Unverified50% confidenceFactExact time
This study evaluated the dLLM mapping method under two attack settings: sampling-based RareTrap and optimization-based ThinkTrap.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/23/2026
First Seen
Valid until: 1/21/2027
Sources
Related Entities
Related Claims
UnverifiedThe method proposed in this study consists of two steps: the first step is context anomaly detection, where the LLM simultaneously reasons over both text and sensor modalities; the second step is generating a binary mask, converting the judgment results into a filtering operation on the training dataset.71% similarUnverifiedThe paper conducted experiments on two tasks, a synthetic label-efficient prediction game and LLM routing, and E2D.PMSide outperforms previous methods on both tasks.71% similarUnverifiedA second engineering implementation recommendation: when evaluating fine-tuned checkpoints, not only test the target capability B₁ but also test against a decoupled matrix of adversarial behaviors, making 'whether unrelated capability drift appears after fine-tuning' a routine metric.70% similarUnverifiedPGD（Projected Gradient Descent）通过多步迭代生成更强的对抗样本，C&W攻击通过优化框架寻找最小扰动量，AutoAttack提供标准化鲁棒性评估基准68% similarUnverifiedThe study systematically scanned 384 hyperparameter configurations on modular arithmetic tasks using a two-hidden-layer MLP.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1088161API
curl https://kongchang.com/api/v1/knowledge/claims/1088161MCP
get_claim(id=1088161)