Heng Tian

Tsung-Ming Huang, Tiexiang Li, Wei-De Li et al.

Calculation of band structure of three dimensional photonic crystals amounts to solving large-scale Maxwell eigenvalue problems, which are notoriously challenging due to high multiplicity of zero eigenvalue. In this paper, we try to address this problem in such a broad context that band structure of three dimensional isotropic photonic crystals with all 14 Bravais lattices can be efficiently computed in a unified framework. We uncover the delicate machinery behind several key results of our work and on the basis of this new understanding we drastically simplify the derivations, proofs and arguments in our framework. In this work particular effort is made on reformulating the Bloch boundary condition for all 14 Bravais lattices in the redefined orthogonal coordinate system, and establishing eigen-decomposition of discrete partial derivative operators by systematic use of commutativity among them, which has been overlooked previously, and reducing eigen-decomposition of double-curl operator to the canonical form of a 3x3 complex skew-symmetric matrix under unitary congruence. With the validity of the novel nullspace free method in the broad context, we perform some calculations on one benchmark system to demonstrate the accuracy and efficiency of our algorithm.

Haibo Zhao, Yu Qi, Boce Hu et al.

We present Generalizable Hierarchical Skill Learning (GSL), a novel framework for hierarchical policy learning that significantly improves policy generalization and sample efficiency in robot manipulation. One core idea of GSL is to use object-centric skills as an interface that bridges the high-level vision-language model and the low-level visual-motor policy. Specifically, GSL decomposes demonstrations into transferable and object-canonicalized skill primitives using foundation models, ensuring efficient low-level skill learning in the object frame. At test time, the skill-object pairs predicted by the high-level agent are fed to the low-level module, where the inferred canonical actions are mapped back to the world frame for execution. This structured yet flexible design leads to substantial improvements in sample efficiency and generalization of our method across unseen spatial arrangements, object appearances, and task compositions. In simulation, GSL trained with only 3 demonstrations per task outperforms baselines trained with 30 times more data by 15.5 percent on unseen tasks. In real-world experiments, GSL also surpasses the baseline trained with 10 times more data.

Heng Tian

Existing approaches often enhance the performance of single-image super-resolution (SISR) methods by incorporating auxiliary structures, such as specialized loss functions, to indirectly boost the quality of low-resolution images. In this paper, we propose a plug-and-play module called Learnable Separable Kernels (LSKs), which are formally rank-one matrices designed to directly enhance image frequency components. We begin by explaining why LSKs are particularly suitable for SISR tasks from a frequency perspective. Baseline methods incorporating LSKs demonstrate a significant reduction of over 60\% in both the number of parameters and computational requirements. This reduction is achieved through the decomposition of LSKs into orthogonal and mergeable one-dimensional kernels. Additionally, we perform an interpretable analysis of the feature maps generated by LSKs. Visualization results reveal the capability of LSKs to enhance image frequency components effectively. Extensive experiments show that incorporating LSKs not only reduces the number of parameters and computational load but also improves overall model performance. Moreover, these experiments demonstrate that models utilizing LSKs exhibit superior performance, particularly as the upscaling factor increases.