Xiuxuan Shen, Zhongyuan Jiang, Junsan Zhang, Junxiao Han, Yao Wan, Chengjie Guo, Bingcheng Liu, Jie Wu, Renxiang Li, Philip S. Yu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

In geological exploration, due to various factors, raw seismic data often corrupted by random noise, posing significant challenges to data fidelity, signal-to-noise ratio (SNR), and resolution. Traditional denoising techniques are limited by their inability to fully consider spatiotemporal correlations within seismic data, reliance on fixed filter parameters, and inadequate modeling of complex noise, and they have limited effectiveness and applicability when processing complex seismic signals. Therefore, this study proposes an innovative seismic data denoising model named seismic transformer generative adversarial network (STGAN). This model combines Transformer technology with generative adversarial networks (GANs) to effectively remove noise from seismic data. By incorporating the GAN, the model can effectively learn complex noise features in seismic data and generate cleaner, more realistic data. Meanwhile, the adoption of Transformers makes it possible to captures the temporal dependencies of seismic signals, thus significantly improving the accuracy and efficiency of data processing. The model employs serial and parallel structures in the generator to effectively extract multiscale features and achieve a balance between denoising effect and computational efficiency. Additionally, traditional batch normalization (BN) is replaced with batch renormalization (BRN) to further improve the stability of the model and optimize model performance. Comparative experiments on synthetic and field datasets demonstrate that the STGAN model has better denoising performance than classical methods such as bicubic interpolation, nonlocal means algorithm, and DnCNN, providing a more accurate basis for further interpretation and analysis of seismic data.

Automatic generation of medical reports can provide diagnostic assistance to doctors and reduce their workload. To improve the quality of the generated medical reports, injecting auxiliary information through knowledge graphs or templates into the model is widely adopted in previous methods. However, they suffer from two problems: 1) The injected external information is limited in amount and difficult to adequately meet the information needs of medical report generation in content. 2) The injected external information increases the complexity of model and is hard to be reasonably integrated into the generation process of medical reports. Therefore, we propose an Information Calibrated Transformer (ICT) to address the above issues. First, we design a Precursor-information Enhancement Module (PEM), which can effectively extract numerous inter-intra report features from the datasets as the auxiliary information without external injection. And the auxiliary information can be dynamically updated with the training process. Secondly, a combination mode, which consists of PEM and our proposed Information Calibration Attention Module (ICA), is designed and embedded into ICT. In this method, the auxiliary information extracted from PEM is flexibly injected into ICT and the increment of model parameters is small. The comprehensive evaluations validate that the ICT is not only superior to previous methods in the X-Ray datasets, IU-X-Ray and MIMIC-CXR, but also successfully be extended to a CT COVID-19 dataset COV-CTR.

The approaches based on spatio-temporal features for video action recognition have emerged such as two-stream based methods and 3D convolution based methods. However, current methods suffer from the problems caused by partial observation, or restricted to single information modeling, and so on. Segment-level recognition results obtained from dense sampling can not represent the entire video and, therefore lead to partial observation. And a single model is hard to capture the complementary information on spacial, temporal and spatio-temporal information from video at the same time. Therefore, the challenge is to build the video-level representation and capture multiple information. In this paper, a video-level multi-model fusion action recognition method is proposed to solve these problems. Firstly, an efficient video-level 3D convolution model is proposed to get the global information in the video which assembling segment-level 3D convolution models. Secondly, a multi-model fusion architecture is proposed for video action recognition to capture multiple information. The spatial, temporal and spatio-temporal information are aggregate with SVM classifier. Experimental results show that this method achieves the state-of-the-art performance on the datasets of UCF-101(97.6%) without pre-training on Kinetics.