Back to feed
arXiv cs.LG
arXiv cs.LG
7/10/2026
LiST: Lipschitz Scaling Training for Robust and Calibrated Neural Networks

LiST: Lipschitz Scaling Training for Robust and Calibrated Neural Networks

Short summary

Researchers introduce Lipschitz Scaling Training (LiST), a novel method that simultaneously optimizes neural networks for accuracy, robustness, and calibration—three properties typically difficult to achieve together. By iteratively adjusting Lipschitz constraints and connecting them to temperature scaling, LiST creates out-of-the-box calibrated models without manual tuning. Validated on CIFAR-10/100 and Tiny-ImageNet with competitive results, code available on GitHub.

  • •LiST solves the core challenge of balancing accuracy, robustness, and calibration in neural networks simultaneously
  • •Method iteratively adjusts Lipschitz constraints linked to temperature scaling for principled operating-point selection
  • •Achieves competitive performance on standard benchmarks with full code available; data can be reintegrated for improved sample efficiency

Generated with AI, which can make mistakes.

#research-breakthrough#ai-tools#open-source#certification-education
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

When Calibration Rankings Reverse: Accuracy-Controlled Evaluation for Fair Comparison of LLMs
arXiv cs.CL
Gradient Smoothing: Coupling Layer-wise Updates for Improved Optimization
arXiv cs.LG
When Implausible Tokens Get Reinforced: Tail-Aware Credit Calibration for LLM Reinforcement Learning
arXiv cs.CL
Operator-Aware Mixed-Precision Tolerance Calibration for Tensor Kernels
arXiv cs.LG