This paper analyzes PEFT of LLMs via linearization, showing how an explicit inductive bias connects fine-tuning dynamics to NTKs and offering theoretical and empirical insights into when linearization accurately predicts adaptation performance.

We approximate ReLU networks for hidden-state DP analysis; the results on practical noisy cyclic GD are on par with DP-SGD

VLMLight leverages vision-language scene understanding and dual-branch reasoning to achieve safe and efficient traffic signal control, especially in critical scenarios.