TriLens: Per-Layer Logit-Lens Entropy for White-Box Hallucination Detection

Bohan Yang, Yijun Gong, Zhi Zhang, Ge Zhang, Wenpeng Xing^#, Meng Han^#

EMNLP'26 First Author Findings of EMNLP 2026

Reads MHSA, FFN, and residual-stream outputs at every layer through the model's own logit lens, then compresses their entropy trajectories into a 3L-dimensional representation. The three module-wise signals provide complementary evidence for hallucination detection without storing hidden states or sampling multiple generations.

[Paper] [Project]

DECENT: Distribution-aware Prototypical Calibration with Fine-grained Voting for VLM Misclassification Detection

Yijun Gong*, Bohan Yang*, Zhi Zhang, Yupeng Qin, Wenpeng Xing, Changting Lin, Meng Han, Ziyue Qiao^#, Xiao Luo^#

EMNLP'26 Co-First Author EMNLP 2026 (Main Conference)

Calibrates VLM predictions against distribution-aware class prototypes and aggregates fine-grained voting evidence across the decision space. The two-stage design targets subtle misclassifications that may remain hidden behind a single confident output.

Latent Fusion Jailbreak: Blending Harmful and Harmless Representations to Elicit Unsafe LLM Outputs

Wenpeng Xing, Bohan Yang, Zaifeng Yang, Changting Lin, Meng Han^#

Student First Author Under review

Pairs harmful queries with structurally similar benign counterparts and uses refusal-loss gradients to choose layers and token positions for hidden-state interpolation. Across five open-weight models and four safety benchmarks, it reaches 94.13% average attack success; tailored adversarial training lowers re-optimized attack success to 12.37%.

[Paper]