Sijia Liu

Addressing ethical concerns through selective memory removal in AI models.

Zeroth-order optimization offers memory efficiency for large language models in NLP tasks.

New dataset enhances evaluation methods for machine unlearning in image generation.

A novel approach to find backdoor samples without needing clean data.

AdvUnlearn enhances image generation safety by combining unlearning with adversarial training.

UGradSL offers a practical solution for removing sensitive information from machine learning models.

A new method improves Large Language Models' ability to forget sensitive information.

A new framework evaluates how well language models recognize and respond to emotions.

BrainKing assesses language models' problem-solving skills under limited information.

Exploring vulnerabilities in face recognition caused by adversarial watermarking attacks.

Researchers find effective ways to remove unwanted knowledge from language models.