Ebook: Machine Learning and Artificial Intelligence
Machine learning (ML) and intelligent systems are now comprehensively applied for the solving of practical problems. Emerging techniques such as big data analysis, deep neural networks, AI, and IoT have been adopted and integrated into the development and application of machine learning and intelligent systems, and their wide application in industry, medicine, engineering, education and other mainstream domains have made them a part of everyday life.
This book presents the proceedings of MLIS 2023, the 5th International Conference on Machine Learning and Intelligent Systems, held as a hybrid event from 17-20 November 2023 in Macau, China. This annual conference aims to provide a platform for a knowledge exchange of the most recent scientific and technological advances in the field of ML and intelligent systems, and to strengthen links within the scientific community in related research areas. A total of 80 submissions were received for the conference, of which 20 papers were selected for presentation and publication in these proceedings following a rigorous peer-review process. Papers were assessed on originality, scientific/practical significance, compelling logical reasoning and language, and the selected papers cover a wide range of topics, and provide innovative and original ideas or results of general significance in the field of ML and intelligent systems.
Providing a current overview of developments in the fields of machine learning and intelligent systems, the book will be of interest to all those working in this field.
Over recent decades, Machine learning (ML) and intelligent systems have become an entrenched part of everyday life and are now comprehensively applied for solving practical problems. Emerging techniques such as big data analysis, deep neural networks, AI, and IoT have been adopted and integrated into the development and application of machine learning and intelligent systems. The broad application of machine learning and intelligent systems in the fields of industry, medicine, engineering, education and other mainstream domains greatly benefit our work and life.
The International Conference on Machine Learning and Intelligent Systems (MLIS) is an annual conference that aims to provide a platform for knowledge exchange of the most recent scientific and technological advances in the field of ML and intelligent systems, and to strengthen links in the scientific community in related research areas. MLIS 2023 is to be held in Macau, China as well as online via MS TEAMS from 17–20 November 2023. The plenary session of MLIS2023 will include Keynote Speeches, Invited Speeches, Poster Presentations and Oral Presentations. Scientific topics covered at MLIS 2023 include data mining, image processing, neural networks, natural language processing, video processing, computational intelligence, expert systems, human-computer interaction, and deep learning.
This book contains the proceeding of the 5th International Conference on Machine Learning and Intelligent Systems (MLIS 2023). Rigorous peer review was conducted on each paper by at least two reviewers, and 20 accepted papers were selected from more than 80 submitted full papers, based on the review results on originality, scientific/practical significance, compelling logical reasoning and language. Comments from active reviewers like Prof. S.P. Thirumuruganandham, Dr. Manuela Freire from the University of Coimbra, Dr. Yan Yang from Cornell University, Dr. Seethalakshmi V from Anna University, Dr. Imran Iqbal from Peking University, among others, engaged authors to respond to pertinent questions for the camera-ready versions of their papers. We believe the 20 selected papers included here will provide innovative and original ideas or results of general significance in the field of ML and intelligent systems.
On behalf of the conference organizers, we would like to take this opportunity to express our sincere thanks to the Guest Editor and reviewers for their tremendous efforts and dedication to the conference and to the authors for their relevant contributions to the conference, as well as to all the colleagues from the publishers, IOS Press, for their support and their tireless efforts towards the publication of the conference proceedings. We believe that with their earnest support and contributions, future MLIS Conferences will continue to scale new heights.
Prof. Jon-Lark Kim
Sogang University, South Korea
An auto-encoder which can be split into two parts is designed. The two parts can work well separately. The top half is an abstract network which is trained by supervised learning and can be used to classify and regress. The bottom half is a concrete network which is accomplished by inverse function and trained by self-supervised learning. It can generate the input of abstract network from concept or label. It is tested by tensorflow and mnist dataset. The abstract network is like LeNet-5. The concrete network is the inverse of the abstract network. A picture can change to label that is compression, then change it back from label that is decompression. So compression and decompression can be realized by the autoencoder. Through test, the absolute function can do generation task well, while the leaky relu function can do classification task well. Lossy compression can be achieved by abstract network and concrete network with absolute function. With one-hot encoding, the compression ratio is 5.1% when decompression quality is good. With binary encoding, the compression ratio is 2% when decompression quality is ok. The mean PNSR of five pictures is 19.48 dB. When jump connection and negative feedback are used, the decompression performance is good. The mean PNSR of five pictures is 29.86 dB. The compression ratio is 56.1%.
An auto-encoder which can be split into two parts is designed. The two parts can work well separately. The top half is an abstract network which is trained by supervised learning and can be used to classify and regress. The bottom half is a concrete network which is accomplished by inverse function and trained by self-supervised learning. It can generate the input of abstract network from concept or label. It is tested by tensorflow and mnist dataset. The abstract network is like LeNet-5. The concrete network is the inverse of the abstract network. Through test, encryption and decryption can be achieved by abstract network and concrete network add jump connection and negative feedback with absolute function. When binary encoding is used, although the encrypted vector is four bits, the result of decryption has the same quality as one-hot encoding because of the jump connection and negative feedback. The parameter of DNN is secondary key, the architecture of DNN is primary key. Secondary key can be shared by all the people, primary key can be shared by sender and receiver. The key can be generated by training the DNN. When big dataset is used for encryption, the classes are far bigger, the label may be something in the world, numbers, words, or attributes represented by float number. The label can use the mix of one-hot encoding and binary encoding, it is harder to attack. Through analysis, it is safe for most situations.
Understanding the perceptions and experiences of a community regarding disasters is crucial in effectively planning and implementing disaster strategies. There are two known approaches to analyzing perceptions, qualitative and automated approaches for thematic analysis. This paper aims to investigate the strengths and limitations of the mentioned approaches. Thus, using both approaches, this study analyzed data from a focus group discussion about disaster risk reduction and management and climate change adaptation conducted in a typhoon-prone city in Legazpi City, Philippines. An inductive-deductive approach for the qualitative method while a language model-assisted approach for the automatic method of extracting prominent themes from the collected responses. The results show (dis) similarities regarding themes obtained from the two approaches, specifically the emphasis on concerns about the proper distribution of relief goods and donations, proper early monitoring of potentially powerful typhoons, and other forms of threat, including politically motivated ones. From these findings, we conclude the importance of incorporating a combined manual and automatic approach for thematic analysis of natural language.
Human anatomy is the foundation of medical education. Comprehensive mastering of anatomy allows medical students to focus on their follow-up courses in greater depth. This teaching reform implements the spirit of service teaching and “red education”. The construction of a network-based platform for human structure-rich media resources aims to promote the concept of “all-media teaching” and implement a hybrid teaching approach that combines guidance and learning in anatomy education. This mode develops a new teaching technique of “one center, three carriers, five integration” which promotes the pattern of medical education, and enhances the quality of education and teaching.
China is the world’s largest apple growing country; Apples are affected by various diseases during the planting process, and timely detection of fruit tree diseases is of great significance for improving apple yield. With the development of artificial intelligence technology, The use of deep learning to detect the condition of fruit trees has become a research hotspot in forest science planting. ResNet50 is used to solve the bottleneck problem of recognition accuracy of traditional neural networks, and data enhancement is used to improve the quality of data samples. transfer learning is used to solve the application of small-scale data samples, and loss function is used to balance sample errors. Improve the ability to obtain important feature information by using a dual channel mixed attention mechanism. Finally, a comprehensive network model combining residual neural network, dual channel mixed attention mechanism, transfer learning and loss function is designed. The experimental results show that the effect of this comprehensive ResNet50 model is better than that of the original ResNet50 model and VGG19 model, and its classification accuracy rate can reach 96.87%. This method improves the performance in fruit disease detection, and has wider applicability in practical applications.
In this paper, we propose a method for finding the similarity of sentence pairs. The method combines two modules – a modified Latent Semantic Analysis and a semantic similarity computation. The proposed method makes use of the syntactic structure and semantic information contained in the sentence pairs. The syntactic structure in the form of dependency triplets is extracted and a semantic similarity calculation is carried out. The semantic similarity between words is calculated using Wu & Palmer similarity measure and Wordnet synonym relation is used in modified Latent Semantic Analysis. The proposed method is evaluated on the Microsoft Research Paraphrase Corpus dataset and the accuracy obtained on the dataset is 73.19% which is better than existing statistical and zero shot domain adaptation methods. The proposed method is also tested on Li et al. text similarity dataset and the Pearson correlation coefficient of 0.9021, Spearman correlation of 0.9103 and mean deviation of 0.105 with the human judgement show that the method outperforms state-of-the-art methods.
In recent years, the automatic classification of leaves has attracted more and more researches. The accurate classification of leaves enables the development of smart solutions in agriculture. The paper analyzes and compares the classification of leaf images using deep neural networks and handcrafted feature extraction methods on public datasets. Both powerful deep neural network models and handcrafted feature extraction methods have been optimized and employed to classify leaf images accurately without using specific domain knowledge. The highest classification accuracies of 94% and 97.78% are achieved on the public (Plant Village) datasets that consist of grayscale and color plant leaf images, respectively. The obtained classification accuracy of apple leaf disease images is 95.53%. Obtained results show the efficiency and robustness of the deep neural networks on the classification of leaf images. The comparison results allow to develop the applications of the automatic classification of leaf imagery in reality.
Nowadays, there is a large amount of symbolic data in various fields, and people can fully utilize these symbolic data for clustering, providing a better foundation and direction for data mining and analysis. Currently, clustering algorithms for symbol data have emerged one after another, but there are still shortcomings in computational cost and algorithm robustness. Therefore, it is urgent to study an algorithm with stable clustering results, less time consumption, and low I/O overhead. The following proposes a symbolic cluster analysis algorithm that approaches to the optimal value. It reduces the size of the original data by generating a symbolic association graph from a large number of symbolic data samples, effectively solves the problem of high computing costs caused by the huge amount of data, and proves the clustering effect of the algorithm through empirical analysis.
This paper explores the way to apply ChatGPT to mitigate the side effects of Personal Learning Environments (PLEs) in higher education from the perspectives of teachers and teaching. An interview with six university professors and two Information and Communication Technology (ICT) experts was conducted. The thematic analysis reveals two main categories of side effects, including educational philosophy and teaching skills. The data analysis also contends that ChatGPT offers a valuable tool to aid educators in overcoming these challenges by improving the personalization of PLEs, aligning PLEs with formal education requirements, and designing innovative assessments and enhancing learning engagement in PLEs. This study provides insights beneficial to practitioners, researchers, and policymakers by elucidating the challenges and opportunities associated with integrating ChatGPT with PLEs in higher education.
In this study, we focused on analyzing customer-generated data on Facebook to explore how textual content on a social web can provide valuable information for decision support. To accomplish this goal, we used several techniques that included social network analysis (SNA), natural language processing (NLP), data mining (DM), and machine learning (ML), integrating them with artificial intelligence approaches. Our analysis aimed to harness the information generated during the Volkswagen pollutant emissions situation in a case study that was conducted using the textual content from 10,642 posts, that represented the interactions of 25,877 users over a span of twenty-two weeks. The results demonstrated that monitoring online social networks (OSNs) can significantly enhance decision-making processes and might help to mitigate potential damages to brands/businesses. By leveraging the proposed methodological approach, a set of orientations for decision-making was extracted, providing valuable guidance for brand management and reputation protection. Overall, this study highlights the importance of analyzing textual content on OSNs and leveraging advanced computational techniques to improve decision support.
The aim of global distribution network optimization is to optimize the flow of goods between logistics nodes, leading to more efficient and compact packing. As a result, this optimization helps to reduce the shipping cost, which is calculated based on the weight and volume of package after cartonization. Shipping cost is calculated from chargeable weight. In this optimization problem, the routing or rerouting of products or raw materials would result in a new shipment network. To compare the performance of various shipment networks, we use the logistics cost of all shipments within a past time window as the evaluation criteria. Hence, when dealing with the routing/rerouting of numerous types of products/raw materials and having to consider numerous central distribution centers (CDC), a multitude of shipment network configurations would arise. The logistics cost for each routing/rerouting affects other networks and in turn, requires the logistics cost of all other shipment networks to be recomputed as well. Given the enormity of shipments in each network, it is infeasible to employ a cartonization solver to pack and then compute the chargeable weight of the shipments. Chargeable weight is the greater of actual weight and the volumetric weight of the carton after packing. In this paper, a neural network model is applied to predict the chargeable weight of shipments. Conventional machine learning models, such as random forest and support vector regression are used as the benchmark models. Moreover, to further reduce the overall mean error ratio, we propose using exact algorithm and Red Jasper’s cartonization solver to calculate the chargeable weight for small shipments as this combined method runs fast and results in minimal error. As for large complex shipments, we propose using machine learning method to approximate the chargeable weight. Based on real data provided by one of the top five semiconductor equipment makers in the world for experimentation, results suggest that our method achieves a significant improvement in computational speed while maintaining a low mean error.
In the age of Information Disorder, Satire is one of its phenomena mainly occurring in the context of social media. Satire represents an interesting study subject given that it can be easily confused with further forms of the disorder. The present work proposes and evaluates a set of linguistic features to build classifiers able to distinguish satires from other textual contents. The adopted features are firstly identified within the scientific literature and, secondly, ranked and filtered by means of the Information Gain index. Several experimentation activities show good performance for the aforementioned classifiers and an acceptable ability to generalize for the models trained with such features.
Digital transformation is one of the major challenges thrown up with the technological and societal evolution of recent years. It represents an interesting research area nowadays, especially with its increasing number of papers in this field. However, the current literature approaches the implementation of digital transformation in a fragmented way. In addition, several managers report their inability to succeed in the digital transition due to a lack of visibility and a non-existent integrative approach. To overcome these limits, this work proposes a transversal and integrative approach to the implementation of digital transformation and its main stages. We suggest adding the integration of the PDCA approach into the whole process of digital transformation to optimize the results of this transformation. Thus, this work can serve as a guide for any company wishing to consider a technological investment and digital transformation project.
Cause-effect graphs (CEGs) are usually applied for black-box testing of complex industrial systems. The specification process is time-consuming and can result in many errors. In this work, machine learning methods were applied for predicting the feasibility of CEG elements. All information was extracted from graphs contained in CEGSet, a dataset of CEGs. The data was converted to two different formats. The Boolean features format represents relations as separate data rows, whereas the Term-Frequency times Inverse-Document-Frequency (TF-IDF) format represents graphs as data rows. Eight machine learning models were trained on this data. The results of testing by using the 80–20 holdout method indicate that important information is lost when converting the graphs to the TF-IDF format, whereas the Boolean feature format enables 100%-accurate predictions of ensemble methods. The achieved results indicate that pre-trained models can be used as help for domain experts during the CEG specification process.
We live in an era in which the preservation of the environment is being widely discussed, driven by growing concerns over climate issues. One major factor contributing to this situation is the lack of attention societies give to maintaining high sustainability levels. Data plays a crucial role in understanding and assessing sustainability impacts in both urban and rural areas. However, obtaining comprehensive data on a country’s sustainability is challenging due to the lack of simple and accessible sources. Existing solutions for sustainability analysis are limited by high costs and implementation difficulties, which restrict their spatial coverage. In this paper, we propose a solution using low-cost hardware and open-source technologies to collect data about the movement of people and vehicles. This solution involves low-cost video-based meters that can be flexibly deployed to various locations. Specifically, we developed a prototype using Raspberry Pi and YOLO which is able to correctly classify 91% of the vehicles by type, and 100% of the events (entering of leaving). The results indicate that this system can effectively and affordably identify and count people and vehicles, allowing for its implementations namely in remote sensitive areas such as natural parks, in which the access of people and vehicles must be controlled and monitored.
With the development of technology and the increase in the need to use the internet and the transmission of personal data and save it on the cloud and personal computers and with the increase in security risks represented by penetrations and cyber threats, and among these threats are those represented by the use of deep learning techniques to produce deep fake images that are mostly used to violate cyber security through threats such as the ransom threat or publishing fake news and other cyber risks, so there was an urgent need to establish systems to detect deep forgery in images or videos, in this paper a system was proposed to detect deep forgery based on the principle of using the diffusion model based on the graph based on image segmentation where the system transforms the image to graph and then segment it into three areas which are fake, real and background. And do purification of the adversarial content in these areas using diffusion technology, and then identify the fake areas in the image and use GCN in order to distinguish between real and fake images. In order to evaluate the proposed method, it was compared with the other methods on different data sets (CELEB-DF (V2) data set, Face Forensics++ data set, and wild deep fake). The result shows that the proposed system achieved better results than the approaches in the literature.
Supply chain finance has brought new choices to agricultural enterprises facing funding shortages. However, due to the different credit ratings and frequent default events of participating enterprises in the chain, the promotion of agricultural supply chain finance business is hindered. This can be solved by the technical characteristics of blockchain. This study discusses the rationality and feasibility of applying blockchain technology to agricultural supply chain finance through literature review and mechanism analysis, using data from Chinese industries. The application of blockchain technology helps to build a good supply chain financial ecosystem, helping Chinese agricultural enterprises explore new financing channels and overcome financial difficulties.
Bayesian inference, a statistical methodology rooted in Bayes’ theorem, offers the ability to compute probability distributions of unobserved phenomena, given observed information. To this end, this technique has proven useful in disease diagnosis based on equipment measurement. This paper proposes an innovative Bayesian inference strategy capable of rapidly estimating capillary oxygen supply capability in muscle tissues by leveraging uncertainty quantification techniques. Specifically, the oxygen supply capability is formulated with Krogh Erlang’s equation along with Fick’s second law. Moreover, the prior distribution of the early-time capillary oxygen supply capability is updated using acquired measurements of oxygen concentration within capillary to yield the posterior distribution. The resulting data with supportive simulation indicates that the cellular dimension can be efficiently updated, thereby facilitates the accurate uncertainty quantification of cellular environment estimate.
Cable-stayed bridges are critical components of modern transportation infrastructure, but their vulnerability to seismic events presents substantial risks to their long-term performance and safety. This paper focuses on the assessment of seismic damage in the lifecycle risk cost calculation for cable-stayed bridge infrastructure. The fragility method is utilized as a key approach to quantify seismic vulnerability and estimate associated risk costs throughout the bridge’s lifespan. Seismic fragility curves are developed by analyzing the structural response to various earthquake scenarios, providing a probabilistic representation of the bridge’s performance based on ground shaking intensity. This facilitates the evaluation of potential damage levels and their corresponding costs. The risk cost calculation encompasses direct expenses, including repair and replacement of damaged components, and employs statistical methods to estimate expected expenses. Stakeholders and decision-makers can utilize this approach to make informed choices regarding risk reduction investments, maintenance planning, and long-term sustainability considerations. A specific case study is conducted on a cable-stayed bridge, focusing on longitudinal seismic waves. The findings reveal that the main tower exhibits greater resilience compared to the tower abutment. The failure probabilities of slight, moderate, severe, and absolute damage for the main tower are determined as 9.8%, 1.1%, 0.3%, and 0.2% respectively, while the corresponding failure probabilities for the main abutment are 35.2%, 21.1%, 7.4%, and 2.9%. The maintain life cycle cost associated with seismic events for this bridge is estimated at 0.548 million USD. These results provide valuable insights for decision-making processes regarding risk reduction strategies, maintenance planning, and long-term sustainability considerations.
Exercise is an indispensable part of people’s lives and is closely related to their health. Human Activity Recognition (HAR), which involves detects and analyzes human body activity, has become the focus of current research. Photoplethysmography (PPG) has advantages such as convenience for detection and low cost, and is widely used in wearable devices becoming an ideal choice for HAR. In this study, we used wavelet scattering transform (WST) to extract features from PPG and then performed activity recognition on it. We achieved excellent classification accuracy of 92.54% and 97.76% respectively in the experiments of three-class and four-class exercise detection. The results showed this method based on wavelet scattering transform and PPG can accurately detect exercise types and provide effective support for HAR.