Search

Biosignals provide information about the onset of certain diseases or health conditions, aiding in diagnosing diseases and monitoring health conditions quickly and accurately. However, the inter-subject variability within biosignals hampers the model performance. In this study, we propose an inter-subject similar loss to learn representations robust to inter-subject variability. This loss promotes subject invariance, improves the generalizability of the representation, and allows better representations to be learned even with fewer training subjects. The proposed framework consists of two complementary loss functions: (1) supervised contrastive loss and (2) inter-subject similar loss. We evaluated the classification performance of the proposed method on three public biosignal datasets. The experimental results demonstrate that the proposed method outperforms the comparison methods and that the subject-invariant representation performs well on unseen subjects. Code is available at: https://github.com/KimHyeon-Ji/CSGR-Bio.git [ABSTRACT FROM AUTHOR]

Over the past few years, the growing interest in adapting wearable and mobile devices for smart healthcare and living systems has led to an expansion of studies on human activity recognition (HAR). Recently, many HAR studies have focused on deep supervised learning, which automatically extracts meaningful features through extensive labeled data, is widely employed and demonstrated impressive progress. However, several issues, such as privacy sensitivity, cost intensity, and restricted laboratory conditions in HAR, impede the scalability of human activity labels, and the lack of labeled data degrades the performance of supervised learning. To address these challenges, we aim to extract useful representations without labels through self-supervised learning. Especially, we propose signal-wise self-supervised learning (SWISS) to learn signal interactions, taking into account the characteristics of inertial measurement unit used in sensor-based HAR data. First, we design data-level and feature-level reconstruction tasks for masked signals. These tasks compel the model to learn signal interactions by reconstructing masked signals in both the original data space and the low-dimensional feature space. Additionally, we design a new model architecture called the signal token transformer (STT), which is suitable for encoding both temporal features and signal interactions. We demonstrate the effectiveness of our approach on four publicly available human activity datasets. Furthermore, we show that STT can identify important signals and facilitate the learning of signal interactions through a self-attention mechanism. The code is available at https://github.com/DMQA-SR/HAR_SWISS. [ABSTRACT FROM AUTHOR]

Player evaluation is a key component of the question-answering (QA) system in sports. Since existing player evaluation methods heavily rely on game statistics, they cannot capture the qualitative impact of each player during a game, which can be exploited using news articles after the game. In this paper, we propose a deep learning-based player evaluation model by combining both quantitative game statistics and the qualitative analyses provided by news articles. Players are classified as positive or negative based on their performance during certain periods, and news articles in the same period are annotated using the player's class. Then, the relationship between news articles and the annotated polarity is investigated by a deep neural network, which can deal with the high dimensionality of the text data. Since there is no explicit polarity label for news articles, we use the change in game statistics in target periods to annotate related sentences. The proposed system is applied to a Korean professional baseball league (KBO) and it is shown to be capable of understanding the sentence polarity of news articles on player performances. [ABSTRACT FROM AUTHOR]