Qi Li

A look into the safety concerns surrounding large language models.

CleanSheet advances model hijacking without altering training processes.

A new system enhances detection of Internet routing anomalies using BGP semantics.

Ruledger provides a secure solution for smart home system vulnerabilities.

This framework allows collaborative model training while ensuring data privacy and model integrity.

A new dataset improves action recognition for badminton, aiding performance analysis.

Dust and gas play crucial roles in shaping galaxies and star formation.

Machine unlearning allows models to forget data, essential for data privacy compliance.

A novel approach to protect image privacy in technology while maintaining model performance.

PrintListener exposes vulnerabilities in fingerprint authentication through sound analysis.

An overview of testing methods for deep learning libraries and their significance.

Examining the risks and defenses against JNI Global Reference attacks in Android.

New methods improve simulations of gas mixtures in low-pressure environments.

MIVPG improves how models interpret images and text together.

A method to forecast non-factual answers from language models before they generate responses.

New methods improve 3D models from freehand ultrasound images.

This study examines watermarking methods for machine-generated text and their effectiveness against removal attacks.

Examine various jailbreak attacks on language models and their defenses.

KiOP offers a seamless way to share knowledge from multiple models.

A new framework enhances privacy and data quality in federated learning.

Insights into star formation and dust reveal galaxy evolution.

New method combines expert knowledge with large models to analyze industrial vibration signals.

This article discusses fairness issues in Graph Neural Networks and introduces a new method.

A framework combining structural and semantic information enhances knowledge graph completion.

This guide discusses keeping data safe while training AI effectively.

Learn how DWCL improves data grouping through quality views.

Learn how smart systems organize complex data efficiently.

A new method improves AI's ability to answer complex questions accurately.

CrOCl shows promise for smarter, energy-saving technologies through unique magnetic properties.

Discover how AVATAR cleverly disguises harmful intents in language models.

Engorgio exploits language models, raising concerns about service reliability and performance.