Searcharxiv⌕ Search

arXiv subjects

Shunqin Zhang

Publications and source records attributed to Shunqin Zhang.

3 recordsLinked to original sources

DyTransRM: A Unified Simulation-to-Real Framework for Dynamic Radio Map Prediction

Dynamic radio map (RM) prediction enables proactive wireless network control by forecasting future propagation conditions from historical observations. However, collecting temporally continuous and spatially dense real-world RMs is prohibitively expensive, while models trained on large-scale simulated data often suffer from limited transferability to real environments. To address this issue, we propose DyTransRM, a unified simulation-to-real framework for dynamic RM prediction under sparse real-world measurements. DyTransRM first introduces RadioCLIP to learn physics-semantic representations by aligning environmental evolution with the corresponding propagation evolution, thereby improving both prediction accuracy and cross-domain transferability. To further bridge the simulation-to-real gap, SparseUOT employs sparse-mask simulation refinement to mimic realistic measurement patterns and unbalanced optimal transport to selectively align compatible propagation representations across domains. Experiments on RadioMapMotion and real-world datasets demonstrate that RadioCLIP achieves an NMSE of 0.0018 and an RMSE of 0.012 in simulation-domain prediction. After sparse real-domain adaptation, DyTransRM achieves an NMSE of 0.023, RMSE of 0.079, and MAE of 0.064, outperforming representative adaptation baselines without requiring spatially complete real RMs. These results demonstrate the effectiveness of DyTransRM for transferring simulation-learned dynamic propagation knowledge to practical sparse-measurement scenarios.

eess.SP↗

Global well-posedness and scattering of energy-critical defocusing nonlinear Schrödinger equation on high-dimensional hyperbolic spaces

We prove global well-posedness and scattering for the energy-critical defocusing nonlinear Schrödinger equation \begin{equation*} \begin{cases} i\partial_t u + Δu = |u|^{\frac{4}{d-2}} u, u(0) = u_0, \end{cases} \end{equation*} on hyperbolic space $\mathbb{H}^d(d \geq 4)$. We mainly utilize the concentration-compactness/rigidity method. We establish a linear profile decomposition on high-dimensional hyperbolic spaces and prove that the nonlinear profile on hyperbolic spaces can be approximated by the solutions of the energy-critical defocusing nonlinear Schrödinger equation on Euclidean spaces, then obtaining the global well-posedness and scattering of the energy-critical defocusing nonlinear Schrödinger equation on the high-dimensional hyperbolic space.

math.AP↗

Deep Transformation Model

There has been a significant recent surge in deep neural network (DNN) techniques. Most of the existing DNN techniques have restricted model formats/assumptions. To overcome their limitations, we propose the nonparametric transformation model, which encompasses many popular models as special cases and hence is less sensitive to model mis-specification. This model also has the potential of accommodating heavy-tailed errors, a robustness property not broadly shared. Accordingly, a new loss function, which fundamentally differs from the existing ones, is developed. For computational feasibility, we further develop a double rectified linear unit (DReLU)-based estimator. To accommodate the scenario with a diverging number of input variables and/or noises, we propose variable selection based on group penalization. We further expand the scope to coherently accommodate censored survival data. The estimation and variable selection properties are rigorously established. Extensive numerical studies, including simulations and data analyses, establish the satisfactory practical utility of the proposed methods.

stat.ME↗