Papers by Zhihua Wen
AGD: Adversarial Game Defense Against Jailbreak Attacks in Large Language Models (2025.acl-long)
Copied to clipboard
Shilong Pan, Zhiliang Tian, Zhen Huang, Wanlong Yu, Zhihua Wen, Xinwang Liu, Kai Lu, Minlie Huang, Dongsheng Li
| Challenge: | Existing defenses, including post-training alignment and prompt engineering, struggle with adaptability to out-of-distribution (OOD) attacks. |
| Approach: | They propose an adversarial game-based defense method that dynamically adjusts LLMs’ internal representations to achieve a balanced trade-off between helpfulness and harmlessness. |
| Outcome: | The proposed method improves LLMs’ safety over all baselines. |
POMP: Probability-driven Meta-graph Prompter for LLMs in Low-resource Unsupervised Neural Machine Translation (2024.acl-long)
Copied to clipboard
| Challenge: | Low-resource languages (LRLs) face challenges in supervised neural machine translation due to limited parallel data. |
| Approach: | They propose a method that uses a dynamic graph to organize auxiliary languages in prompts to improve LRL translations. |
| Outcome: | The proposed method improves translation accuracy in low-resource languages (LRLs) using auxiliary language pairs and synthetic pseudo-parallel data. |
Mitigating Action-Relation Hallucinations in LVLMs via Relation-aware Visual Enhancement (2026.acl-long)
Copied to clipboard
| Challenge: | Existing research has focused on mitigating object hallucinations but often overlooks more complex relation hallucines, especially action relations involving interactions between objects. |
| Approach: | They propose a framework to locate action-relevant image regions and enhance the LVLM’s attention to those regions by using a Relation-aware Visual Enhancement method. |
| Outcome: | The proposed method achieves superior performance in mitigating action-relation hallucinations with negligible additional inference cost. |
Two-stage Generative Question Answering on Temporal Knowledge Graph Using Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Temporal knowledge graph question answering (TKGQA) is one of the most challenging QA tasks due to the temporal constraints hidden in questions and the answers sought from dynamic structured knowledge. |
| Approach: | They propose a generative temporal knowledge graph question answering framework which guides LLMs to answer temporal questions through two phases: Subgraph Retrieval and Answer Generation. |
| Outcome: | The proposed framework exploits LLM’s intrinsic knowledge to mine temporal constraints and structural links in the questions without extra training, thus narrowing down the subgraph search space in both temporal and structural dimensions. |
GRACE: Gradient-guided Controllable Retrieval for Augmenting Attribute-based Text Generation (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for controlling the generation of pre-trained language models infuse domain bias into the generation process, making it difficult to generate out-of-domain texts. |
| Approach: | They propose a retrieval-augmented generation framework that uses retrieval to generate fluent sentences with high attribute relevance. |
| Outcome: | The proposed method can generate fluent sentences with high attribute relevance while keeping domain bias out of the model. |
GROVE: A Retrieval-augmented Complex Story Generation Framework with A Forest of Evidence (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for generating stories with complex plots rely on detailed prompts, which inadvertently limit the creative potential of the generated stories. |
| Approach: | They propose a retrieval-auGmented stoRy generation framework with a fOrest of eVidEnce to enhance stories’ complexity. |
| Outcome: | The proposed framework enables generating more diverse plotlines from human-written stories. |
Knowledge Injection Exists in MoE? Exploring Expert-Aware Contrast Decoding in MoE for Mitigating LLMs’ Hallucinations (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods to mitigate hallucinations include prompt engineering and model optimization, but lack domain generalization and potential errors in fine-tuning data may exacerbate the hallucism. |
| Approach: | They propose an expert-aware adaptive contrast decoding that uses expert differences in MoE’s higher layers to mitigate hallucinations on QA tasks. |
| Outcome: | The proposed method outperforms baseline models on four datasets Large language models (LLMs) show strong performance but suffer from hallucinations, limiting their application. |