Alain Josué Ratovondrahona, Hanitriniaina Marielle Rakotozanany, Thomas Mahatody, Victor Manantsoa
Programming an application requires multiple people with skills and experience in that field. It will also take a lot of time with multiple steps before achieving the final result of an application. Today, developers are assisted by various tools, software, or applications based on Artificial Intelligence (AI) such as OpenAI's ChatGPT. These AI that automatically generates source code helps developers to develop applications much faster. However, although code generators are numerous and very helpful, we are not yet at the stage where we can generate a fully functional application, but just generate pieces of source code. And we don’t know yet how to understand textual descriptions of Software Requirements to generate an application directly. Or where to find data to train an AI capable of generating a functional application from textual descriptions. Therefore, we created a new architecture composed of virtual intelligent agents called SPADE BDI to create virtual developers. The virtual intelligent agents were responsible for keyword extraction, Software Requirements synthesis, and source file creation. Then we used a transformer based on pre-trained GPT-3 for source code generation. This transformer is orchestrated by a virtual intelligent agent. To solve the problem of training data, we collected and created a new dataset called WSBL. The data came from several projects developed with the Laravel Framework over 4 years. The result allowed us to have a functional application directly from a textual description. Each intelligent virtual agent played a role like a developer by analyzing textual of Software Requirements and then generating source code. With a 15% reduction in time to develop an application compared to brute development. Our new architecture allows for processing textual descriptions (Software Requirements) step by step using intelligent virtual agents named SPADE BDI and source code generation is done by a transformer based on pre-trained GPT-3 to have a directly functional application
{"title":"Human like programming using SPADE BDI agents and the GPT-3-based Transformer","authors":"Alain Josué Ratovondrahona, Hanitriniaina Marielle Rakotozanany, Thomas Mahatody, Victor Manantsoa","doi":"10.54941/ahfe1002939","DOIUrl":"https://doi.org/10.54941/ahfe1002939","url":null,"abstract":"Programming an application requires multiple people with skills and experience in\u0000 that field. It will also take a lot of time with multiple steps before achieving the\u0000 final result of an application. Today, developers are assisted by various tools,\u0000 software, or applications based on Artificial Intelligence (AI) such as OpenAI's\u0000 ChatGPT. These AI that automatically generates source code helps developers to develop\u0000 applications much faster. However, although code generators are numerous and very\u0000 helpful, we are not yet at the stage where we can generate a fully functional\u0000 application, but just generate pieces of source code. And we don’t know yet how to\u0000 understand textual descriptions of Software Requirements to generate an application\u0000 directly. Or where to find data to train an AI capable of generating a functional\u0000 application from textual descriptions. Therefore, we created a new architecture composed\u0000 of virtual intelligent agents called SPADE BDI to create virtual developers. The virtual\u0000 intelligent agents were responsible for keyword extraction, Software Requirements\u0000 synthesis, and source file creation. Then we used a transformer based on pre-trained\u0000 GPT-3 for source code generation. This transformer is orchestrated by a virtual\u0000 intelligent agent. To solve the problem of training data, we collected and created a new\u0000 dataset called WSBL. The data came from several projects developed with the Laravel\u0000 Framework over 4 years. The result allowed us to have a functional application directly\u0000 from a textual description. Each intelligent virtual agent played a role like a\u0000 developer by analyzing textual of Software Requirements and then generating source code.\u0000 With a 15% reduction in time to develop an application compared to brute development.\u0000 Our new architecture allows for processing textual descriptions (Software Requirements)\u0000 step by step using intelligent virtual agents named SPADE BDI and source code generation\u0000 is done by a transformer based on pre-trained GPT-3 to have a directly functional\u0000 application","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121431152","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Personalized Learning is an evolving trend in many schools in the United States and globally. However, an earlier study showed that personalized tutoring positively affected students' achievement. A tutor can quickly and competently evaluate students' capacities and needs and suggest appropriate instruction, resulting in students' academic performance. Studies have found that digital tools in education are efficient, such as digital tutors, digital assessments, and student-centric curricula can support student achievement similar to what is done by skilled human tutors. A PLP App developed with AI, specifically to address issues relevant to India, is presented in this paper that provides precise help to students from across the spectrum who need additional support in understanding any subject and concepts and wish to improve academic performance. This PLP App helps teachers identify gaps in knowledge and understanding of subjects among students and support them with technology-enabled tools to bridge the gap. This is done using Coherence maps between different levels of learning in concepts in specific subjects, which address gaps in learning that cannot be easily addressed in any other manner by both students and teachers. It doesn't just tailor learning, keeping the differences among learners in mind; it also shifts the weight of students' progress from the teacher and divides it between the students and teachers. The PLP App considers the conditions of Learning, such as the motivation of the student, the associated feelings of autonomy, ability, and relevance of the Learning. Setting goals and receiving feedback are essential parts of the learning process. The learning path created by the Coherence maps is a concrete, visualized, and easily understandable list of goals designed to guide students from their current level of knowledge to a higher level of competence. Self-assessment and peer review, coupled with the learning path, help students better understand their skills and increase their sense of autonomy and ownership in Learning. Students should have personal learning paths to encourage them to set and manage their academic goals. The data relating to each student is captured on an ongoing basis by the PLP App to ensure all student performance data is recorded in the system to provide most accurate understanding of the level of knowledge. The PLP software also supports teachers' plans and students' preferences by keeping past track records. Observation and monitoring of benchmarks allow the teacher to assign additional content to the student for better performance. The drop-out of students from schools in India has many reasons. They include understanding the subject or content, personal reasons, economic reasons, and many other reasons. However, it has been established by earlier studies that a significant part of the reason for drop-outs is a failure in specific courses, such as Mathematics and English
{"title":"Personalized Learning Path (PLP) – \"App\" for improving academic performance and\u0000 prevention of dropouts in India","authors":"J. Kallakurchi, P. Banerji","doi":"10.54941/ahfe1002935","DOIUrl":"https://doi.org/10.54941/ahfe1002935","url":null,"abstract":"Personalized Learning is an evolving trend in many schools in the United States\u0000 and globally. However, an earlier study showed that personalized tutoring positively\u0000 affected students' achievement. A tutor can quickly and competently evaluate students'\u0000 capacities and needs and suggest appropriate instruction, resulting in students'\u0000 academic performance. Studies have found that digital tools in education are efficient,\u0000 such as digital tutors, digital assessments, and student-centric curricula can support\u0000 student achievement similar to what is done by skilled human tutors. A PLP App developed\u0000 with AI, specifically to address issues relevant to India, is presented in this paper\u0000 that provides precise help to students from across the spectrum who need additional\u0000 support in understanding any subject and concepts and wish to improve academic\u0000 performance. This PLP App helps teachers identify gaps in knowledge and understanding of\u0000 subjects among students and support them with technology-enabled tools to bridge the\u0000 gap. This is done using Coherence maps between different levels of learning in concepts\u0000 in specific subjects, which address gaps in learning that cannot be easily addressed in\u0000 any other manner by both students and teachers. It doesn't just tailor learning, keeping\u0000 the differences among learners in mind; it also shifts the weight of students' progress\u0000 from the teacher and divides it between the students and teachers. The PLP App considers\u0000 the conditions of Learning, such as the motivation of the student, the associated\u0000 feelings of autonomy, ability, and relevance of the Learning. Setting goals and\u0000 receiving feedback are essential parts of the learning process. The learning path\u0000 created by the Coherence maps is a concrete, visualized, and easily understandable list\u0000 of goals designed to guide students from their current level of knowledge to a higher\u0000 level of competence. Self-assessment and peer review, coupled with the learning path,\u0000 help students better understand their skills and increase their sense of autonomy and\u0000 ownership in Learning. Students should have personal learning paths to encourage them to\u0000 set and manage their academic goals. The data relating to each student is captured on an\u0000 ongoing basis by the PLP App to ensure all student performance data is recorded in the\u0000 system to provide most accurate understanding of the level of knowledge. The PLP\u0000 software also supports teachers' plans and students' preferences by keeping past track\u0000 records. Observation and monitoring of benchmarks allow the teacher to assign additional\u0000 content to the student for better performance. The drop-out of students from schools in\u0000 India has many reasons. They include understanding the subject or content, personal\u0000 reasons, economic reasons, and many other reasons. However, it has been established by\u0000 earlier studies that a significant part of the reason for drop-outs is a failure in\u0000 specific courses, such as Mathematics and English","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134172490","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
ChatGPT is getting increasing attention in both academic and professional settings. Since its release, there has been a discussion on how services such as ChatGPT may change education. Many teachers have shared that they use ChatGPT to help them generate assignment prompts, questions, and lesson plans in various subject areas. Mixed opinions have been shared with regard to the quality of materials created by ChatGPT. While some teachers believe that the materials are of reasonable quality, others worry that ChatGPT may not always generate accurate or reliable information and may reproduce biases and stereotypes that exist in the data it was trained on. In this study, I explore the research question of whether ChatGPT can really replace teachers in generating high-quality assessment questions. Specifically, I compare ChatGPT-generated questions with instructor-written questions that have been used in two classes at a public research University in the US. The preliminary results show that although ChatGPT can produce logically sensible questions, the quality is not always comparable to instructor-written ones. The ChatGPT-generated questions are not specific to student misconceptions and do not align with the learning objectives instructors have in mind, which often leads to such questions being relatively obvious and easy to answer. I further discuss the capabilities and limitations of ChatGPT in generating high-quality assessment questions. This study provides insights into how we may leverage advanced AI tools such as ChatGPT to support education.
{"title":"Can ChatGPT Help College Instructors Generate High-quality Quiz Questions?","authors":"K. Lu","doi":"10.54941/ahfe1002957","DOIUrl":"https://doi.org/10.54941/ahfe1002957","url":null,"abstract":"ChatGPT is getting increasing attention in both academic and professional\u0000 settings. Since its release, there has been a discussion on how services such as ChatGPT\u0000 may change education. Many teachers have shared that they use ChatGPT to help them\u0000 generate assignment prompts, questions, and lesson plans in various subject areas. Mixed\u0000 opinions have been shared with regard to the quality of materials created by ChatGPT.\u0000 While some teachers believe that the materials are of reasonable quality, others worry\u0000 that ChatGPT may not always generate accurate or reliable information and may reproduce\u0000 biases and stereotypes that exist in the data it was trained on. In this study, I\u0000 explore the research question of whether ChatGPT can really replace teachers in\u0000 generating high-quality assessment questions. Specifically, I compare ChatGPT-generated\u0000 questions with instructor-written questions that have been used in two classes at a\u0000 public research University in the US. The preliminary results show that although ChatGPT\u0000 can produce logically sensible questions, the quality is not always comparable to\u0000 instructor-written ones. The ChatGPT-generated questions are not specific to student\u0000 misconceptions and do not align with the learning objectives instructors have in mind,\u0000 which often leads to such questions being relatively obvious and easy to answer. I\u0000 further discuss the capabilities and limitations of ChatGPT in generating high-quality\u0000 assessment questions. This study provides insights into how we may leverage advanced AI\u0000 tools such as ChatGPT to support education.","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117065695","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
With the increasing availability of digital resources for on-the-job training, competence management in the industry requires new tools to identify training opportunities for the continuous development of the skills of employees. Our emphasis is to determine which digital courses or further learning resources suit the actual employee’s competence in combination with the skills and knowledge she or he aspires to achieve. In this paper, we describe the role of ontologies and, in particular, the ESCO ontology for the development of suitable profiles for learners, learning goals, and learning resources. We describe the matching processes operating on these profiles in order to identify the training opportunities that match best the learner’s capacity and aspirations.
{"title":"Operationalising ontologies for competence management in the industry","authors":"Kurt Englmeier, Pedro Contreras","doi":"10.54941/ahfe1002951","DOIUrl":"https://doi.org/10.54941/ahfe1002951","url":null,"abstract":"With the increasing availability of digital resources for on-the-job training,\u0000 competence management in the industry requires new tools to identify training\u0000 opportunities for the continuous development of the skills of employees. Our emphasis is\u0000 to determine which digital courses or further learning resources suit the actual\u0000 employee’s competence in combination with the skills and knowledge she or he aspires to\u0000 achieve. In this paper, we describe the role of ontologies and, in particular, the ESCO\u0000 ontology for the development of suitable profiles for learners, learning goals, and\u0000 learning resources. We describe the matching processes operating on these profiles in\u0000 order to identify the training opportunities that match best the learner’s capacity and\u0000 aspirations.","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"103 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122388824","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Digitized photographs are commonly employed by archaeologists to assist in uncovering ancient artefacts. However, locating a specific image within a vast collection remains a significant obstacle. The metadata associated with images is often sparse, marking keyword-based searches difficult. In this paper, we propose a new visual search method to improve retrieval performance by utilizing visual descriptors generated from a feature pyramid network. This network is a convolutional neural network (CNN) model that incorporates additional modules for feature extraction and enhancement. The first module encodes an image into regional features through spatial pyramid pooling, while the second module emphasizes distinctive spatial features. Additionally, we introduce a two-stage feature attention to enhance feature quality and a compact descriptor is then formed by aggregating these features for searching the image. We tested our proposed method on benchmark datasets and a public vast collection of Thailand’s ancient artefacts. Results from our experiments show that the proposed method achieves 77.9% of mean average precision, which outperforms existing CNN-based visual descriptors.
{"title":"Visual Instance Retrieval for Cultural Heritage Artifacts using Feature Pyramid\u0000 Network","authors":"Luepol Pipanmekaporn, Suwatchai Kamonsantiroj","doi":"10.54941/ahfe1002933","DOIUrl":"https://doi.org/10.54941/ahfe1002933","url":null,"abstract":"Digitized photographs are commonly employed by archaeologists to assist in\u0000 uncovering ancient artefacts. However, locating a specific image within a vast\u0000 collection remains a significant obstacle. The metadata associated with images is often\u0000 sparse, marking keyword-based searches difficult. In this paper, we propose a new visual\u0000 search method to improve retrieval performance by utilizing visual descriptors generated\u0000 from a feature pyramid network. This network is a convolutional neural network (CNN)\u0000 model that incorporates additional modules for feature extraction and enhancement. The\u0000 first module encodes an image into regional features through spatial pyramid pooling,\u0000 while the second module emphasizes distinctive spatial features. Additionally, we\u0000 introduce a two-stage feature attention to enhance feature quality and a compact\u0000 descriptor is then formed by aggregating these features for searching the image. We\u0000 tested our proposed method on benchmark datasets and a public vast collection of\u0000 Thailand’s ancient artefacts. Results from our experiments show that the proposed method\u0000 achieves 77.9% of mean average precision, which outperforms existing CNN-based visual\u0000 descriptors.","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125911302","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Drone technology is prevailing in the mainstream market with its promising innovative potential across different application scenarios. While the technological capacity of drones is explored and developed, many have addressed the societal perceptions and reactions towards its use. Recent literature inclines towards more neutral if not positive perception by the general public. This paper, performed within a European Union project ADACORSA, explores the most relevant concepts for drone technology acceptance and presents a detailed overview of the survey-based research conducted in 2022. Data was collected from a total of 601 participants across Europe and ADACORSA partner countries largely from Germany, Austria, France, Greece and Turkey. To make the survey as accessible as possible, participants could take the survey in 16 different languages. The performed risk analysis showed highest level of concerns related to security/privacy in terms of misuse and invasion of private spaces. Safety and privacy concerns are perceived as equally risky. Benefits analysis on the other hand revealed general public anticipates greater economic advantages but significantly lesser societal and environmental benefits. Apart from emergencies and humanitarian aid, and purposes to facilitate services that benefit society, industrial applications curtailed most support from the general public. Highest opposition was established for hobby/recreation-related drone use, primarily from individuals who have never used a drone. The objective of this paper is both to understand general public’s acceptance towards the use of drones and to provide a nuanced overview to drone operators of which purposes are perceived as reasonable and are accepted by the general public.
{"title":"Public's Perspective on Civil Drones: Reasons to support and oppose","authors":"Vaishnavi Upadrasta, Rodney Leitner","doi":"10.54941/ahfe1002961","DOIUrl":"https://doi.org/10.54941/ahfe1002961","url":null,"abstract":"Drone technology is prevailing in the mainstream market with its promising\u0000 innovative potential across different application scenarios. While the technological\u0000 capacity of drones is explored and developed, many have addressed the societal\u0000 perceptions and reactions towards its use. Recent literature inclines towards more\u0000 neutral if not positive perception by the general public. This paper, performed within a\u0000 European Union project ADACORSA, explores the most relevant concepts for drone\u0000 technology acceptance and presents a detailed overview of the survey-based research\u0000 conducted in 2022. Data was collected from a total of 601 participants across Europe and\u0000 ADACORSA partner countries largely from Germany, Austria, France, Greece and Turkey. To\u0000 make the survey as accessible as possible, participants could take the survey in 16\u0000 different languages. The performed risk analysis showed highest level of concerns\u0000 related to security/privacy in terms of misuse and invasion of private spaces. Safety\u0000 and privacy concerns are perceived as equally risky. Benefits analysis on the other hand\u0000 revealed general public anticipates greater economic advantages but significantly lesser\u0000 societal and environmental benefits. Apart from emergencies and humanitarian aid, and\u0000 purposes to facilitate services that benefit society, industrial applications curtailed\u0000 most support from the general public. Highest opposition was established for\u0000 hobby/recreation-related drone use, primarily from individuals who have never used a\u0000 drone. The objective of this paper is both to understand general public’s acceptance\u0000 towards the use of drones and to provide a nuanced overview to drone operators of which\u0000 purposes are perceived as reasonable and are accepted by the general public.","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134235579","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Moeko Okawara, Junji Fukuhara, M. Takimoto, Tsutomu Kumazawa, Y. Kambayashi
Various machine learning (ML) techniques have been developed widely over the last decade. Especially, deep learning (DL) contributes to ML for creating a lot of structured data such as tables from unstructured data such as images and sounds. The results have led to a lot of successes in engineering, but most of their decisions and actions are hard to be explained or verified. On the other hand, as a perfectly explainable ML approach, i.e., inductive logic programming (ILP), has been used in data mining. ILP, which is based on the first order predicate logic, is one of the symbolic approaches that is useful to deal with structured data and the relations between them. In a practical sense, we can add the results generated by ILP into given background knowledge, and make the knowledge database rich. Thus, ILP becomes more important for data mining than before, and we can extract meaningful relations between the structured data. However, contrary to DL, it is not easy for ILP to perform a learning process efficiently, because we cannot make ILP processes uniformly executable in parallel on GPU. One learning process corresponds to an inductive prediction process, where training samples correspond to positive and negative examples. In the process, ILP explores hypothesis candidates while calculating a cover set that is a set of examples deduced from each candidate. Notice that from the finally obtained hypothesis, the positive examples should be deduced, and the negative ones should not be deduced with the background knowledge. The cover set is known to be uniformly calculated in relational operations, which are executed on GPU or a relational database management system (RDBMS) such as SQL. Since modern RDBMSs can not only manage memory operations safely but also execute SQL in parallel utilizing GPU. Thus, we can partially execute ILP in parallel. But the overhead of launching the procedure for each cover set calculation is heavy and we cannot ignore the significance of the total overhead. In order to mitigate this problem, we propose an extension of the algorithm for searching a hypothesis in Progol, which is one of the most popular ILP systems. Progol uses A*-like algorithm for searching a hypothesis. The algorithm incrementally refines each hypothesis candidate through adding a literal to it, calculating its cover set in order to check whether it satisfies the condition as a hypothesis. In our approach, our algorithm simultaneously performs several refinements with high possibility as a hypothesis. We call it predictive refinement. Even though the refinements may include redundant ones because the same hypothesis may be found earlier, the predictive refinement reduces a lot of overhead cost for launching the procedure of cover set calculation. Thus our algorithm can generate a hypothesis more efficiently than the conventional search algorithm. We have extended Progol to implement the predictive refinement and cover
{"title":"Efficient Inductive Logic Programming based on Predictive A*-like Algorithm","authors":"Moeko Okawara, Junji Fukuhara, M. Takimoto, Tsutomu Kumazawa, Y. Kambayashi","doi":"10.54941/ahfe1002934","DOIUrl":"https://doi.org/10.54941/ahfe1002934","url":null,"abstract":"Various machine learning (ML) techniques have been developed widely over the last\u0000 decade. Especially, deep learning (DL) contributes to ML for creating a lot of\u0000 structured data such as tables from unstructured data such as images and sounds. The\u0000 results have led to a lot of successes in engineering, but most of their decisions and\u0000 actions are hard to be explained or verified. On the other hand, as a perfectly\u0000 explainable ML approach, i.e., inductive logic programming (ILP), has been used in data\u0000 mining. ILP, which is based on the first order predicate logic, is one of the symbolic\u0000 approaches that is useful to deal with structured data and the relations between them.\u0000 In a practical sense, we can add the results generated by ILP into given background\u0000 knowledge, and make the knowledge database rich. Thus, ILP becomes more important for\u0000 data mining than before, and we can extract meaningful relations between the structured\u0000 data. However, contrary to DL, it is not easy for ILP to perform a learning process\u0000 efficiently, because we cannot make ILP processes uniformly executable in parallel on\u0000 GPU. One learning process corresponds to an inductive prediction process, where training\u0000 samples correspond to positive and negative examples. In the process, ILP explores\u0000 hypothesis candidates while calculating a cover set that is a set of examples deduced\u0000 from each candidate. Notice that from the finally obtained hypothesis, the positive\u0000 examples should be deduced, and the negative ones should not be deduced with the\u0000 background knowledge. The cover set is known to be uniformly calculated in relational\u0000 operations, which are executed on GPU or a relational database management system (RDBMS)\u0000 such as SQL. Since modern RDBMSs can not only manage memory operations safely but also\u0000 execute SQL in parallel utilizing GPU. Thus, we can partially execute ILP in parallel.\u0000 But the overhead of launching the procedure for each cover set calculation is heavy and\u0000 we cannot ignore the significance of the total overhead. In order to mitigate this\u0000 problem, we propose an extension of the algorithm for searching a hypothesis in Progol,\u0000 which is one of the most popular ILP systems. Progol uses A*-like algorithm for\u0000 searching a hypothesis. The algorithm incrementally refines each hypothesis candidate\u0000 through adding a literal to it, calculating its cover set in order to check whether it\u0000 satisfies the condition as a hypothesis. In our approach, our algorithm simultaneously\u0000 performs several refinements with high possibility as a hypothesis. We call it\u0000 predictive refinement. Even though the refinements may include redundant ones because\u0000 the same hypothesis may be found earlier, the predictive refinement reduces a lot of\u0000 overhead cost for launching the procedure of cover set calculation. Thus our algorithm\u0000 can generate a hypothesis more efficiently than the conventional search algorithm. We\u0000 have extended Progol to implement the predictive refinement and cover","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133886522","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
E. Conceicao, Mª Inês Conceição, Mª Manuela Lúcio, João Gomes, H. Awbi
In this paper the numerical study of the airflow around the occupant using confluent jets system is made. This study uses a software that considers a coupling between Computational Fluid Dynamics and the Human Thermal Modelling numerical models with the inputs from the Building Thermal Modelling numerical model. The coupling of numerical models is used to evaluate the human temperature distribution, using the Human Thermal Modelling numerical model, the airflow around the occupants, using the Computational Fluid Dynamic numerical model, and the room surrounding temperatures, using the Building Thermal Modelling numerical model numerical model. In this numerical work, developed for winter conditions, the airflow around the occupants and inside the space are evaluated for a confluents jets ventilation system, build with one exhaust and one inlet ventilation systems. The study is made inside a virtual chamber, occupied with two virtual manikins and equipped with one tables and two chairs. The air velocity, air temperature, Draught Risk, carbon dioxide concentration and air exchange rate field around the occupants and inside the space are evaluated.
{"title":"Numerical study of the airflow around the occupant using confluent jets system","authors":"E. Conceicao, Mª Inês Conceição, Mª Manuela Lúcio, João Gomes, H. Awbi","doi":"10.54941/ahfe1002941","DOIUrl":"https://doi.org/10.54941/ahfe1002941","url":null,"abstract":"In this paper the numerical study of the airflow around the occupant using\u0000 confluent jets system is made. This study uses a software that considers a coupling\u0000 between Computational Fluid Dynamics and the Human Thermal Modelling numerical models\u0000 with the inputs from the Building Thermal Modelling numerical model. The coupling of\u0000 numerical models is used to evaluate the human temperature distribution, using the Human\u0000 Thermal Modelling numerical model, the airflow around the occupants, using the\u0000 Computational Fluid Dynamic numerical model, and the room surrounding temperatures,\u0000 using the Building Thermal Modelling numerical model numerical model. In this numerical\u0000 work, developed for winter conditions, the airflow around the occupants and inside the\u0000 space are evaluated for a confluents jets ventilation system, build with one exhaust and\u0000 one inlet ventilation systems. The study is made inside a virtual chamber, occupied with\u0000 two virtual manikins and equipped with one tables and two chairs. The air velocity, air\u0000 temperature, Draught Risk, carbon dioxide concentration and air exchange rate field\u0000 around the occupants and inside the space are evaluated.","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"88 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134165180","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Oleksandr Burov, E. Lavrov, S. Lytvynova, Olha P. Pinchuk, K. Horska, Oleksii M. Tkachenko, N. Kovalenko, Y. Chybiriak
IntroductionHybrid job and learning create new opportunities and set new requirements to control a human-machine interaction. It is important to keep in mind that modern and future participants of these activi-ties can include the artificial intellect (AI-actor) as well. One of the critical features of their interaction could be the rate of the information exchange, because an AI-actor can accept and produce tasks in a quite stable rate, in contrast with a human-actor whose performance quality can very in time. As a result, their interaction needs to be adjusted in many cases from viewpoint of complexity and rate. It is supposed that the process of the information task flow should correspond an individual or moderate rate, in the best case. But according to our preliminary data (Burov, 1990, 1996), the moderate rate (even individually adapted) of perceptual and cognitive task flow was accompanied by a higher physiological strain than slow and fast ones. Because the cognitive component of the mental work becomes more and more significant both for a job and for a teaching/learning, it is useful for adaptive systems’ design to clarify if the free (“auto”) and moderate rates have the same and/or similar influence on a human performance quality (reliability and speed) and health consequences. GoalTo carry out the comparison analysis of the speed and reliability of cognitive activity by subjects performing computer tasks at a free and fixed pace, considering the physiological "cost" of such activities. Discussion of ResultsThe methodological basis of our research are models and methods for assessment a human ability to cognitive work using the computer system for psychophysiological research developed by authors. The survey included cognitive test task performance, blood pressure and heart rate before and after the test performance, as well as electropuncture diagnostics (EPD) after Nakatani (including 3 stress-points) for each subject. 47 subjects participated in experiments, 4 times per month (three times performing tests in the fix pace, one time in the free pace, each test session in a week). The duration of each test session was 3 continuous hours. Variation of the cognitive test task performance (accuracy and reliability) over the research period were studied and compared with changes of psychological and physiological indices, namely heart rate, blood pressure, vegetative stress index after Bayevsky as well as stress indices after Nakatani. It has been revealed strong increase of the stress by physiological and EPD indices and deterioration in activity (task performance time, reliability) in test sessions with fixed pace. Individual and inter-dividual variations are considered.Significance of the Proposed PresentationThe results can be applied to optimize a human and digital system interaction accounting a hu-man cognitive and psychophysiological limitations in interaction pace. The optimization goal can be to
{"title":"On the way to hybrid intelligence: influence of the human-system interaction rate\u0000 on the human cognitive performance","authors":"Oleksandr Burov, E. Lavrov, S. Lytvynova, Olha P. Pinchuk, K. Horska, Oleksii M. Tkachenko, N. Kovalenko, Y. Chybiriak","doi":"10.54941/ahfe1002925","DOIUrl":"https://doi.org/10.54941/ahfe1002925","url":null,"abstract":"IntroductionHybrid job and learning create new opportunities and set new\u0000 requirements to control a human-machine interaction. It is important to keep in mind\u0000 that modern and future participants of these activi-ties can include the artificial\u0000 intellect (AI-actor) as well. One of the critical features of their interaction could be\u0000 the rate of the information exchange, because an AI-actor can accept and produce tasks\u0000 in a quite stable rate, in contrast with a human-actor whose performance quality can\u0000 very in time. As a result, their interaction needs to be adjusted in many cases from\u0000 viewpoint of complexity and rate. It is supposed that the process of the information\u0000 task flow should correspond an individual or moderate rate, in the best case. But\u0000 according to our preliminary data (Burov, 1990, 1996), the moderate rate (even\u0000 individually adapted) of perceptual and cognitive task flow was accompanied by a higher\u0000 physiological strain than slow and fast ones. Because the cognitive component of the\u0000 mental work becomes more and more significant both for a job and for a\u0000 teaching/learning, it is useful for adaptive systems’ design to clarify if the free\u0000 (“auto”) and moderate rates have the same and/or similar influence on a human\u0000 performance quality (reliability and speed) and health consequences. GoalTo carry out\u0000 the comparison analysis of the speed and reliability of cognitive activity by subjects\u0000 performing computer tasks at a free and fixed pace, considering the physiological \"cost\"\u0000 of such activities. Discussion of ResultsThe methodological basis of our research are\u0000 models and methods for assessment a human ability to cognitive work using the computer\u0000 system for psychophysiological research developed by authors. The survey included\u0000 cognitive test task performance, blood pressure and heart rate before and after the test\u0000 performance, as well as electropuncture diagnostics (EPD) after Nakatani (including 3\u0000 stress-points) for each subject. 47 subjects participated in experiments, 4 times per\u0000 month (three times performing tests in the fix pace, one time in the free pace, each\u0000 test session in a week). The duration of each test session was 3 continuous hours.\u0000 Variation of the cognitive test task performance (accuracy and reliability) over the\u0000 research period were studied and compared with changes of psychological and\u0000 physiological indices, namely heart rate, blood pressure, vegetative stress index after\u0000 Bayevsky as well as stress indices after Nakatani. It has been revealed strong increase\u0000 of the stress by physiological and EPD indices and deterioration in activity (task\u0000 performance time, reliability) in test sessions with fixed pace. Individual and\u0000 inter-dividual variations are considered.Significance of the Proposed PresentationThe\u0000 results can be applied to optimize a human and digital system interaction accounting a\u0000 hu-man cognitive and psychophysiological limitations in interaction pace. The\u0000 optimization goal can be to ","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133003921","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Amandeep Singh, S. Samuel, Jagmeet Singh, Yash Kumar Dhabi
This study aimed to develop a non-intrusive smart monitoring system that could identify and prevent drowsy driving, reducing the risk of accidents. The study developed a system that uses video processing to measure the Euclidean distance of the eye and an eye aspect ratio (EAR) in order to detect drowsiness. The system employed face recognition to accurately identify the driver's eye aspect ratio. An Internet of Things (IoT) module used for remote assessment of the driver's drowsiness response in real-time. If the driver is in a drowsy state, the system sends an alert/warning to the driver and relevant authorities. In addition, if a crash occurs, the system sends a warning message with the location of the collision. The system was tested on 20 participants, achieving an overall eye detection accuracy of 99.98% (with glasses), 99.89% (without glasses), and a drowsiness detection accuracy of 98.05% (with glasses) and 99.05% (without glasses). This system has the potential to be implemented in a variety of driving applications, where expensive technologies are often difficult to adopt.
{"title":"Internet of Things (IoT) based Drowsiness Detection and Intervention System","authors":"Amandeep Singh, S. Samuel, Jagmeet Singh, Yash Kumar Dhabi","doi":"10.54941/ahfe1002955","DOIUrl":"https://doi.org/10.54941/ahfe1002955","url":null,"abstract":"This study aimed to develop a non-intrusive smart monitoring system that could\u0000 identify and prevent drowsy driving, reducing the risk of accidents. The study developed\u0000 a system that uses video processing to measure the Euclidean distance of the eye and an\u0000 eye aspect ratio (EAR) in order to detect drowsiness. The system employed face\u0000 recognition to accurately identify the driver's eye aspect ratio. An Internet of Things\u0000 (IoT) module used for remote assessment of the driver's drowsiness response in\u0000 real-time. If the driver is in a drowsy state, the system sends an alert/warning to the\u0000 driver and relevant authorities. In addition, if a crash occurs, the system sends a\u0000 warning message with the location of the collision. The system was tested on 20\u0000 participants, achieving an overall eye detection accuracy of 99.98% (with glasses),\u0000 99.89% (without glasses), and a drowsiness detection accuracy of 98.05% (with glasses)\u0000 and 99.05% (without glasses). This system has the potential to be implemented in a\u0000 variety of driving applications, where expensive technologies are often difficult to\u0000 adopt.","PeriodicalId":383834,"journal":{"name":"Human Interaction and Emerging Technologies (IHIET-AI 2023): Artificial\n Intelligence and Future Applications","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132674788","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：