MOTIVATION: Accurate survival prediction is crucial for personalized cancer treatment but remains challenging for rare cancers due to limited data. Most deep learning models require large training datasets, which are unavailable for rare cancer types, creating a significant clini-cal bottleneck. RESULTS: We propose MoESurv, a zero-sample survival prediction framework that leverages a mix-ture-of-experts architecture to extract generalizable prognostic patterns from pan-cancer data. The model integrates shared experts, cancer-specific experts, and routing experts within an autoencoder to disentangle common and type-specific survival features. Evaluated on seven rare TCGA cancer types, MoESurv achieved state-of-the-art performance, improving the average C-index by 4 percentage points over the best baseline. Further-more, external validation across diverse populations and independent cohorts-including a Chinese glioma cohort (CGGA mRNAseq_693, C-index = 0.7433), a rare GBM IDH-mutant subtype (C-index = 0.8064), and the pan-cancer PCAWG cohort (C-index = 0.7090)-demonstrated that MoESurv possesses the most robust predictive per-formance, highlighting its generalizability. MoESurv also effectively stratified high- and low-risk patient groups and identified potential survival-associated genes, demonstrating both clinical utility and biological interpretability. AVAILABILITY: The code is freely available at https://github.com/HuaYC666/MoESurv and https://zenodo.org/records/20785500.

Uncertainty quantification (UQ) in scientific machine learning is increasingly critical as neural networks are widely adopted to tackle complex problems across diverse scientific disciplines. For physics-informed neural networks (PINNs), a prominent model in scientific machine learning, uncertainty is typically quantified using Bayesian or dropout methods. However, both approaches suffer from a fundamental limitation: the prior distribution or dropout rate required to construct honest confidence sets cannot be determined without additional information. In this paper, we propose a novel method within the framework of extended fiducial inference (EFI) to provide rigorous uncertainty quantification for PINNs. The proposed method leverages a narrow-neck hyper-network to learn the parameters of the PINN and quantify their uncertainty based on imputed random errors in the observations. This approach overcomes the limitations of Bayesian and dropout methods, enabling the construction of honest confidence sets based solely on observed data. This advancement represents a significant breakthrough for PINNs, greatly enhancing their reliability, interpretability, and applicability to real-world scientific and engineering challenges. Moreover, it establishes a new theoretical framework for EFI, extending its application to large-scale models, eliminating the need for sparse hyper-networks, and significantly improving the automaticity and robustness of statistical inference.

Remote sensing change detection (CD) has garnered extensive research and application due to its ability to identify changes in land features within the same area across different periods. CD tasks require features with strong intraclass distinctions and precise spatial boundary details. Existing methods enhance the extraction of difference features but significantly increase computational complexity in high-resolution remote sensing imagery. Moreover, these methods focus on pixel-level difference extraction while neglecting feedback from the overall change object. As a result, they lack global information perception, leading to blurred edges and fragmented interiors in the change areas. To address these challenges, we propose a feature enhancement and feedback network (FEFNet) for CD. First, we designed a multilevel dual-feature fusion enhancement module (DFFM) to improve the representation of latent features between the bitemporal images. Second, we developed a feature coupling feedback module (FCFM) that efficiently decodes multiscale change features to generate extraction results. The experimental results show that FEFNet outperforms recent models in both computational efficiency and detection performance. With only 10.56G FLOPs and 2.26M parameters, FEFNet achieves an${F}_{1}$score of 92.32% on the LEVIR-CD dataset and 93.77% on the WHU-CD dataset. The code will be available athttps://github.com/XiaoJ058/RS-CD.

Semantic Change Detection (SCD) plays a critical role in intelligent information extraction from remote sensing images. However, its performance in detecting weak-information targets, such as small-sized ground objects, is still constrained by semantic representation ambiguity, information loss, and complex background interference. This paper proposes a Semantic Guidance Inference Network (SGI-Net) that establishes a collaborative optimization mechanism through semantic-driven feature coupling, adaptive reasoning, and attention guidance to overcome these challenges. Specifically, we design a Hierarchical Feature Coupling Module (HFCM). This module enhances geometric detail representation for weak-information targets through cross-layer feature aggregation and dual-path decoupling. It generates discriminative semantic priors, establishing geometric-semantic correlations across multi-level features. Furthermore, a Multi-dimensional Adaptive Inference Module (MAIM) is developed to achieve adaptive feature aggregation of local-global patterns via non-overlapping patch partitioning, preserving spatial continuity of weak-information targets through multi-scale feature fusion. Additionally, the Change-Guided Attention (CGA) module dynamically adjusts channel-wise and spatial attention weights based on prior knowledge, amplifying cross-hierarchical responses to weak change signals. Experimental results on the SECOND and SJH_SCD datasets demonstrate that SGI-Net achieves values of 73.86% and 77.43%, respectively, surpassing state-of-the-art (SOTA) methods by 0.73-1.37% in mIoU and 1.26-1.77% in SeK. Visualization analysis further confirms that SGI-Net effectively enhances the feature responses of weak-information targets in complex object scenes, improving the discrimination accuracy of fine-grained change boundaries. These results provide reliable technical support for high-precision remote sensing change detection.

Singing voice synthesis and singing voice conversion have significantly advanced, revolutionizing musical experiences. However, the rise of "Deepfake Songs" generated by these technologies raises concerns about authenticity. Unlike Audio DeepFake Detection (ADD), the field of song deepfake detection lacks specialized datasets or methods for song authenticity verification. In this paper, we initially construct a Chinese Fake Song Detection (FSD) dataset to investigate the field of song deepfake detection. The fake songs in the FSD dataset are generated by five state-of-the-art singing voice synthesis and singing voice conversion methods. Our initial experiments on FSD revealed the ineffectiveness of existing speech-trained ADD models for the task of song deepfake detection. Thus, we employ the FSD dataset for the training of ADD models. We subsequently evaluate these models under two scenarios: one with the original songs and another with separated vocal tracks. Experiment results show that song-trained ADD models exhibit a 38.58% reduction in average equal error rate compared to speech-trained ADD models on the FSD test set.