LaplacianFormer:Rethinking Linear Attention with Laplacian Kernel

Zhe Feng, Sen Lian, Changwei Wang, Muyang Zhang, Tianlong Tan, Rongtao Xu, Weiliang Meng, Xiaopeng Zhang

Keywords: Linear Attention, Transformer, Kernel learning

Abstract: The quadratic complexity of softmax attention presents a major obstacle for scaling Transformers to high-resolution vision tasks. Existing linear attention variants often replace the softmax with Gaussian kernels to reduce complexity, but such approximations lack theoretical grounding and tend to oversuppress mid-range token interactions. We propose LaplacianFormer, a Transformer variant that employs a Laplacian kernel as a principled alternative to softmax, motivated by empirical observations and theoretical analysis. To address expressiveness degradation under low-rank approximations, we introduce a provably injective feature map that retains fine-grained token information. For efficient computation, we adopt a Nyström approximation of the kernel matrix and solve the resulting system using Newton--Schulz iteration, avoiding costly matrix inversion and SVD. We further develop custom CUDA implementations for both the kernel and solver, enabling high-throughput forward and backward passes suitable for edge deployment. Experiments on ImageNet show that LaplacianFormer achieves strong performance-efficiency trade-offs while improving attention expressiveness. Code is available at the following site: \href{https://mike7472727.github.io/laplacianformer.github.io/}{\textcolor{black}{LaplacianFormer }}.

Primary Area: foundation or frontier models, including LLMs

Submission Number: 8404