Yusuke Yamasaki

Robust Backdoor Removal by Reconstructing Trigger-Activated Changes in Latent Representation

Plausible Token Amplification for Improving Accuracy of Differentially Private In-Context Learning Based on Implicit Bayesian Inference