We propose a Weak-form Physics-Informed Neural Operator (WINO), a data-free framework that combines the efficiency of neural operators with the geometric flexibility of the φ-finite element method (φ-FEM). φ-FEM is an unfitted method that accommodates geometric variations without body-fitted meshes, where the domain geometry is represented by the level-set function φ. To impose the boundary conditions, Dirichlet problems adopt the φ-FEM lifting so only the homogeneous displacement contribution is learned, whereas traction-driven Neumann problems additionally predict the auxiliary fields necessary for the unfitted weak formulation. Parameters are trained by minimizing squared weak-form residuals aligned with φ-FEM together with squared penalties on the cut-cell auxiliary equations, which removes the need for large paired datasets of converged reference solutions. When labeled reference data are available, an optional data-augmented variant (WINO+data) can further combine this physics-informed loss with a supervised term. After training, WINO outputs can seed the nonlinear φ-FEM solvers as neural operator warm starts (NOWS), which reduce iteration counts relative to traditional cold-started solvers. Numerical benchmarks show substantial accuracy of WINO together with total training times of about 15%–70% of those of supervised φ-FEM-FNO across all cases, without requiring reference-solution generation.
Image deblurring aims to restore high-quality images from blurry ones. Although existing methods have achieved remarkable progress in restoration performance, most approaches still mainly rely on designing complex modules in the spatial domain, leading to the gradual saturation of performance improvements. This paper proposes a novel CNN–Transformer hybrid architecture for image deblurring, dubbed DSCTNet, by analyzing the differences between blurry and sharp images in the fourier and wavelet domains. Specifically, DSCTNet comprises two stages. The first stage is the local restoration stage, which progressively removes different levels of blur through multiple iterations. During this process, the multi-scale features generated at each iteration are fused into enhanced prior features, which are then forwarded along with the restored outputs to the second stage. The second stage is the global restoration stage, which employs a dual-branch design in the spatial and wavelet domains and incorporates a Transformer-based self-attention mechanism to capture long-range dependencies. By leveraging the prior features from the first stage, this stage effectively compensates for the Transformer’s limitations in local modeling, further improving structural restoration and detail reconstruction based on the preceding stage’s results. Additionally, a dual-domain feature enhancement module is designed to extract features at different frequencies in the fourier domain and features with multiple receptive fields in the spatial domain, subsequently fusing these features to enhance representational richness and diversity. Extensive experiments demonstrate that the proposed DSCTNet achieves superior image restoration performance compared with existing state-of-the-art methods.
Multimodal learning aims to exploit cross-modal complementarity by integrating heterogeneous modalities such as text, audio, and vision. However, multimodal models often suffer from the imbalanced multimodal learning problem, where the dominant modality monopolizes optimization resources while weaker modalities remain insufficiently learned. Empirical studies show that, unlike unimodal training, gradients across modalities exhibit highly similar evolution trends in multimodal settings, even with existing balancing strategies, indicating that gradient magnitude adjustment alone fails to address the underlying optimization dynamics. In this work, we reanalyze imbalanced multimodal learning from an optimization perspective by incorporating Cognitive Load Theory. We show that, under limited model capacity, jointly optimizing multiple modalities induces a persistent high load state, where gradients from different modalities compete within a constrained update subspace, resulting in gradient conflicts and a persistent imbalanced state. Inspired by human cognitive behavior, we propose a multimodal balanced learning framework that integrates a plug-and-play multimodal alternating learning strategy to alleviate the learning load of simultaneous multimodal optimization and a Gaussian noise based optimizer to prevent overly similar gradient magnitude change trends. Extensive experiments on IEMOCAP, CMU-MOSEI, and AVE demonstrate the rationality and effectiveness of the proposed framework.
Drone delivery, as a low-altitude logistics application of urban air mobility (UAM), may reshape last-mile distribution and urban service accessibility. However, its societal diffusion cannot be fully understood through conventional binary acceptance models, because consumers may differ substantially in their readiness to progress from awareness to repeated use. This study investigates heterogeneous consumer readiness for drone delivery using 1,018 valid survey responses collected in Shenzhen, China, where urban drone delivery has entered early deployment. Drawing on the Transtheoretical Model, adoption is conceptualized as an ordered five-stage readiness process: precontemplation, contemplation, preparation, action, and maintenance. An integrated framework combining the Technology Acceptance Model and the Theory of Planned Behavior, extended by environmental concern and perceived risk, is used to identify attitudinal latent constructs. A hybrid model integrating a Multiple Indicators and Multiple Causes model and a Partial Proportional Odds Model examines how socio-demographic attributes and psychological perceptions shape stage-specific readiness, with marginal effects quantifying heterogeneous influences across stages. Results show that drone delivery adoption is not a linear extension of general acceptance but a staged readiness process shaped by different psychological mechanisms. Socio-demographic attributes mainly explain baseline differences in adoption propensity, while attitudinal constructs exert stronger and more stage-dependent effects. Early-stage readiness is constrained by perceived risk and uncertainty, but is also sensitive to social influence and environmental concern; progression toward action and maintenance is more closely associated with perceived usefulness, ease of use, perceived behavioral control, attitude, and behavioral intention. This pattern indicates a pathway from risk reduction and cognitive activation to technology evaluation, experiential confidence, and behavioral consolidation. Findings provide behavioral evidence for stage-targeted risk communication, service design, and demand cultivation strategies in early urban drone delivery deployment.
During seismic events, the underground components of bridge pile foundations are particularly susceptible to damage. Because subsurface damage cannot be directly observed, assessing the functional degradation of affected bridges becomes both time-consuming and uncertain. Therefore, reliable and rapid assessment of post-earthquake pile bearing capacity is critical for seismic recovery. To address this issue, this study proposes an innovative framework that integrates an attention-enhanced gated recurrent unit (AttGRU) network with the beam on nonlinear Winkler foundation (BNWF) approach to calibrate the residual load-carrying capacity of piles after earthquakes. The AttGRU network serves as a metamodel to accurately reproduce the nonlinear seismic time-history responses of the pile–soil system. The data-driven predictions of soil pressure and soil–pile relative displacement are used to modify the p–y curves in the BNWF model. The predicted pile-head inertial force and soil displacement time histories are then quasi-statically applied to the modified BNWF model to capture pile damage induced by both inertial and kinematic loading mechanisms. Subsequent pushdown analysis of the damaged specimens under a suite of representative ground motions is conducted to evaluate the post-earthquake vertical bearing capacity. Finally, an optimal degradation model of vertical bearing capacity is developed.
