Search papers, labs, and topics across Lattice.
Tony
4
0
9
0
Forget static defenses: CEMMA's co-evolutionary approach uses genetic algorithms to evolve jailbreaking attacks, creating a constantly adapting red-teaming strategy that significantly boosts attack success rates and strengthens model robustness.
A hybrid GAN architecture with RoPE transformers and dual-path RNNs achieves state-of-the-art music source restoration with only 7.1M parameters.
The trustworthiness of LLM-enabled applications hinges not on further model improvements, but on establishing system-level threat monitoring to detect post-deployment anomalies.
By explicitly disentangling target features with MLLM guidance, MeGU achieves superior unlearning performance without sacrificing model utility, outperforming existing methods that struggle with the inherent entanglement of semantic concepts in model representations.
