Anomaly detection without employing dedicated sensors for each industrial machine is recognized as one of the essential techniques for preventive maintenance and is especially important for factories with low automatization levels, a number of which remain much larger than autonomous manufacturing lines. We have based our research on the hypothesis that real-life sound data from working industrial machines can be used for machine diagnostics. However, the sound data can be contaminated and drowned out by typical factory environmental sound, making the application of sound data-based anomaly detection an overly complicated process and, thus, the main problem we are solving with our approach. In this paper, we present a noise-tolerant deep learning-based methodology for real-life sound-data-based anomaly detection within real-world industrial machinery sound data. The main element of the proposed methodology is a generative adversarial network (GAN) used for the reconstruction of sound signal reconstruction and the detection of anomalies. The experimental results obtained in the Malfunctioning Industrial Machine Investigation and Inspection (MIMII) show the superiority of the proposed methodology over baseline approaches based on the One-Class Support Vector Machine (OC-SVM) and the Autoencoder–Decoder neural network. The proposed schematics using the unscented Kalman Filter (UKF) and the mean square error (MSE) loss function with the L2 regularization term showed an improvement of the Area Under Curve (AUC) for the noisy pump data of the pump.

Automated generation of medical reports that describe the findings in the medical images helps radiologists by alleviating their workload. Medical report generation system should generate correct and concise reports. However, data imbalance makes it difficult to train models accurately. Medical datasets are commonly imbalanced in their finding labels because incidence rates differ among diseases; moreover, the ratios of abnormalities to normalities are significantly imbalanced. We propose a novel reinforcement learning method with a reconstructor to improve the clinical correctness of generated reports to train the data-to-text module with a highly imbalanced dataset. Moreover, we introduce a novel data augmentation strategy for reinforcement learning to additionally train the model on infrequent findings. From the perspective of a practical use, we employ a Two-Stage Medical Report Generator (TS-MRGen) for controllable report generation from input images. TS-MRGen consists of two separated stages: an image diagnosis module and a data-to-text module. Radiologists can modify the image diagnosis module results to control the reports that the data-totext module generates. We conduct an experiment with two medical datasets to assess the data-to-text module and the entire two-stage model. Results demonstrate that the reports generated by our model describe the findings in the input image more correctly.

Knowledge graphs (KGs) are generally used for various NLP tasks. However, as KGs still miss some information, it is necessary to develop Knowledge Graph Completion (KGC) methods. Most KGC researches do not focus on the Out-of-KGs entities (Unseen-entities), we need a method that can predict the relation for the entity pairs containing Unseen-entities to automatically add new entities to the KGs. In this study, we focus on relation prediction and propose a method to learn entity representations via a graph structure that uses Seenentities, Unseen-entities and words as nodes created from the descriptions of all entities. In the experiments, our method shows a significant improvement in the relation prediction for the entity pairs containing Unseen-entities.

In this study, we propose inning summarization methods to generate a simple and sophisticated summary of baseball games using play-by-play data. We focus on the two information sources of inning reports and game summaries. First, we generate a basic sentence using an inning report; further, the basic sentence is integrated with an explanatory phrase to generate inning summaries that contain expressions such as "the long-awaited first score" in game summaries. We refer to these phrases as the game-changing phrases (GPs). Readers can easily understand the situation of a game using GPs. In this study, we investigate the template-and neural-based methods of summary generation. Additionally, we evaluate the two methods and discuss their advantages and disadvantage.