Self-Calibrating Dense Displacement Fields for Reliable Co-Registration of Large Optical Satellite Imagery
本文提出SCDF方法，无需训练和GPU，通过预测-测量-过滤循环处理大尺寸光学卫星图像配准问题，提高精度并减少误差。
0 citationsRead paper
Selected work
Representative Papers
U-CFR: Uncertainty-Guided Cascade Forward Refinement for Interactive Segmentation
This work addresses the limitations of existing interactive image segmentation methods, which typically require numerous user clicks and exhibit slow convergence. The authors propose a self-correction framework operating at inference time that automatically generates internal pseudo-clicks by integrating segmentation uncertainty, contour gradients, and explicit edge predictions into a boundary-aware uncertainty scoring mechanism. A dual-head network architecture with a shared encoder–decoder is introduced to jointly optimize region consistency and boundary sharpness, enabling a cascaded forward refinement process. This approach substantially improves initial mask quality, boundary accuracy, and click efficiency, reducing the number of required user interactions by over 10% on challenging benchmarks such as the Berkeley dataset.
0 citationsRead paper
Machine Learning-based Correlation of Charpy Impact Properties Between Sub-sized and Standard-sized Specimens for Nuclear Structural Materials
This study addresses the challenge of accurately correlating impact toughness between miniature and standard Charpy specimens in nuclear structural material assessment, where spatial and material constraints often necessitate the use of small-scale samples yet lack a high-precision, universally applicable mapping method. The authors propose a machine learning–based cross-specimen-size framework that aligns the impact energy curves of miniature specimens to the standard response through temperature shifting and scaled residual projection, followed by hyperbolic tangent model fitting to extract the upper shelf energy (USE) and ductile-to-brittle transition temperature (DBTT). Notably, this approach enables full-transition-region correlation without requiring any reference data from standard specimens, making it suitable for material surveillance and accelerated irradiation testing. Validated on 389 datasets of SA533B steel, the method achieves R² values of 0.942 and 0.892 for USE and DBTT predictions, respectively, substantially outperforming conventional analytical techniques.
0 citationsRead paper
Differentiate the Evaluator, Not the Program: An Efficient Runtime Representation for Neuro-Symbolic Learning
This work addresses the inefficiency of joint program-and-parameter search in neuro-symbolic learning, where each candidate program requires separate parameter optimization. To overcome this bottleneck, the authors propose the Neural Differentiable Virtual Machine (NDVM), which applies automatic differentiation to the interpreter rather than individual programs. By decoupling symbolic structure from differentiable numeric state, NDVM preserves program dynamics while enabling precise backpropagation through execution traces. The design incorporates dense batched numeric buffers and runtime symbolic environment management, substantially amortizing evaluation overhead. Experiments demonstrate that NDVM achieves approximately 60× amortized speedup per-channel batching over baseline methods, exhibits near-linear multi-core scaling, and accelerates the discovery of high-quality solutions by roughly 24× under a fixed computational budget.
0 citationsRead paper
Superspace Concentration and Adversarial Robustness in Quantum Algorithms
This work investigates how to quantify the capacity of quantum systems to concentrate information in an extended degree-of-freedom space to enhance adversarial robustness. It introduces a focus measure \( F(\rho) \) and establishes, for the first time, a complete resource theory of hyperspace concentration, rigorously distinguishing it operationally from \( U(d_S) \)-asymmetry. The study further uncovers a direct link between \( F(\rho) \) and the success probability of the marked state in Grover’s algorithm. Through GPU-accelerated simulations across six system configurations and over ten thousand random states, \( F(\rho) \) demonstrates strict monotonicity under various noise channels with no violations observed; analytical decoherence predictions achieve an accuracy of \( 1.11 \times 10^{-16} \). Experiments show that focused states maintain \( F > 0.9 \) even under attack strength \( \varepsilon = 0.302 \), substantially outperforming conventional fidelity-based metrics (\( \varepsilon = 0.174 \)), with the focus capacity gap \( \Delta F \) obeying a \( \log_2(d_S) \) scaling law.
0 citationsRead paper
Recent publications
Latest Papers
Self-Calibrating Dense Displacement Fields for Reliable Co-Registration of Large Optical Satellite Imagery
本文提出SCDF方法，无需训练和GPU，通过预测-测量-过滤循环处理大尺寸光学卫星图像配准问题，提高精度并减少误差。
0 citationsRead paper
U-CFR: Uncertainty-Guided Cascade Forward Refinement for Interactive Segmentation
This work addresses the limitations of existing interactive image segmentation methods, which typically require numerous user clicks and exhibit slow convergence. The authors propose a self-correction framework operating at inference time that automatically generates internal pseudo-clicks by integrating segmentation uncertainty, contour gradients, and explicit edge predictions into a boundary-aware uncertainty scoring mechanism. A dual-head network architecture with a shared encoder–decoder is introduced to jointly optimize region consistency and boundary sharpness, enabling a cascaded forward refinement process. This approach substantially improves initial mask quality, boundary accuracy, and click efficiency, reducing the number of required user interactions by over 10% on challenging benchmarks such as the Berkeley dataset.
0 citationsRead paper
Machine Learning-based Correlation of Charpy Impact Properties Between Sub-sized and Standard-sized Specimens for Nuclear Structural Materials
This study addresses the challenge of accurately correlating impact toughness between miniature and standard Charpy specimens in nuclear structural material assessment, where spatial and material constraints often necessitate the use of small-scale samples yet lack a high-precision, universally applicable mapping method. The authors propose a machine learning–based cross-specimen-size framework that aligns the impact energy curves of miniature specimens to the standard response through temperature shifting and scaled residual projection, followed by hyperbolic tangent model fitting to extract the upper shelf energy (USE) and ductile-to-brittle transition temperature (DBTT). Notably, this approach enables full-transition-region correlation without requiring any reference data from standard specimens, making it suitable for material surveillance and accelerated irradiation testing. Validated on 389 datasets of SA533B steel, the method achieves R² values of 0.942 and 0.892 for USE and DBTT predictions, respectively, substantially outperforming conventional analytical techniques.
0 citationsRead paper
Differentiate the Evaluator, Not the Program: An Efficient Runtime Representation for Neuro-Symbolic Learning
This work addresses the inefficiency of joint program-and-parameter search in neuro-symbolic learning, where each candidate program requires separate parameter optimization. To overcome this bottleneck, the authors propose the Neural Differentiable Virtual Machine (NDVM), which applies automatic differentiation to the interpreter rather than individual programs. By decoupling symbolic structure from differentiable numeric state, NDVM preserves program dynamics while enabling precise backpropagation through execution traces. The design incorporates dense batched numeric buffers and runtime symbolic environment management, substantially amortizing evaluation overhead. Experiments demonstrate that NDVM achieves approximately 60× amortized speedup per-channel batching over baseline methods, exhibits near-linear multi-core scaling, and accelerates the discovery of high-quality solutions by roughly 24× under a fixed computational budget.
0 citationsRead paper
Superspace Concentration and Adversarial Robustness in Quantum Algorithms
This work investigates how to quantify the capacity of quantum systems to concentrate information in an extended degree-of-freedom space to enhance adversarial robustness. It introduces a focus measure \( F(\rho) \) and establishes, for the first time, a complete resource theory of hyperspace concentration, rigorously distinguishing it operationally from \( U(d_S) \)-asymmetry. The study further uncovers a direct link between \( F(\rho) \) and the success probability of the marked state in Grover’s algorithm. Through GPU-accelerated simulations across six system configurations and over ten thousand random states, \( F(\rho) \) demonstrates strict monotonicity under various noise channels with no violations observed; analytical decoherence predictions achieve an accuracy of \( 1.11 \times 10^{-16} \). Experiments show that focused states maintain \( F > 0.9 \) even under attack strength \( \varepsilon = 0.302 \), substantially outperforming conventional fidelity-based metrics (\( \varepsilon = 0.174 \)), with the focus capacity gap \( \Delta F \) obeying a \( \log_2(d_S) \) scaling law.
0 citationsRead paper