Lung sounds remain vital in clinical diagnosis as they reveal associations with pulmonary pathologies. With COVID-19 spreading across the world, it has become more pressing for medical professionals to better leverage artificial intelligence for faster and more accurate lung auscultation. This research aims to propose a feature engineering process that extracts the dedicated features for the depthwise separable convolution neural network (DS-CNN) to classify lung sounds accurately and efficiently. We extracted a total of three features for the shrunk DS-CNN model: the short-time Fourier-transformed (STFT) feature, the Mel-frequency cepstrum coefficient (MFCC) feature, and the fused features of these two. We observed that while DS-CNN models trained on either the STFT or the MFCC feature achieved an accuracy of 82.27% and 73.02%, respectively, fusing both features led to a higher accuracy of 85.74%. In addition, our method achieved 16 times higher inference speed on an edge device and only 0.45% less accuracy than RespireNet. This finding indicates that the fusion of the STFT and MFCC features and DS-CNN would be a model design for lightweight edge devices to achieve accurate AI-aided detection of lung diseases.

In this study, we propose using a thermal imaging camera (TIC) with a deep learning model as an intelligent human detection approach during emergency evacuations in a low-visibility smoky fire scenarios. We use low-wavelength infrared (LWIR) images taken by a TIC qualified with the National Fire Protection Association (NFPA) 1801 standards as input to the YOLOv4 model for real-time object detection. The model trained with a single Nvidia GeForce 2070 can achieve >95% precision for the location of people in a low-visibility smoky scenario with 30.1 frames per second (FPS). This real-time result can be reported to control centers as useful information to help provide timely rescue and provide protection to firefighters before entering dangerous smoky fire situations.

Contract production and transaction are very common and important since it can greatly reduce future uncertainty and is well-organized by buyers, sellers, and trusted third parties (TTPs). However, current commodity trading systems (such as spot, futures, and forward contract) are still traditional centralized structures, which means that there may always be concerned about the single point of failure and data corruption. Besides, even though there already exists fragmentary decentralized applications (DApps) for the marketplace, order management, resale, delivery, financing, and insurance, they are not integrated for operating contract production and transaction comprehensively. In this work, a blockchain-enabled integrated marketing platform (BeIMP) is proposed for contract production and transactions. BeIMP is a consortium blockchain (CsBc) framework that enables better safety, efficiency, and interoperability among stakeholders. BeIMP can directly match producers and consumers and reduce the problem of intermediaries’ improper market operation. BeIMP has a three-tier risk diversification mechanism. First, both parties can know the quantity and price according to the contract specification (CSpec) in advance to avoid future uncertainty. Second, the buyers can resell the established order if they need, thereby diversifying the risk. Third, the concept of insurance is introduced to reduce force majeure and other systemic risks. BeIMP can also help producers get the loan from the financial institution (FI) if they need fund for production. We implement and demonstrate the prototype in contract farming (CF) first and discuss its generalizability to other scenarios. Experiment results show that the smart contract (SC) function is stable enough and the proof of authority (PoA) has the advantage in throughput and can give users a better experience.

We study the problem of controllable citation text generation by introducing a new concept to generate citation texts. Citation text generation, as an assistive writing approach, has drawn a number of researchers’ attention. However, current research related to citation text generation rarely addresses how to generate the citation texts that satisfy the specified citation intents by the paper’s authors, especially at the beginning of paper writing. We propose a controllable citation text generation model that extends a pre-trained sequence to sequence models, namely, BART and T5, by using the citation intent as the control code to generate the citation text, meeting the paper authors’ citation intent. Experimental results demonstrate that our model can generate citation texts semantically similar to the reference citation texts and satisfy the given citation intent. Additionally, the results from human evaluation also indicate that incorporating the citation intent may enable the models to generate relevant citation texts almost as scientific paper authors do, even when only a little information from the citing paper is available.

Digital advertising on social media officially surpassed traditional advertising and became the largest marketing media in many countries. However, how to maximize the value of the overall marketing budget is one of the most concerning issues of all enterprises. The content of the Facebook photo post needs to be analyzed effectively so that the social media companies and managers can concentrate on handling their fan pages. This research aimed to use text mining techniques to find the audience accurately. Therefore, we built a topic model recommendation system (TMRS) to analyze Facebook posts by sorting the target posts according to the recommended scores. The TMRS includes six stages, such as data preprocessing, Chinese word segmentation, word refinement, TF-IDF word vector conversion, creating model via Latent Semantic Indexing (LSI), or Latent Dirichlet Allocation (LDA), and calculating the recommendation score. In addition to automatically selecting posts to create advertisements, this model is more effective in using marketing budgets and getting more engagements. Based on the recommendation results, it is verified that the TMRS can increase the engagement rate compared to the traditional engagement rate recommended method (ERRM). Ultimately, advertisers can have the chance to create ads for the post with potentially high engagements under a limited budget.

To improve data transmission efficiency, image compression is a commonly used method with the disadvantage of accompanying image distortion. There are many image restoration (IR) algorithms, and one of the most advanced algorithms is the generative adversarial network (GAN)-based method with a high correlation to the human visual system (HVS). To evaluate the performance of GAN-based IR algorithms, we proposed an ensemble image quality assessment (IQA) called ATDIQA (Auxiliary Transformer with DISTS IQA) to give weights on multiscale features global self-attention transformers and local features of convolutional neural network (CNN) IQA of DISTS. The result not only performed better on the perceptual image processing algorithms (PIPAL) dataset with images by GAN IR algorithms but also has good model generalization over LIVE and TID2013 as traditional distorted image datasets. The ATDIQA ensemble successfully demonstrates its performance with a high correlation with the human judgment score of distorted images.