Searcharxiv⌕ Search

arXiv subjects

Wenzheng Hu

Publications and source records attributed to Wenzheng Hu.

3 recordsLinked to original sources

Singular Global Compactness for Logarithmically Coupled \(p\)-Laplacian Systems

We establish global compactness for logarithmically coupled p-Laplacian systems on RN, where p=N>4 and bounded positive potentials converge to positive constants at infinity. The finite-energy domain need not be closed under subtraction, obstructing direct profile subtraction. We develop a scalar localisation method using slowly expanding cutoffs and residual commutator estimates uniform in the regularisation parameter. As the regularisation vanishes, residual Palais-Smale sequences split into a solution of the original system and finitely many separating autonomous profiles, with a strongly vanishing remainder and splitting of the logarithmic densities. A converse construction characterises all levels of compactness failure. Applications include convergence of ground-state levels, a compactness-escape alternative for minimising sequences, and a directional response formula for the ground-state energy.

math.AP↗

Logarithmically Coupled p-Laplacian Systems: From Discrete Ground States to Rigidity and Critical Continuous Limits

We introduce a new class of logarithmically coupled \(p\)-Laplacian systems on locally finite graphs and in the critical case \(p=N>4\). The coupling poses two challenges: the energy functional is ill-defined on the natural Sobolev space, and loss of the \(L^\infty\)-embedding prevents standard control of logarithmic terms. To resolve this non-separable logarithmic singularity, we develop an \textbf{exponent calibration technique} that converts logarithmic terms into strictly lower-order power estimates. The technique plays an essential role in two critical steps: the existence proof for the regularised problem and the uniform boundedness of regularised ground states that restores compactness. Thus, we prove solvability of the original ill-posed continuous critical equation. In the discrete setting, we identify five new intrinsic structures, including an exact constrained Hessian factorisation and a scalar rigidity index. We complete a discrete--continuous double loop, providing a unified variational framework for logarithmically coupled systems.

math.AP↗

CATRO: Channel Pruning via Class-Aware Trace Ratio Optimization

Deep convolutional neural networks are shown to be overkill with high parametric and computational redundancy in many application scenarios, and an increasing number of works have explored model pruning to obtain lightweight and efficient networks. However, most existing pruning approaches are driven by empirical heuristic and rarely consider the joint impact of channels, leading to unguaranteed and suboptimal performance. In this paper, we propose a novel channel pruning method via Class-Aware Trace Ratio Optimization (CATRO) to reduce the computational burden and accelerate the model inference. Utilizing class information from a few samples, CATRO measures the joint impact of multiple channels by feature space discriminations and consolidates the layer-wise impact of preserved channels. By formulating channel pruning as a submodular set function maximization problem, CATRO solves it efficiently via a two-stage greedy iterative optimization procedure. More importantly, we present theoretical justifications on convergence of CATRO and performance of pruned networks. Experimental results demonstrate that CATRO achieves higher accuracy with similar computation cost or lower computation cost with similar accuracy than other state-of-the-art channel pruning algorithms. In addition, because of its class-aware property, CATRO is suitable to prune efficient networks adaptively for various classification subtasks, enhancing handy deployment and usage of deep networks in real-world applications.

cs.LG↗