Search papers, labs, and topics across Lattice.
Wuhan University
4
0
7
PixJail achieves a remarkable 2.1% error rate in reproducing T2I jailbreak results, setting a new standard for evaluation consistency in a fast-paced research area.
Watermarking VLA and WAM models can now be done without sacrificing performance or revealing detectable signals to adversaries.
Adversarial attacks on ASR systems can achieve a +26.6 WER improvement by targeting feature representations instead of raw audio, exposing a significant blind spot in current robustness evaluations.
Current MLLMs fail to detect covert advertisements, revealing a critical gap in social media moderation that could mislead consumers and pose ethical risks.
