors' caused by special address formats, preventing unsuccessful positioning Successful address parsing through AI tokenization technology In the past, while handling address location services, manual preprocessing for data standardization was required, hence it was not solely marketed as a product, but included in project plans that offered address location services However, after incorporating address normalization and AI tokenization technology, it has become a complete product, significantly reducing the time users spend on manual adjustments and achieving the intended location accuracy Furthermore, the AI-enhanced Address Locator is now introduced on the SongXu Information Co Ltd website, including product descriptions and official listings After four months of testing and modifications, AI technology was successfully incorporated into the existing address location product From selecting the tokenization tools, building the corpus, training the model, and interfacing with product features, to complete test planning, collection from 'Government Data Open Platform' and 'Taichung City Government Data Open Platform,' including over 62 datasets and more than 300,000 addresses, achieving a complete match rate of 9008 and a fuzzy match rate of 98, greatly surpassing the original product in match rates and processing time To promote AI technology applications in the information services sector, the AI-enhanced address location service is positioned as a new solution and showcased on the SongXu company website starting from product function introductions, explaining address regularization methods and address location features subsequently, guiding potential customers to envision applicable scenarios including decision analytics, precision marketing, and other applications The product will aid various sectors’ data by assigning spatial information to addresses, delving into the context and trends of data in two-dimensional space Address Location Solution Providing spatial coordinates for attractions, intersections, and points of interest Successful development and implementation of AI-enhanced products in companies focused on smart transportation systems in the domestic market revealed that, while effectively solving address location issues, they also recognized that descriptions of spatial information, beyond addresses inclusive During their progress, integrating AI more broadly into 'Entity Recognition' is set to be an important future application not limited to address location In an era of information overload, collecting data is straightforward identifying keywords of interest is key Future development directions aim to optimize these products and create more business opportunities「Translated content is generated by ChatGPT and is for reference only Translation date：2024-05-19」

Business email fraud causes over ten billion USD in losses annually Business Email Compromise BEC has become a global threat, causing losses of over ten billion dollars every year Criminal organizations target victim companies in various ways, infiltrating and lurking over long periods to steal information, eventually orchestrating scams to induce victims to make payments or transact with the wrong parties In such crimes, key fraudulent emails often appear indistinguishable from ordinary business or personal correspondence because they match business contexts and daily transaction behaviors BEC messages may not even contain URLs or attachments that could be detected by technical security measures, rendering antivirus and click-prevention strategies ineffective With email fraud being rampant, effective prevention of such threats has become a top priority A domestic biomedical manufacturer, one of the government's five major innovation industries and a recipient of the 13th Startup Business Award from the Ministry of Economic Affairs, features high innovation and high risk in the biotechnology sector Handling sensitive information related to new drugs, experimental materials, or market development, and even confidential personal data tied to medical and clinical trials, the company faces substantial threats from BEC scams They hope to leverage AI's interpretation capabilities for more accurate and comprehensive alerts against malicious emails, ultimately enhancing productivity and avoiding scams Utilizing AI recognition to preemptively deter threatening emails, effectively boosting corporate productivity 'Artificial Intelligence-assisted BEC transaction intent perception' is a feature developed by NetzEngine Information Software Co, Ltd and Dupont Digital Security Ltd, capable of identifying emails with transactional intents and incorporating them into NetzEngine's MailGates email behavior analysis module to detect threatening messages and improve the detection accuracy of suspicious threat emails This case uses two functionalities from the aforementioned AI technologies, 'Email Fraud Protection by Mail Header Security Policy' and 'Email Fraud Protection by Email Behavior Analysis Policy' Openfind MailGates' 'Mail Header Security Policy' feature In MailGates' 'Email Fraud Protection' features, the 'Mail Header Security Policy' can be adjusted For example, all emails from hotmailcom must have a correct From header, but the Reply-to header will be blank, this is the correct format for Hotmail emails, the same applies to Gmail If the email comes from another source, however, it should adhere to the filtering rules recommended by MailGates Both From and Reply-to should come from the correct and same domain, otherwise, it is likely to be a fraudulent message The third rule in the image represents all emails from openfindcomtw should use the aforementioned default rules for inspection If an email does not comply with the Mail Header Security Policy as per this setup, users will receive a 'Email Fraud Warning' notification on the subject line, helping to prevent BEC email fraud Openfind MailGates' 'Email Behavior Analysis Policy' feature, can set 'Protection Level' Openfind MailGates' 'Email Behavior Analysis Policy' feature, can set 'Operational Actions' In MailGates' 'Email Fraud Protection' feature, the 'Email Behavior Analysis Policy' can be applied as needed This feature's design and settings might seem complex for average users, who could opt for the 'Intelligent Detection' method instead, simply choosing among 'Loose, Standard, Strict' levels The system will determine the actual settings for these levels based on recently collected feedback data More skilled managers can use the 'Custom' mode, to set all behavior analysis functionalities in detail For example, by listing commonly impersonated domain names under 'Similar External Domains', the system will automatically consider similar but unequal domains, intended to impersonate and deceive users, as higher threat sources This feature allows users to set alerts for such emails, including using title and content warnings, and through behavior analysis, if the system deems the email likely to be a BEC fraud, it will clearly prompt users to be vigilant 'Email Fraud Protection by Mail Header Security Policy' and 'Email Fraud Protection by Email Behavior Analysis Policy' are actually planned and designed functionalities incorporated into the MailGates email protection system, combined with the aforementioned AI research outcomes, all MailGates users will be able to utilize these two functions against BEC scam emails For corporate clients, 'missed threat emails' represent the most significant information security threat and the aspect most needing improvement With the adoption of NetzEngine's BEC smart sensing mechanisms, they can immediately and effectively reduce the threats of BEC scams within the unit, avoiding scams and boosting corporate productivity Comprehensively guarding客户 client email security, expanding the value of the domestic information security industry NetzEngine Information Software Co, Ltd, in collaboration with AI startup Dupont Digital Security Ltd, adopts NLP and more specialized threat analysis technologies, capable of intelligently perceiving emails with transaction intentions and increasing the interception rate of BEC emails This not only enhances client value and maintains their cybersecurity capabilities, but as a leader in domestic email security, Openfind NetzEngine Information will continue to develop solutions for email and messaging communication security In the future, NetzEngine plans to integrate BEC protection and APT sandbox defense technologies, continuously expanding into a derivative product line Advanced Threat Protections By addressing customer needs against the continually increasing cybersecurity threats with a robust and comprehensive protection solution, it aims to deliver greater value to customers, while also expanding the domestic information security industry's problem-solving options and enhancing its value In the current high-end cybersecurity protection market, many clients can only use products from foreign manufacturers These products' designs, usage processes, and, most crucially, the sources and processes of samples or policy settings often do not suit the specific requirements of domestic government agencies or enterprises Therefore, through products and services offered by NetzEngine, enterprises will be helped to safeguard email security「Translated content is generated by ChatGPT and is for reference only Translation date：2024-05-19」

Life is tough for lettuce village farmers who are at the mercy of the elements Global market trends are volatile In terms of agriculture, it must compete with cities for land and water, and also face other crises, such as mitigating climate change and protecting natural habitats In particular, climate change makes it more difficult for farmers to plan cultivation in traditional ways For organic lettuce exporters, they have to overcome problems such as climate and pests to ensure that the lettuce can meet the standards of overseas customers This is an issue that the industry is facing Difficulties and needs of Taiwan Lettuce Village Although Taiwan Lettuce Village currently uses an internationally certified standard process GGAP for cultivation, and also uses the "Intelligent Agriculture Management System" developed by Info-Link Services for cultivation management, it still faces the dilemma of not being able to control crop yields and quality due to climate abnormalities Efforts to solve pests and production problems in recent years not only consumes labor, but also doubled the use of pesticides However, using cultivation and production models of the past will cause the industry to stagnate or even face elimination Therefore, it hopes to add value through AI, and make the lettuce village can be more information-based, intelligent, analytical, and predictive in cultivation, so as to expand the industryrsquos exports and diversify the industry's development in the future Current Demand of the Lettuce Village The agriculture industry in Taiwan Lettuce Village currently only implements "information management" Even though it has the concept of data application, there are no implementation methods and direction, and fields are still manually inspected and the dosage of pesticides is determined based on experience Since crop production varies due to environmental factors each season, the accumulated temperature conditions required for lettuce growth can be estimated by comparing crop yield and harvest date based on historical meteorological data, thereby establishing an accumulated temperature calculation module to estimate the cultivation schedule, allows the system to automatically analyze and make prediction based on the current temperature and humidity of the overall environment During the crop harvesting period, it assists field personnel in optimizing their work, reducing the need for daily inspections to determine when to schedule the next task Agricultural information system AI allows lettuce to grow smoothly "The stability of the cultivation environment" plays an important role in the growth process of crops Understanding the growing conditions of crops can greatly increase production and maintain a certain level of quality Combined with "smart equipment in the field" and "linebot," field management can be carried out and warnings can be received at any time, allowing managers to respond quickly to reduce potential losses, and assist in disease prevention, growth period, and harvest prediction It can be further integrated with data from the Central Weather Administration to establish a "cultivation database," and conduct agricultural analysis through data collection, such as fertilizer dosage planning, analysis of lettuce growth days in different months, analysis of the quality and weight of lettuce output based on temperature, and even disease prevention predictions Comparison of differences before and after digitization Compiled the cultivation data collected by the Lettuce Village from field equipment and external data , such as temperature, humidity, sunlight, and farmland fertility, and applied the data in four aspects, including 1 establishing crop progress and growth obstacle information, analyzing the temperature ranges that are suitable for growth and hinder growth, importing open data real-time and future weather forecast data to establish forecast standards, and using weather sensing equipment for field monitoring, in order to achieve real-time warning notifications and preventive effects 2 Utilize cultivation data for growth predictions, in order to achieve the goal of estimating harvest date 3 Using mobile phones for weather monitoring achieves the goal of real-time control and adjustment of field operations, allowing Lettuce Village to effectively manage manpower, material costs, and crop quality 4 Compiled farmland fertility data to provide the fertilizer ratios for suitable for farmland and reduce the frequency of fertilization, improving farmland fertility while improving the overall environment Description of Data Applications The system will continue to be optimized and promoted it to more units The "Intelligent Agriculture Big Data Application Platform" allows farmers in Lettuce Village to no longer be limited to the traditional agricultural business model, achieve systematic cultivation and production management, and standardize specifications to improve quality, stabilize output, and reduce labor consumption and material costs Improved pest and disease detection accuracy from 80 to 100 In the future, we hope to increase the accuracy of pest and disease detection, so that farmers can monitor the status of crops in real time, making the system more complete We also hope to apply this system model to more crops, and allow more farmers to consistently grow high-quality crops at low cost through government promotion

Avoid unnecessary invasive biopsy examinations, all thanks to the professional judgment of radiologists Medical imaging recognition is a crucial task for radiologists, who must make professional judgments based on patient's examination data Upon identifying a tumor, it must be determined whether it is cancerous feasible methods include 'non-invasive medical imaging' and 'invasive biopsy examination' The advantage of biopsy examinations is that they can provide very accurate diagnoses, however, as they are invasive, doctors and patients will avoid this method if the probability of severe conditions is low One of the responsibilities of radiologists is to provide related professional judgments to aim for the most ideal situation Radiologists are overwhelmed, standards for judging tumor benignity or malignancy fluctuate, exposing a crisis in medical quality With the popularization of medical imaging examinations and the gradual flourishing of preventive medicine concepts, the burden on radiologists has been increasing A single doctor needs to handle multiple patients at once, and under conditions of long working hours and multiple patients, the standard for judging the benignity or malignancy of tumors based on images can fluctuate, resulting in patients not receiving optimal medical quality Tatung Science and Technology National Taiwan University Develops 'AI Medical Imaging Recognition System', Introduced to Medical Institutions, Effectively Enhances Tumor Interpretation Efficiency and Accuracy Tatung World Technology Co, Ltd and the Research Team of the Institute of Biomedical Electronics and Informatics at National Taiwan University jointly developed the 'AI Medical Imaging Recognition System' The trained model can interpret the benignity and malignancy of breast X-rays, with an accuracy rate reaching 85 This system has been introduced to the radiology department of a central medical institution for POC verification, helping to reduce the workload of radiologists and the waiting time for patients' examination reports Breast Tumor AI Interpretation System Diagram In the future, the correlation between the breast imaging report, data system BI-RADS grading, and AI benignmalignant interpretation will also be further defined, transforming the imaging interpretation from a binary system to a probabilistic BI-RADS grading This will assist the institution in establishing a common standard and enhance the efficiency of cooperation across different medical specialties Benefits of Introducing AI Identification System Replicating successful models, laying the foundation for the AI medical imaging big data era The development model of this AI Medical Imaging Identification System can be applied to different types of medical imaging, including computed tomography scans, ultrasound imaging, etc and can integrate natural language processing capabilities with pathology analysis reports, laying the foundation for the AI medical imaging big data era「Translated content is generated by ChatGPT and is for reference only Translation date：2024-05-19」

IBM's supercomputer Watson can predict when employees are likely to resign, with an accuracy rate of 95, saving IBM up to 300 million a year in retaining employees Moreover, through cloud computing services and modernization, IBM has streamlined 30 of personnel costs, allowing the remaining employees to earn higher salaries and engage in more valuable work So, in Taiwan, how can we ensure that 'employees who stay can receive higher salaries and perform more valuable work' The key lies in the 'competence setting' for each position According to the 'iCAP Competency Development Application Platform' established by the Ministry of Labor's Workforce Development Agency, every position has its main responsibilities, work tasks, behavioral indicators, work outputs, knowledge, skills, and attitudes Only by establishing 'competency' for each position can enterprises effectively apply this in employee recruitment, education and training, and performance management Without this, not knowing what employees should do is like groping in the dark, which can pose risks to business operations Competency Benchmark Example Currently, on the 'iCAP Competency Development Application Platform', there are 872 established competency benchmarks, including 553 items completed by various ministries This includes 253 items from the Ministry of Labor and 66 items from the Ministry of Education If companies want to establish their own 'competency benchmarks', they need to search for reference materials on the 'iCAP Competency Development Application Platform' Suppose a company wants to recruit 'sales' personnel but doesn't know what 'sales personnel' should do they should first search for 'sales personnel' as shown in the figure below Searching for 'sales' on the 'iCAP Competency Development Application Platform' You can find that there are 18 types of sales personnel At this point, the company needs to go through each one, check, read, and organize into the 'competency benchmarks' they need however, if we search what should be a common position in any company, 'general affairs', the result is unexpectedly zero items Searching for 'general affairs' on the 'iCAP Competency Development Application Platform' As seen above, although the 'iCAP Competency Development Application Platform' established by the Ministry of Labor's Workforce Development Agency can solve some of the 'competency benchmarks' for positions, the division of labor within each company is different, and some positions might not be found on the 'iCAP Competency Development Application Platform' Secondly, in small and medium enterprises, there are often 'multi-skilled workers', meaning many job responsibilities are on a single employee For example, in small enterprises with less than 30 people, usually, accounting, general affairs, and HR are handled by the same person If you want to establish competency benchmarks for this person, you have to search separately for 'accounting', 'general affairs', and 'HR', and then integrate these three types of job competencies, which is often time-consuming and ineffective This 'Small and Medium Enterprises AI Competency Evaluation System' aims to let 'people fully utilize their capabilities', by introducing AI to more accurately establish basic competency standards for employees, and to track their competency performance at any time Competency models are all generated and adjusted manually, which is time-consuming A domestic exporter of screws, nuts, fasteners, etc, had all its competency models generated and adjusted manually The execution process was time-consuming and insufficient to meet company needs due to personnel changes, such as previously, Qiao Mai Enterprise had specialized 'production control personnel', but after this personnel resigned, this job had to be done by other employees, meaning other employees' competency models needed to be adjusted immediately Or if the company needed to set up a development department due to future development, but previously no one had relevant experience, not only did they not know how to select from within, but also did not understand how to describe on a recruitment website to find the talent they really wanted Besides, the CEO of this company has always been troubled by internal performance management Due to the lack of precise standards and systems to measure employee performance, the results of each performance assessment did not accurately reflect the true performance of the employees, forming assessment blind spots and unable to identify truly deserving employees Thus, it is hoped that with the AI competency evaluation system, the necessary competencies for the development department can be immediately clarified, as well as how recruitment and performance appraisals should be conducted, so as to effectively solve the pain of unclear responsibilities and inaccurate assessments within the company Thus, its benefits are significant AI Competency System Establishment X Deep Learning This 4-month HR field competency system project has a clear execution direction, but the introduction of explanatory models such as Seq2Seq, Deep Keyphrase Generation, Tf-IDF keyword extraction algorithms, and PageRank are new attempts in the HR field During the process, open-source big data architecture is used for natural language processing to complete Word2Vector and index, and inverted index to establish keyword weight and relevance Due to the inability to process like image data with continuous numbers, it is necessary to simplify the feature values with related keywords such as skills, knowledge, and job categories Basic steps are briefly described as follows 1 Establish a Propagation model using Google's long-used LTR mixed Pointwise recommendation engine 2 months 2 Establish a Back Propagation model 2 months, adjust the hyperparameters of the loss function 3 Adjust the hyperparameters of the CF model 4 Establish a human-machine collaboration mechanism to obtain more data to feed the Model 5 Repeat the above steps During the development process of the competency model, Lianhe Trend Co, Ltd and Weiguang International Information Co, Ltd held multiple discussions, believing that there are interconnections between competencies After establishing the knowledge graph, further upload the competency scale to the Neo4j graph database for processing complex relational data structures with excellent performance Currently, 500 competency scales have been uploaded to the Neo4j relationship analysis platform Using python for wor2vector natural language analysis In addition to describing a position with a tensor after word2vector, finding out the appearance of this position's knowledge graph, according to this knowledge graph, one can understand the relevance between different positions and the similarity performance of their dimensions Finally, this knowledge graph is used to establish the company's 'competency model' and train it with deep learning AI Competency Evaluation System Interface In the future, in addition to establishing their own competency models, companies can also be opened to end-users Individuals can analyze their own competency performance to understand their possibilities for job change and their market value, as well as identify skills needing enhancement If companies respond to this knowledge graph, they can develop cross-industry products in the future 1 Short-term Analyze the competency scales iCAP, iPAS published by the government with natural language and keyword models, and cooperate with unsupervised learning to establish 'Native Competency Base Unit Models' 2 Medium-term Tailor-made exclusive competency models for enterprises Based on the existing 'Native Competency Base Unit Models', experts use supervised learning to train the individual company's 'Distributed Derivative Competency Models' 3 Long-term Establish 'Reinforcement Learning' models, incorporating employee career cognition and planning Competency model recommendations, comparable to professional human resource consultants Through the dynamic learning of the competency knowledge graph through unsupervised learning, individual companies' competency models are quickly established Internal human resources personnel or external professional HR consultants can then use the generated competency models to assess and apply aspects of talent recruitment, competency inventory, performance management, and education and training The system will automatically suggest competencies to be strengthened according to the company's existing job structure, including related knowledge, skills, and attitudes Through the continuous introduction and training of data, the system learns the employer's actual view of the model for that profession and feeds back to the cloud competency scale, completing the dynamic learning of the knowledge graph through transfer learning In the future, it can be comparable to professional HR consultants, thereby rapidly assisting many cross-disciplinary or technologically diverse companies in training employee competencies「Translated content is generated by ChatGPT and is for reference only Translation date：2024-05-19」

In response to current trends, digital learning and mobile educational materials have attracted widespread attention With rapid technological advancements, effectively nurturing professionals who can 'adapt to developmental changes' is a critical concern that many businesses continually consider Over recent years, various enterprises have progressively integrated 'digital learning' into employee training programs to enhance educational outcomes, thus bringing 'digital learning' and 'mobile educational materials' into the limelight Outsourced narration is costly and cannot handle large volumes of demand Differences in the digital educational material production process before and after the implementation of the AI voice synthesis system Strategic Breakthrough Corporation of Taiwan has assisted companies in converting many seminars, physical courses, and training events conducted by public sectors into digital materials in the past years However, during the conversion process, it required inviting teachers, finding and renting filming locations, and post-production of recordings and videos During recording, issues such as speakers' nervousness, discomfort in front of cameras, or mispronunciations might lead to poor recording quality or constant retakes Though there was an option to provide customer-specific educational material narration, the outsourcing costs were high and could not handle the demand efficiently Therefore, there was a hope to introduce AI speech synthesis technology and develop an 'Intelligent Voice Synthesis Module' to instantly convert text on slides into natural, human-like voice files, thus saving on narration costs Realistic Intelligent Voice Synthesis Module, providing a diversified selection of voices AI Voice Synthesis Module Illustration Strategic Corporation of Taiwan collaborated with the AI technology team, Magic Cube Digital Ltd, using Tacotron2 combined with WaveNet and Tacotron features Characters are embedded into Mel-scale spectrogram plots, then a modified WaveNet model acting as the vocoder synthesizes waveform in the time domain from these spectrograms, finally developing an MOS Mean Opinion Score for voice quality evaluation that approximates human-like intelligent voice synthesis modules This AI Intelligent Voice Synthesis Module, after being tested by testers using the MOS voice quality evaluation standard, received a score of 43, meeting the initial project target score of 421 and surpassing WaveNet's score of 408, thereby demonstrating exceptional effectiveness AI Intelligent Voice Synthesis Module, reducing costs and increasing profits, will effectively enhance Taiwan's digital learning industry environment Costs have been significantly reduced after the implementation of the AI voice system, and profits have increased relatively This AI Intelligent Voice Synthesis Module not only reduces the cost of producing digital educational materials but also solves the difficulties faced by Taiwan's industry, government, and academia in spreading digital educational materials It can effectively enhance the efficiency of customers in producing digital teaching materials, significantly reduce labor shortages, and cost structural risks, and improve profitability Strategic Corporation of Taiwan will also continue to develop the 'Intelligent Transcription Module' and introduce Robotic Process Automation RPA to replace the current manual processes, such as captioning, dubbing, and file conversion in the production of digital educational materials, assisting in the transformation and enhancement of the domestic digital learning industry「Translated content is generated by ChatGPT and is for reference only Translation date：2024-05-19」

Taiwan is located in the subtropics and has a diverse geographic environment that is very suitable for growing fruit Bananas and pineapples were once extremely popular export commodities that we are proud of However, farmers in consuming countries gradually obtained the excellent seeds of Taiwanrsquos fruits, and were able to grow the same quality fruit but at a more affordable price, causing our fruit exports to face a major crisis At present, although Taiwan's fruits such as mango and guava still have certain competitive advantages, if they fail to make further progress compared with other countries, they will still encounter the same problem over time and cannot be ignored Fruit quality and brand value are the only ways for Taiwan's fruit industry to remain competitive internationally Fruit screening is the main link in fruit production and marketing that determines quality Currently, the industry is highly dependent on aging rural manpower, resulting in rising fruit screening costs due to labor shortage and making it extremely difficult to maintain stable yield Therefore, the automation of fruit screening work has become a very important and urgent issue Professor Chi-Chun Lee at the Department of Electrical Engineering of National Tsing Hua University led a team to develop an automatic fruit screening system that combines cameras, conveyor belts, and AI The system currently has an accuracy reaching 93 One production season can increase the output value of mango by NT17 billion With the gradual development of the AI system, the accuracy is expected to improve in the future, and the same system can also be applied to other fruits, further promoting traceable fruit and driving the technological upgrading of Taiwan's fruit industry Fruit screening relies heavily on scarce manpower, and the aging of the rural population makes the situation even worse Professor Chi-Chun Lee learned about the fruit industryrsquos dilemma from his classmate Yu alias, who had studied together in the United States Yu is the young second-generation successor of one of Taiwan's largest fruit import and export companies According to Yu's observations in the industry over numerous years, Taiwan's fruit production and export usually generated good profits at first, but after fruit farmers in the consuming countries obtained the seeds, they will often attempt to grow the fruit locally to reduce costs and obtain greater profits If Taiwanese fruits cannot surpass the products of fruit farmers in consuming countries in terms of quality or brand value, they will be eliminated because competitors' costs are indeed lower Fruit screening is used to divide fruits according to quality If they cannot pass the minimum specification, they will be discarded as waste products In practice, the work of screening fruits will be carried out by farmers' goods yards and distributor' packaging yards respectively However, if it is not properly handled by the collection freight yards and the packaging yards do not do a good job in sampling in the early stage, it will result in a loss for distributors and cause 30 of AA grade fruits to be eliminated This job relies heavily on experienced fruit screeners More experienced fruit screeners can not only control the quality and reduce the chance of fruit damage in the fruit screening process, but also have the ability to pick out about 10 more A grade fruits, which adds great value What worries the industry is that experienced fruit screeners are gradually decreasing due to the aging population in rural areas, making them a very rare resource Such rare human resources are often in high demand during busy farming periods Farmers or distributors who fail to hire experienced fruit screeners have to settle for less experienced one, taking on the risk of additional losses and paying greater costs The most unfortunate situation suffering a loss of 30 mentioned above Fruit screening is an important process in the later stages of fruit production when packaging and selling Failure to properly control quality will result in huge losses AI is very suitable for assisting in fruit screening, but it is difficult to obtain data sets After understanding Yu's difficulties, Professor Lee found that this was a problem that could be solved using AI - although fruit screening relies heavily on experienced fruit screeners, it is a highly repetitive task Handling repetitive tasks with a large amount of data has always been a strength of AI However, the first problem appeared even before research and development work started Which fruit do we start with First of all, a suitable fruit must reach a certain export volume, and the fruit must still have considerable room for growth For some fruits that lack international competitiveness, such as bananas and pineapples, companies no longer have the ability to invest more funds to purchase equipment, let alone sponsor RampD or assist the RampD team in experiments When you have an idea, you need to pick up the pace and put it into practice as soon as possible Therefore, Irwin mango, which still has a certain advantage in terms of scale, was selected as the first experimental subject of the automatic fruit screening systemThe first step after harvesting mangoes is to screen the fruits for the first time at the goods yard After the fruits are screened, they are sent to the packaging yard for fumigation and disinfection, and preparation for sale or loaded into containers for export However, exporters with a deeper understanding of the target market will have stricter quality requirements and will often screen the fruit again to ensure the quality of the fruit before fumigation at the packaging site Since employees at the goods yard are paid based on the number of mangoes screened rather than on the quality of the mangoes, they focus on quantity when working As a result, to ensure the quality of the selected fruits, the subsequent packaging factory has to screen the fruit again, increasing labor The solution seems simple and clear - A camera, machine conveyor belts for grading and sorting, and an AI that can distinguish the quality of mangoes from their appearance are all that are needed to achieve automatic fruit screening However, the hard part is how can AI distinguish the quality of mangoes Thatrsquos right, you must start by establishing a training data set In order to create the data set, Professor Lee's team established a website that allows anyone to upload photos of mangoes and rate them Once the data sets are refined, they can be used to train AI The fruit screening machine developed by Professor Lee's team uses AI image recognition to select the best looking mangoes The accuracy of the trained AI reaches 93, which can increase the output value by NT17 billion in one season In 2019, the assistance of the Industrial Development Bureau now the Industrial Development Administration of the Ministry of Economic Affairs and AI HUB accelerated the verification of the technology Professor Lee's team accumulated 100,000 entries of data during the 2-month empirical period, and the accuracy of the trained AI reached 93 This is far higher than the manual screening accuracy of 70, resulting in a clear difference in quality In terms of export value, the output value of mango is expected to be increased by NT17 billion in one season It can also reduce labor costs by NT1866 million and avoid the seasonal labor shortage problem mentioned above In addition, since it is no longer necessary to screen the fruit once at the goods yard and packaging yard each, it also reduces losses caused by human error in the fruit screening process When the technology becomes more mature, the same system can be applied to other fruits exported by Taiwan, such as wax apple and guava, in the future, taking Taiwan's fruit industry to the next level Since it is AI, accuracy can be improved through continuous training, and continuous adjustment of algorithms and cooperation with equipment manufacturers can significantly improve production capacity In addition, Professor Lee is also organizing the AI Cup competition with the sponsorship of manufacturers and the government, allowing more teams to use the same data set to continue to develop the algorithm, in hopes of facilitating further cooperation with companies that are interested Irwin mango grade identification system on AI HUB Professor Lee's team hopes to use the power of AI to achieve complete traceability of fruits from production to packaging and transportation, thereby increasing the brand value of Taiwan's fruits Besides hoping to allow Taiwan's fruits to seize a place in the fiercely competitive foreign markets, with high-quality supply, Taiwan's fruits can also shine internationally and become the pride of Taiwan Taiwan's fruits still have certain competitive advantages in the international market, but they also face competitive pressure from fruit farmers in consuming countries as they are exported Easily save NT1866 million per mango season and significantly improve quality nbsp nbsp nbsp nbsp nbsp nbsp