Human activity recognition (HAR) based on sensor data is a significant problem in pervasive computing. In recent years, deep learning has become the dominating approach in this field, due to its high accuracy. However, it is difficult to make accurate identification for the activities of one individual using a model trained on data from other users. The decline on the accuracy of recognition restricts activity recognition in practice. At present, there is little research on the transferring of deep learning model in this field. This is the first time as we known, an empirical study was carried out on deep transfer learning between users with unlabeled data of target. We compared several widely-used algorithms and found that Maximum Mean Discrepancy (MMD) method is most suitable for HAR. We studied the distribution of features generated from sensor data. We improved the existing method from the aspect of features distribution with center loss and get better results. The observations and insights in this study have deepened the understanding of transfer learning in the activity recognition field and provided guidance for further research.

Text-to-SQL semantic parsing is an important NLP task, which facilitates the interaction between users and the database. Much recent progress in text-to-SQL has been driven by large-scale datasets, but most of them are centered on English. In this work, we present MultiSpider, the largest multilingual text-to-SQL semantic parsing dataset which covers seven languages (English, German, French, Spanish, Japanese, Chinese, and Vietnamese). Upon MultiSpider we further identify the lexical and structural challenges of text-to-SQL (caused by specific language properties and dialect sayings) and their intensity across different languages. Experimental results under various settings (zero-shot, monolingual and multilingual) reveal a 6.1% absolute drop in accuracy in non-English languages. Qualitative and quantitative analyses are conducted to understand the reason for the performance drop of each language. Besides the dataset, we also propose a simple schema augmentation framework SAVe (Schema-Augmentation-with-Verification), which significantly boosts the overall performance by about 1.8% and closes the 29.5% performance gap across languages.

Aeromagnetic compensation is a technology used to reduce aircraft magnetic interference, which plays an important role in aeromagnetic surveys. In addition to maneuvering interferences, the onboard electronic (OBE) interference has been proven to be a significant part of aircraft interference, which must be reduced before further interpretation of aeromagnetic data. In the past, most researchers have focused on establishing linear models to compensate for OBE magnetic interference. However, such methods can only work using accurate reference sensors. In this paper, we propose a data-driven OBE interference compensation method, which can reduce OBE interference without relying on any other reference sensor. This network-based method can integrally detect and repair the OBE magnetic interference. The proposed method builds a prediction model by combining wavelet decomposition with a long short-term memory (LSTM) network to detect and predict OBE interference, and then estimates the local variation of the magnetic field to remove the drift of the interference. In our tests, we construct 10 semi-real datasets to quantitatively evaluate the performance of the proposed method. The F1 score of the proposed method for OBE interference detection is over $0.79$, and the RMSE of the compensated signal is less than $0.009$ nT. Moreover, we also test our method on real signals, and the results show that our method can detect all interference and significantly reduce the standard deviation of the magnetic field.

Deep learning-based bearing fault diagnosis methods typically require a substantial amount of labeled data, which are expensive and time-consuming to obtain. Semi-supervised learning (SSL) provides a solution by leveraging both labeled and unlabeled data, but its performance can be compromised by the presence of unknown classes in real-world unlabeled data, reducing robustness and diagnostic accuracy. To address this issue, we propose Open-set Semi-supervised Contrastive Learning (OSCL), a novel framework that combines contrastive learning with open-set recognition. OSCL first utilizes contrastive pre-training to extract discriminative feature representations from vibration signals. Furthermore, it jointly optimizes an open-set classifier (to perform open-set tasks) and a closed-set classifier (to perform closed-set tasks) using both known and unknown class data. To further enhance feature representations, raw vibration data is processed using a multi-domain fusion strategy that integrates Short-Time Fourier Transform (STFT), Continuous Wavelet Transform (CWT), and Time-Domain Conversion (TDC). Meanwhile, tailored strong and weak augmentations are also applied to enhance the model’s effectiveness. Experiments on three benchmark datasets show that OSCL achieves state-of-the-art performance in both closed-set classification and open-set scenarios, while remaining robust with limited labeled samples. The framework’s ability to generalize across datasets and handle unknown classes demonstrates its practical applicability in industrial settings.