Papers by Cheng Bi
Decoding by Contrasting Knowledge: Enhancing Large Language Model Confidence on Edited Facts (2025.acl-long)
Copied to clipboard
| Challenge: | In-context knowledge editing (ICE) is currently the most effective method for knowledge editing, but it is constrained by the black-box modeling of LLMs and lacks interpretability. |
| Approach: | They propose a method to decode new knowledge by comparing logits with unedited knowledge to improve the accuracy of LLMs. |
| Outcome: | The proposed method improves the performance of LLaMA3-8B-instruct on MQuAKE by up to 219%. |
Gated Differentiable Working Memory for Long-Context Language Modeling (2026.acl-long)
Copied to clipboard
Lingrui Mei, Shenghua Liu, Yiwei Wang, Yuyao Ge, Baolong Bi, Jiayu Yao, Jun Wan, Ziling Yin, Jiafeng Guo, Xueqi Cheng
| Challenge: | Long contexts break transformers, attention scores dilute, model cannot adapt to novel patterns at inference time. |
| Approach: | They propose a framework that gates the memory consolidation process by estimating Contextual Utility . they propose GDWM to maintain a form of working memory with constant contexts . |
| Outcome: | The proposed framework achieves comparable or superior performance on sparse-information tasks with 4 fewer gradient steps compared to uniform baselines. |
Towards Fully Exploiting LLM Internal States to Enhance Knowledge Boundary Perception (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) exhibit impressive performance across diverse tasks but struggle to accurately gauge their knowledge boundaries. |
| Approach: | They propose Consistency-based Confidence Calibration (C3) which assesses confidence consistency through question reformulation to improve LLMs’ ability to recognize their knowledge gaps. |
| Outcome: | The proposed method improves the unknown perception rate by 5.6% on NQ and 4.9% on HotpotQA. |
LPNL: Scalable Link Prediction with Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on graph learning with large language models have focused on the link prediction task on large graphs. |
| Approach: | They propose a framework for scalable link prediction on large-scale heterogeneous graphs based on large language models. |
| Outcome: | The proposed framework outperforms baselines in link prediction tasks on large graphs. |
Knowledge-enriched, Type-constrained and Grammar-guided Question Generation over Knowledge Bases (2020.coling-main)
Copied to clipboard
| Challenge: | Existing methods for question generation over knowledge bases have low diversity and poor fluency due to the limited information contained in the subgraphs and semantic drift due to decoder’s oblivion of the semantics of the answer entity. |
| Approach: | They propose a knowledge-enriched, type-constrained and grammar-guided KBQG model that generates natural-language questions over a set of triples in the KB. |
| Outcome: | The proposed model outperforms existing methods on two widely-used benchmark datasets. |
ALiiCE: Evaluating Positional Fine-grained Citation Generation (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing research on citation generation is limited to sentence-level statements . positional fine-grained citations can appear anywhere within sentences . |
| Approach: | They propose a framework that allows LLMs to generate citations from sentences . they use dependency tree-based methods to parse sentence-level claims into atomic claims . |
| Outcome: | The proposed framework evaluates citation quality using three metrics including positional fine-grained citation recall, precision, and coefficient of variation of citation positions. |
Tailoring Table Retrieval from a Field-aware Hybrid Matching Perspective (2025.emnlp-main)
Copied to clipboard
| Challenge: | Empirical results show that a hybrid retrieval approach to table retrieval outperforms state-of-the-art benchmarks. |
| Approach: | They propose a table-tailored HYbrid matching rEtriever which addresses table matching needs from a field-aware hybrid perspective. |
| Outcome: | Empirical results show that the proposed rEtriever outperforms state-of-the-art retrieval methods. |
Evaluating Implicit Bias in Large Language Models by Attacking From a Psychometric Perspective (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that large language models (LLMs) can elicit implicit biases that hurt certain demographics without explicit harmful words. |
| Approach: | They propose three attack approaches to elicit agreements to biased viewpoints from LLMs from a psychometric perspective and built two benchmarks to compare them. |
| Outcome: | The proposed methods elicit agreements to biased viewpoints more effectively than baselines. |
LINKAGE: Listwise Ranking among Varied-Quality References for Non-Factoid QA Evaluation via LLMs (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Non-factoid (NF) question answering is challenging to evaluate due to diverse potential answers and no objective criterion. |
| Approach: | They propose a listwise NFQA evaluation approach that uses Large Language Models to rank candidate answers in a descending list of reference answers sorted by descending quality. |
| Outcome: | The proposed method has higher correlations with human annotations than standard methods. |
Training a Utility-based Retriever Through Shared Context Attribution for Retrieval-Augmented Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Utility-based retrieval has emerged as a promising topic for downstream tasks . however, capturing passage utility accurately remains unexplored due to insufficient understanding . |
| Approach: | They propose a framework for training utility-based retrievers in Retrieval-Augmented Language Models . it incorporates multi-task generalization and inter-passage interaction to improve performance . |
| Outcome: | The proposed framework improves performance on ten datasets across different tasks. |
a1: Steep Test-time Scaling Law via Environment Augmented Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have made remarkable advances in reasoning, yet continue to struggle with hallucinations, logical errors, and inability to self-correct during complex multi-step tasks. |
| Approach: | They propose a framework that enhances LLM reasoning through real-time environmental feedback validating each reasoning step, dynamic branch exploration for investigating alternative solution paths when faced with errors, and experience-based learning from successful reasoning trajectories. |
| Outcome: | The proposed model outperforms comparable models by 24.4 percentage points across benchmarks while outperforming comparable models. |
SLANG: New Concept Comprehension of Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Dynamic nature of language limits the adaptability of Large Language Models (LLMs) Traditionally, LLMs are trained on static data, which limits their adaptability . |
| Approach: | They propose a benchmark to integrate novel data and assess LLMs’ ability to comprehend emerging concepts, alongside a causal inference-based approach to enhance LLM comprehension of new phrases and their colloquial context. |
| Outcome: | The proposed model outperforms baseline models in terms of precision and relevance in the comprehension of Internet slang and memes. |
Breaking the Generator Barrier: Disentangled Representation for Generalizable AI-Text Detection (2026.acl-long)
Copied to clipboard
| Challenge: | a rapid proliferation of large language models (LLMs) generate text that increasingly resembles human writing . this makes it difficult to capture subtle cues that distinguish AI-generated content from human-written content . |
| Approach: | They propose a framework that disentangles AI-detection semantics from generator-aware artifacts by latent encoding and perturbation-based regularization. |
| Outcome: | The proposed framework disentangles AI-detection semantics from generator-aware artifacts on 20 representative LLMs across 7 categories. |
An Iterative Utility Judgment Framework Inspired by Philosophical Relevance via LLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Relevance emphasizes the aboutness of a result to a query, while utility refers to the result’s usefulness or value to an information seeker. |
| Approach: | They propose an Iterative utiliTy judgmEnt fraMework to promote each step in Retrieval-Augmented Generation (RAG) they propose to use relevance ranking, utility judgments, and answer generation to prioritize high-utility results over low-utilitity results. |
| Outcome: | The proposed framework improves relevance, ranking, and answer generation on retrieval (TREC DL, WebAP), utility judgment task (GTI-NQ), and factoid question-answering (NQ) datasets. |
Can Graph Descriptive Order Affect Solving Graph Problems with LLMs? (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have achieved significant success in reasoning tasks, including mathematical reasoning and logical deduction. |
| Approach: | They conduct the first comprehensive analysis of how the order of graph descriptions impacts LLM performance. |
| Outcome: | The results show that graph descriptions significantly improve LLMs’ comprehension of graph structures, and the robustness of LLM models to graph description order varies across different tasks. |
Safety in Large Reasoning Models: A Survey (2025.findings-emnlp)
Copied to clipboard
Cheng Wang, Yue Liu, Baolong Bi, Duzhen Zhang, Zhong-Zhi Li, Yingwei Ma, Yufei He, Shengju Yu, Xinfeng Li, Junfeng Fang, Jiaheng Zhang, Bryan Hooi
| Challenge: | Large Reasoning Models (LRMs) have a high level of advanced reasoning capabilities, but they are vulnerable and vulnerable. |
| Approach: | This paper presents the first comprehensive survey of Large Reasoning Models . it explores the new safety risks, attacks, and defense strategies specific to LRMs based on reasoning . |
| Outcome: | The proposed study examines the safety and security risks of large reasoning models. |
“Not Aligned” is Not “Malicious”: Being Careful about Hallucinations of Large Language Models’ Jailbreak (2025.coling-main)
Copied to clipboard
| Challenge: | “Jailbreak” is a major safety concern of Large Language Models (LLMs). |
| Approach: | They propose a benchmarking framework to evaluate "jailbreak" outputs . they propose specialized validation framework to ensure outputs are useful malicious instructions . |
| Outcome: | The proposed framework enhances existing benchmarks to ensure outputs are useful . it also aims to evaluate the true potential of jailbroken outputs to cause harm to human society. |
Simple or Complex? Complexity-controllable Question Generation with Soft Templates and Deep Mixture of Experts Model (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing work on complex questions does not consider controlling complexity of generated questions. |
| Approach: | They propose an end-to-end neural complexity-controllable question generation model that incorporates a mixture of experts as the selector of soft templates to capture question similarity while avoiding the expensive construction of actual templates. |
| Outcome: | The proposed model is superior to state-of-the-art methods in both automatic and manual evaluations on two benchmark QA datasets. |
SAM3-I: Segment Anything with Instructions (2026.acl-long)
Copied to clipboard
Jingjing Li, Yue Feng, Yuchen Guo, Jincai Huang, Wei Ji, Qi Bi, Yongri Piao, Miao Zhang, Xiaoqi Zhao, Qiang Chen, Shihao Zou, Huchuan Lu, Li Cheng
| Challenge: | Existing methods for concept-level grounding and instruction-level reasoning use coarse representations and iterative mask filtering. |
| Approach: | They propose an instruction-following extension of the Segment Anything Model 3 family that unifies concept-level grounding and instruction-level reasoning within a single segmentation framework. |
| Outcome: | Experiments show that SAM3-I achieves appealing performance across referring and reasoning-based segmentation while maintaining its strong concept recall ability. |
Instruction Pre-Training: Language Models are Supervised Multitask Learners (2024.emnlp-main)
Copied to clipboard
| Challenge: | Unsupervised multitask pre-training has been the key to the success of language models (LMs) however, scaling it in the post-training stage trends towards better generalization. |
| Approach: | They propose a framework that augments massive raw corpora with instruction-response pairs to pre-train LMs. |
| Outcome: | The proposed framework augments massive raw corpora with instruction-response pairs to pre-train LMs. |
ODL-TempLLM: Ontology-Guided and Description Logic-Reasoned Temporal Reasoning with LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Temporal reasoning is crucial for large language models to understand event concurrency and complex temporal interactions in natural language. |
| Approach: | They propose an ontology-guided and description logic–constrained temporal reasoning paradigm that shifts focus from internal inference to the explicit modeling of temporal structure. |
| Outcome: | The proposed method outperforms state-of-the-art methods by 2.07–31.83 F1 points and 1.00–30.73 EM points, exhibiting strong generalization and highlighting the potential of explicit temporal reasoning. |
Adaptive Token Biaser: Knowledge Editing via Biasing Key Entities (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to update parametric knowledge of large language models (LLMs) are outdated and incontext editing (KE) is not effective due to the substantial cost associated with retraining. |
| Approach: | They propose a new decoding technique that enhances in-context editing (ICE) they propose to use parametric knowledge to update the models' knowledge . |
| Outcome: | The proposed technique improves ICE performance while incurring only half the latency. |
Who is in the Spotlight: The Hidden Bias Undermining Multimodal Retrieval-Augmented Generation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing RAG models are sensitive to the order in which evidence is presented, resulting in unstable performance and biased reasoning. |
| Approach: | They propose to quantify position bias in multimodal RAG systems by using position sensitivity index . they also develop a visualization framework to trace attention allocation patterns across decoder layers . |
| Outcome: | The proposed framework shows that multimodal interactions intensify position bias compared to unimodal settings and that this bias increases logarithmically with retrieval range. |
When Do LLMs Need Retrieval Augmentation? Mitigating LLMs’ Overconfidence Helps Retrieval Augmentation (2024.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have difficulty knowing they do not possess certain knowledge and tend to provide specious answers in such cases. |
| Approach: | They propose to use Retrieval Augmentation to enhance LLMs' ability to perceive their knowledge boundaries to reduce overconfidence. |
| Outcome: | The proposed methods reduce overconfidence and improve accuracy in large language models with fewer retrieval calls. |
UPRISE: Universal Prompt Retrieval for Improving Zero-Shot Evaluation (2023.emnlp-main)
Copied to clipboard
Daixuan Cheng, Shaohan Huang, Junyu Bi, Yuefeng Zhan, Jianfeng Liu, Yujing Wang, Hao Sun, Furu Wei, Weiwei Deng, Qi Zhang
| Challenge: | Large Language Models (LLMs) have impressive capabilities but need for task-specific prompt engineering can hinder their generalization. |
| Approach: | They propose a lightweight and versatile retriever that automatically retrieves prompts for a given zero-shot task input. |
| Outcome: | The proposed model is universally applicable across tasks and models . it mitigates hallucination problem in chatGPT, and it improves even the strongest LLMs. |
HiddenGuard: Fine-Grained Safe Generation with Specialized Representation Router (2026.acl-long)
Copied to clipboard
| Challenge: | Current alignment approaches rely on refusal alignment to avoid harmful content . large language models are often overly cautious or overlook subtle harmful content. |
| Approach: | They propose a framework for fine-grained safe generation in Large Language Models that enables real-time, token-level harmfulness detection and redaction without loss in capability. |
| Outcome: | The proposed framework achieves over 90% in F1 score for detecting and redacting harmful content while preserving overall utility and informativeness of the model’s responses. |
Logical Structure as Knowledge: Enhancing LLM Reasoning via Structured Logical Knowledge Density Estimation (2026.findings-acl)
Copied to clipboard
Zhen Bi, Zhenlin Hu, Xueshu Chen, Mingyang Chen, Cheng Deng, Yida Xue, Zhen Wang, Qing Shen, Ningyu Zhang, Jungang Lou
| Challenge: | Existing data-centric paradigms equate quality with factuality or diversity and ignore the internal logical complexity of training samples. |
| Approach: | They propose a density-aware re-cognizing optimization strategy that prioritizes high-density logical samples to align training with the model's reasoning boundary. |
| Outcome: | The proposed metric outperforms existing methods and improves reasoning performance without increasing total data volume. |
MORE: Multi-mOdal REtrieval Augmented Generative Commonsense Reasoning (2024.findings-acl)
Copied to clipboard
| Challenge: | Language Models (LLMs) have gained increasing prominence in artificial intelligence, especially Large Language Model (LLm) due to the well-recognized reporting bias, the recording of commonsense information is significantly less than its existence in reality. |
| Approach: | They propose a Multi-mOdal REtrieval framework to leverage both text and images to enhance commonsense ability of language models. |
| Outcome: | The proposed framework can leverage both text and images to enhance commonsense ability of language models. |
Utility-Focused LLM Annotation for Retrieval and Retrieval-Augmented Generation (2025.emnlp-main)
Copied to clipboard
Hengran Zhang, Minghao Tang, Keping Bi, Jiafeng Guo, Shihao Liu, Daiting Shi, Dawei Yin, Xueqi Cheng
| Challenge: | Existing studies on large language models for document utility annotations have shown that they improve retrieval performance and RAG outcomes compared to models trained on human annotations. |
| Approach: | They propose a model that maximizes their summed marginal likelihood to annotate document utility on multiple positive samples per query. |
| Outcome: | The proposed model maximizes the marginal likelihood of multiple positive samples per query. |