Federated learning enables multiple medical institutions to train a global model without sharing data, yet feature heterogeneity from diverse scanners or protocols remains a major challenge. Many existing works attempt to address this issue by leveraging model representations (e.g., mean feature vectors) to correct local training; however, they often face two key limitations: 1) Incomplete Contextual Representation Learning: Current approaches primarily focus on final-layer features, overlooking critical multi-level cues and thus diluting essential context for accurate segmentation. 2) Layerwise Style Bias Accumulation: Although utilizing representations can partially align global features, these methods neglect domain-specific biases within intermediate layers, allowing style discrepancies to build up and reduce model robustness. To address these challenges, we propose FedBCS to bridge feature representation gaps via domain-invariant contextual prototypes alignment. Specifically, we introduce a frequency-domain adaptive style recalibration into prototype construction that not only decouples content-style representations but also learns optimal style parameters, enabling more robust domain-invariant prototypes. Furthermore, we design a context-aware dual-level prototype alignment method that extracts domain-invariant prototypes from different layers of both encoder and decoder and fuses them with contextual information for finer-grained representation alignment. Extensive experiments on two public datasets demonstrate that our method exhibits remarkable performance.

Dialogue State Tracking (DST) is designed to monitor the evolving dialogue state in the conversations and plays a pivotal role in developing task-oriented dialogue systems.However, obtaining the annotated data for the DST task is usually a costly endeavor.In this paper, we focus on employing LLMs to generate dialogue data to reduce dialogue collection and annotation costs.Specifically, GPT-4 is used to simulate the user and agent interaction, generating thousands of dialogues annotated with DST labels.Then a two-stage fine-tuning on LLaMA 2 is performed on the generated data and the real data for the DST prediction.Experimental results on two public DST benchmarks show that with the generated dialogue data, our model performs better than the baseline trained solely on real data.In addition, our approach is also capable of adapting to the dynamic demands in real-world scenarios, generating dialogues in new domains swiftly.After replacing dialogue segments in any domain with the corresponding generated ones, the model achieves comparable performance to the model trained on real data 1 .

Enhancer-promoter interactions are one of the essential mechanisms in the regulation of gene expression, and Accurate identification of enhancer-promoter interactions (EPIs) is challenging. In recent years, many deep learning methods have been used for EPI prediction. In this study, we propose DeepDualEPI, a dual-channel deep learning model based on genomic signals and DNA sequences, for predicting enhancer-promoter interactions (EPI). We used network architectures such as Dilated CNN, BiLSTM, and Transformer to process genomic signals, and network architectures such as multiscale CNN to extract DNA sequence features, and finally obtained hybrid features and output EPI prediction probabilities. To obtain the best combination of parameters for the model, we conducted several ablation experiments to optimize the model parameters. And to validate the performance of DeepDualEPI, we conducted experiments on four independent test sets to verify the generalization ability of the model. Compared with other state-of-the-art EPI prediction models, the DeepDualEPI model shows significant improvement in both AUC and AUPR evaluation metrics and experimentally demonstrates that better results are achieved on every chromosome, which proves that our model can stably perform EPI prediction across cell lines. And this paper demonstrates through ablation experiments that the inclusion of DNA sequence information can improve the performance of the model. Therefore, the two-channel hybrid feature deep learning approach via genomic signals and DNA sequences proposed in this paper helps to improve the overall accuracy of EPI prediction.

As a foundation task of natural language processing, text classification is widely used in information retrieval, public opinion analysis, and other related tasks.Facing the problem of sparse features of Chinese short texts, which affects the classification accuracy of Chinese short texts, this paper proposes a Chinese short text classification method based on the Character Frequency Sub-word Enhancement (CFSE), which can effectively improve the classification accuracy of Chinese short texts.First, the initial Chinese-character sequence is mapped to the corresponding Character Frequency Sub-word (CFS) sequence based on the global character 1 frequency information.Second, the relationship features among data are extracted based on BiLSTM-Att processing CFS sequence, and the semantic features of the initial Chinese-character sequence are obtained through ERNIE.Finally, these two kinds of features are fused and input into the text classifier to obtain the classification results.Experimental results show that the proposed method can improve the classification accuracy of Chinese short texts.

Dynamic topic models (DTM) are of great use toanalyze the evolution of unobserved topics of a text collectionover time. Recent years have witnessed the explosive growth ofstreaming text data emerging from online media, which createsan unprecedented need for DTMs for timely event analysis. While there have been some matrix factorization methods inthe literature for dynamic topic modeling, further study is stillin great need to model emerging, evolving and fading topicsin a more natural and effective way. In light of this, we firstpropose a matrix factorization model called SONMFSR (SoftOrthogonal NMF with Sparse Representation), which makes fulluse of soft orthogonal and sparsity constraints for static topicmodeling. Furthermore, by introducing the constraints of emerging, evolving and fading topics to SONMFSR, we easily obtain a novel DTM called SONMFSRd for dynamic event analysis. Extensive experiments on two public corpora demonstrate the superiority of SONMFSRd to some state-of-the-art DTMs in both topic detection and tracking. In particular, SONMFSRd shows great potential in real-world applications, where popular topics in Two Sessions 2015 are captured and traced dynamically for possible insights.