Back to feed
arXiv cs.LG
arXiv cs.LG
5/13/2026
Vertex-Softmax: Tight Transformer Verification via Exact Softmax Optimization

Vertex-Softmax: Tight Transformer Verification via Exact Softmax Optimization

Short summary

Vertex-Softmax provides a new algorithm for certified verification of transformer attention with significantly tighter bounds than existing approaches. It proves optimal softmax bounds occur at constraint box vertices, enabling log-linear complexity search. Evaluated on MNIST, Fashion-MNIST, and CIFAR-10, it outperforms alpha-CROWN and branch-and-bound baselines while maintaining formal soundness guarantees.

  • •New Vertex-Softmax algorithm for tight transformer attention verification
  • •Reduces search complexity to log-linear by proving optimum occurs at box vertices
  • •Outperforms alpha-CROWN and branch-and-bound on image classification benchmarks

Generated with AI, which can make mistakes.

#research-breakthrough#ai-tools
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Explore more

Rotation-Preserving Supervised Fine-Tuning
arXiv cs.LG
The Safety-Aware Denoiser for Text Diffusion Models
arXiv cs.LG
Where Reliability Lives in Vision-Language Models: A Mechanistic Study of Attention, Hidden States, and Causal Circuits
arXiv CS.AI
When Does a Language Model Commit? A Finite-Answer Theory of Pre-Verbalization Commitment
arXiv CS.AI