Deep learning from the crowd Fundamentals of morphological galaxy classification
本文通过调整深度神经网络模型，利用众包注释进行星系形态分类，探讨了不同训练方法对准确性和效率的影响。
0 citationsRead paper
Selected work
Representative Papers
Emergence of Complex Structures
This work addresses the apparent tension between structural order and the principle of entropy increase by elucidating how complex structures spontaneously emerge from homogeneous or weakly correlated initial states. Integrating semi-microscopic phase-space dynamics, transport geometry, information theory, and coarse-grained modeling, the study proposes a geometric framework based on Lagrangian–Eulerian transport maps and the spectral properties of deformation tensors to unify the description of nonlocality, anisotropy, and self-organization. Within this framework, a Landau–Ginzburg-type effective free energy theory is constructed, revealing mechanisms of density amplification and anisotropic collapse. Notably, the approach identifies nonlocal tidal effects as the dominant driver of structure formation already at moderate superdensities, successfully generating cosmic density fields that exhibit both non-Gaussian statistics and multiscale coupling.
0 citationsRead paper
A High-Throughput AES-GCM Implementation on GPUs for Secure, Policy-Based Access to Massive Astronomical Catalogs
This work addresses the challenge of reconciling fine-grained access control with high-performance encryption in pre-release large-scale astronomical image catalogs. The authors propose a novel framework that integrates a flexible policy engine with GPU-accelerated AES-GCM authenticated encryption, introducing parallel tree reduction into GHASH computation for the first time. This innovation transforms the traditionally sequential authentication hash process into a logarithmic-time parallel operation, substantially enhancing encryption throughput for petabyte-scale astronomical datasets. The approach ensures both confidentiality and integrity while enabling efficient and secure transition toward FAIR-compliant public archives.
0 citationsRead paper
Recent publications
Latest Papers
Deep learning from the crowd Fundamentals of morphological galaxy classification
本文通过调整深度神经网络模型，利用众包注释进行星系形态分类，探讨了不同训练方法对准确性和效率的影响。
0 citationsRead paper
Emergence of Complex Structures
This work addresses the apparent tension between structural order and the principle of entropy increase by elucidating how complex structures spontaneously emerge from homogeneous or weakly correlated initial states. Integrating semi-microscopic phase-space dynamics, transport geometry, information theory, and coarse-grained modeling, the study proposes a geometric framework based on Lagrangian–Eulerian transport maps and the spectral properties of deformation tensors to unify the description of nonlocality, anisotropy, and self-organization. Within this framework, a Landau–Ginzburg-type effective free energy theory is constructed, revealing mechanisms of density amplification and anisotropic collapse. Notably, the approach identifies nonlocal tidal effects as the dominant driver of structure formation already at moderate superdensities, successfully generating cosmic density fields that exhibit both non-Gaussian statistics and multiscale coupling.
0 citationsRead paper
A High-Throughput AES-GCM Implementation on GPUs for Secure, Policy-Based Access to Massive Astronomical Catalogs
This work addresses the challenge of reconciling fine-grained access control with high-performance encryption in pre-release large-scale astronomical image catalogs. The authors propose a novel framework that integrates a flexible policy engine with GPU-accelerated AES-GCM authenticated encryption, introducing parallel tree reduction into GHASH computation for the first time. This innovation transforms the traditionally sequential authentication hash process into a logarithmic-time parallel operation, substantially enhancing encryption throughput for petabyte-scale astronomical datasets. The approach ensures both confidentiality and integrity while enabling efficient and secure transition toward FAIR-compliant public archives.
0 citationsRead paper