Papers by Yichi Zhang
DANLI: Deliberative Agent for Following Natural Language Instructions (2022.emnlp-main)
Copied to clipboard
Yichi Zhang, Jianing Yang, Jiayi Pan, Shane Storks, Nikhil Devraj, Ziqiao Ma, Keunwoo Yu, Yuwei Bao, Joyce Chai
| Challenge: | Recent work on embodied AI agents that can perform tasks by following human language instructions is limited by reactive methods, which are insufficient for long-horizon complex tasks. |
| Approach: | They propose a neuro-symbolic deliberative agent that, while following language instructions, proactively applies reasoning and planning based on its neural and symbolic representations acquired from past experience. |
| Outcome: | The proposed agent achieves greater than 70% improvement over reactive baselines on the challenging TEACh benchmark. |
Noise-powered Multi-modal Knowledge Graph Representation Framework (2025.coling-main)
Copied to clipboard
| Challenge: | Current efforts to integrate MMKG with pretraining are scarce. |
| Approach: | They propose a method that integrates multi-modal entity features into MMKGs using a Transformer-based architecture equipped with modality-level noise masking. |
| Outcome: | The proposed method achieves SOTA performance across ten datasets. |
Hierarchical Task Learning from Language Instructions with Unified Transformers and Self-Monitoring (2021.findings-acl)
Copied to clipboard
| Challenge: | despite recent progress, learning new tasks through language instructions remains a challenging problem. |
| Approach: | They propose a hierarchical task learning approach that decomposes task learning into three sub-problems and a model that addresses each sub-probability in a unified manner. |
| Outcome: | The proposed model achieves the state-of-the-art performance on the AL-FRED benchmark . it decomposes task learning into three sub-problems and addresses them in a unified manner . |
Collaboration of Fusion and Independence: Hypercomplex-driven Robust Multi-Modal Knowledge Graph Completion (2026.acl-long)
Copied to clipboard
| Challenge: | Existing multi-modal knowledge graphs lack modality-specific information and are limited in their ability to capture nuanced semantic interplay between modalities. |
| Approach: | They propose a multi-modal knowledge graph completion method which integrates both paradigms . they use a fine-grained Entity Representation Factorization module and a Robust Relation-aware Modality Fusion module to obtain robust representations for three independent modalities and one fused modality. |
| Outcome: | The proposed method achieves coexistence and collaboration of fused and independent modality representations while maintaining modality-specific information. |
Interpretable and Low-Resource Entity Matching via Decoupling Feature Learning from Decision Making (2021.acl-long)
Copied to clipboard
| Challenge: | Entity Matching (EM) aims at recognizing entity records that denote the same real-world object. |
| Approach: | They propose a novel EM framework that consists of Heterogeneous Information Fusion and Key Attribute Tree Induction to decouple feature representation from matching decision. |
| Outcome: | The proposed framework outperforms SOTA EM models on 6 public datasets and 3 industrial datasets. |
RL-Guider: Leveraging Historical Decisions and Feedback for Drug Editing with Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | RLs can be used to refine drugs by iterative conversations with domain experts . existing methods do not leverage past knowledge, but human experts develop intuition over time through historical experience . |
| Approach: | They propose a reinforcement-learning agent to provide suggestions to large language models . RL-Guider leverages the “world-level” knowledge of LLMs and historical feedback . |
| Outcome: | a new reinforcement-learning agent improves the performance of large language models . the proposed agent leverages the “world-level” knowledge of LLMs and historical feedback . |
Looking Beyond Text: Reducing Language Bias in Large Vision-Language Models via Multimodal Dual-Attention and Soft-Image Guidance (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large vision-language models (LVLMs) have been criticized for their language bias. |
| Approach: | They propose to use a dual-attention mechanism to construct separate attention for visual and text inputs to enhance integration of visual inputs across models. |
| Outcome: | Experiments show that the proposed model debiases LVLMs from their language bias, enhancing visual comprehension and reducing hallucinations without additional resources. |
Paraphrase Augmented Task-Oriented Dialog Generation (2020.acl-main)
Copied to clipboard
| Challenge: | Neural generative models can perform dialog generation tasks with a large data set, but lack of high-quality data and expensive data annotation process limit their application in real world settings. |
| Approach: | They propose to combine paraphrase and response generation models to improve dialog generation performance by annotating dialog states and dialog act labels. |
| Outcome: | The proposed framework outperforms existing methods significantly in dialog generation tasks, especially under low resource settings. |
Know the Known and the Unknown: Reasonable Answer Generation with Knowledge-Informed Citations (2026.acl-long)
Copied to clipboard
Yichi Zhang, Zhuo Chen, Lingbing Guo, Jun Xu, Mengshu Sun, Zhizhen Liu, Lei Liang, Wen Zhang, Huajun Chen
| Challenge: | Existing approaches focus on generating multi-level citations linked to specific references, making it verifiable and trustworthy. |
| Approach: | They propose a new data construction pipeline and a benchmark to improve citation granularity and awareness of unknown information. |
| Outcome: | The proposed model improves on the existing benchmark and data construction pipeline and provides citation granularity and awareness of unknown information. |
Training Turn-by-Turn Verifiers for Dialogue Tutoring Agents: The Curious Case of LLMs as Your Coding Tutors (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have focused on coding tutoring, but their capabilities in guiding users to solve complex tasks remain underexplored. |
| Approach: | They propose a novel agent workflow, Trace-and-Verify, which combines knowledge tracing to estimate a student’s knowledge state and turn-by-turn verification to ensure effective guidance toward task completion. |
| Outcome: | The proposed agent workflow achieves significantly higher success rates than existing tutoring agents. |
A Probabilistic End-To-End Task-Oriented Dialog Model with Latent Belief States towards Semi-Supervised Learning (2020.emnlp-main)
Copied to clipboard
| Challenge: | Structured belief states are crucial for goal tracking and database query in task-oriented dialog systems. |
| Approach: | They propose a probabilistic dialog model where belief states are represented as discrete latent variables and jointly modeled with system responses given user inputs. |
| Outcome: | The proposed model outperforms supervised-only and semi-supervised baselines on three benchmark datasets. |
Exploring the Generalizability of Factual Hallucination Mitigation via Enhancing Precise Knowledge Utilization (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to improve LLMs' ability to align their responses with objective facts suffer from poor generalization and trade-offs in other different capabilities. |
| Approach: | They propose to introduce PKUE (Precise Knowledge Utilization Enhancement) which fine-tunes the model on self-generated responses to precise and simple factual questions through preference optimization. |
| Outcome: | The proposed enhancements improve LLM’s ability to precisely leverage its knowledge and improve FactualBench, a comprehensive and precise factual QA dataset containing 181k Chinese data spanning 21 domains. |
Tiered Reasoning for Intuitive Physics: Toward Verifiable Commonsense Language Understanding (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Large-scale, pre-trained language models (LMs) have achieved human-level performance on a breadth of language understanding tasks. |
| Approach: | They propose a commonsense reasoning dataset with dense annotations that allows multi-tiered evaluation of machines’ reasoning process. |
| Outcome: | The proposed model can achieve high end performance but struggle to support predictions with valid supporting evidence. |
Knowledgeable Preference Alignment for LLMs in Domain-specific Question Answering (2024.findings-acl)
Copied to clipboard
| Challenge: | Domain-specific question answering (QA) requires a comprehensive understanding of a specific domain to answer specialized questions. |
| Approach: | They propose a new alignment objective to align the LLM preference with different human preferences uniformly to optimize LLM performance in real-world, domain-specific QA settings. |
| Outcome: | The proposed pipeline is superior for real-scenario domain-specific question answering with LLMs. |
Dynamic Anticipation and Completion for Multi-Hop Reasoning over Sparse Knowledge Graph (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing reasoning methods for sparse KGs are incomplete and lack of evidential paths to target entities makes multi-hop reasoning difficult. |
| Approach: | They propose a multi-hop reasoning model over sparse KGs to solve this problem . they use latent prediction of embedding-based models to make the model perform more potential path search over sparses . |
| Outcome: | The proposed method outperforms state-of-the-art models on five datasets from Freebase, NELL and Wikidata. |
Structured and Abstractive Reasoning on Multi-modal Relational Knowledge Images (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on understanding and reasoning with abstractive information from the visual modality have not explored the use of STructured and Abstractive Reasoning (STAR) on such data. |
| Approach: | They propose an automatic STAR data engine to synthesize images with MMRK to build multi-modal instructions with reliable chain-of-thought thinking for various STAR tasks. |
| Outcome: | The proposed framework outperforms GPT-4o in STAR and improves performance across 8 open-source MLLMs. |
CriticLean: Critic-Guided Reinforcement Learning for Mathematical Formalization (2026.acl-long)
Copied to clipboard
Zhongyuan Peng, Yifan Yao, Kaijing Ma, Shuyue Guo, Yizhe Li, Yichi Zhang, Chenchen Zhang, Yifan Zhang, Zhouliang Yu, Luming Li, Minghao Liu, Yihang Xia, Jiawei Shen, Yuchen Wu, Yixin Cao, Zhaoxiang Zhang, Wenhao Huang, Jiaheng Liu, Ge Zhang
| Challenge: | Existing approaches to formalizing mathematical statements face limitations in accuracy, especially in the context of complex, highlevel problems that involve sophisticated mathematical reasoning. |
| Approach: | They propose a CriticLean framework that elevates the role of the critic from a passive validator to an active learning component and introduce a benchmark to measure models’ ability to distinguish semantically correct from incorrect formalizations. |
| Outcome: | The proposed framework outperforms open- and closed-source benchmarks and shows that it significantly outperformed existing models. |
Have We Designed Generalizable Structural Knowledge Promptings? Systematic Evaluation and Rethinking (2025.acl-long)
Copied to clipboard
Yichi Zhang, Zhuo Chen, Lingbing Guo, Yajing Xu, Shaokai Chen, Mengshu Sun, Binbin Hu, Zhiqiang Zhang, Lei Liang, Wen Zhang, Huajun Chen
| Challenge: | Existing methods to integrate external knowledge into LLMs focus on specific problems, lacking a comprehensive exploration of the generalization and capability boundaries of SKP. |
| Approach: | They propose a new paradigm for structural knowledge prompting to integrate external structural knowledge into LLMs by incorporating structural representations. |
| Outcome: | The proposed benchmark SUBARU enables the evaluation of the generalization capabilities of SKP from four perspectives. |
Can Foundation Models Watch, Talk and Guide You Step by Step to Make a Cake? (2023.findings-emnlp)
Copied to clipboard
Yuwei Bao, Keunwoo Yu, Yichi Zhang, Shane Storks, Itamar Bar-Yossef, Alex de la Iglesia, Megan Su, Xiao Zheng, Joyce Chai
| Challenge: | despite advances in AI, it remains a challenge to develop interactive task guidance systems that can offer situated, personalized guidance and assist humans in various tasks. |
| Approach: | They propose to use a multimodal benchmark dataset to study whether interactive task guidance systems can be quickly adapted to perceptually enabled tasks. |
| Outcome: | The proposed models demonstrate fair performances in some cases with no training . the results will provide a stepping stone for future work on situated task guidance . |
Proactive Assistant Dialogue Generation from Streaming Egocentric Videos (2025.emnlp-main)
Copied to clipboard
Yichi Zhang, Xin Luna Dong, Zhaojiang Lin, Andrea Madotto, Anuj Kumar, Babak Damavandi, Joyce Chai, Seungwhan Moon
| Challenge: | Recent advances in conversational AI have been substantial, but developing real-time tasks guidance systems remains a challenge. |
| Approach: | They propose a data curation pipeline that synthesizes dialogues from annotated egocentric videos and a suite of automatic evaluation metrics that validated through extensive human studies. |
| Outcome: | The proposed framework synthesizes dialogues from annotated egocentric videos and validates them through extensive human studies. |
Is Multi-Hop Reasoning Really Explainable? Towards Benchmarking Reasoning Interpretability (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing models for multi-hop reasoning are not able to evaluate their interpretability . a recent study found that many paths are unreasonable . |
| Approach: | They propose a framework to evaluate the interpretability of multi-hop reasoning models . they annotate all possible rules and establish a benchmark . |
| Outcome: | The proposed framework outperforms existing models in terms of performance and interpretability. |
SafetyALFRED: Evaluating Safety-Conscious Planning of Vision Language Models (2026.findings-acl)
Copied to clipboard
Josue Torres-Fonseca, Naihao Deng, Yinpei Dai, Shane Storks, Yichi Zhang, Rada Mihalcea, Casey Kennington, Joyce Chai
| Challenge: | Existing safety evaluations focus on hazard recognition through disembodied question answering (QA) settings, but lack a critical gap in evaluating an agent. |
| Approach: | They evaluate multimodal large language models with six categories of kitchen hazards . they propose a safety-based approach that prioritizes multi-step corrective actions . |
| Outcome: | The proposed model can recognize hazards in QA settings, but average mitigation success rates are low . the proposed model is based on the embodied agent benchmark ALFRED . |
PCA-Bench: Evaluating Multimodal Large Language Models in Perception-Cognition-Action Chain (2024.findings-acl)
Copied to clipboard
Liang Chen, Yichi Zhang, Shuhuai Ren, Haozhe Zhao, Zefan Cai, Yuchi Wang, Peiyi Wang, Xiangdi Meng, Tianyu Liu, Baobao Chang
| Challenge: | a new multimodal decision-making benchmark evaluates the integrated capabilities of multimodal large language models. |
| Approach: | They propose a multimodal decision-making benchmark for evaluating MLLMs . they propose an automatic evaluation protocol to assess 10 prevalent ML models . |
| Outcome: | The proposed benchmark improves performance of multimodal large language models in three scenarios . the model is required to integrate multiple capabilities to make accurate decisions . |
Unleashing the Power of Imbalanced Modality Information for Multi-modal Knowledge Graph Completion (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing MMKGC methods overlook the imbalance problem of modality information among entities, resulting in inadequate modal fusion and inefficient utilization of the raw modality. |
| Approach: | They propose to integrate structural, visual, and textual information of entities into the discriminant models to predict the missing triples. |
| Outcome: | The proposed model outperforms 19 recent methods and achieves state-of-the-art results on three public MMKGC benchmarks. |
Alternating Recurrent Dialog Model with Large-scale Pre-trained Language Models (2021.eacl-main)
Copied to clipboard
| Challenge: | Existing dialog system models require extensive human annotations and are difficult to generalize to different tasks. |
| Approach: | They propose a framework that uses pre-trained language to model each speaker separately . it can be generalized to more challenging, non-collaborative tasks such as persuasion . |
| Outcome: | The proposed framework outperforms or is on par with state-of-the-art methods on two popular datasets: CamRest676 and MultiWOZ. |
Breaking the Ceiling: Exploring the Potential of Jailbreak Attacks through Expanding Strategy Space (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to exploit black-box jailbreaks fail to capture key attack patterns . a novel framework decomposes jailbreak strategies into essential components . |
| Approach: | They propose a framework that decomposes jailbreak strategies into essential components and develops genetic-based optimization with intention evaluation mechanisms. |
| Outcome: | The proposed framework achieves 90% success rate on Claude-3.5, where prior methods completely fail . it also surpasses specialized safeguard models in evaluation accuracy . |
MMEvalPro: Calibrating Multimodal Benchmarks Towards Trustworthy and Efficient Evaluation (2025.naacl-long)
Copied to clipboard
Jinsheng Huang, Liang Chen, Taian Guo, Fu Zeng, Yusheng Zhao, Bohan Wu, Ye Yuan, Haozhe Zhao, Zhihui Guo, Yichi Zhang, Jingyang Yuan, Wei Ju, Luchen Liu, Tianyu Liu, Baobao Chang, Ming Zhang
| Challenge: | Large Multimodal Models (LMMs) exhibit impressive cross-modal understanding and reasoning abilities, but many benchmarks suffer from systematic biases. |
| Approach: | They propose a benchmark to avoid Type-I errors by creating one perception question and one knowledge anchor question through a meticulous annotation process. |
| Outcome: | The proposed benchmark avoids Type-I errors while maintaining reliability of MCQ evaluations. |
Grounding Visual Illusions in Language: Do Vision-Language Models Perceive Illusions Like Humans? (2023.emnlp-main)
Copied to clipboard
| Challenge: | Visual illusions are a phenomenon that is often seen in human perception but are not always faithful to the physical world. |
| Approach: | They build a dataset containing five types of visual illusions and formulate four tasks to examine visual illusion in state-of-the-art VLMs. |
| Outcome: | The proposed dataset reveals that larger models are closer to human perception and more susceptible to visual illusions. |
CogAtom: From Cognitive Atoms to Olympiad-level Mathematical Reasoning in Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for generating high-quality, multi-step reasoning are limited . we present a new framework for synthesising rigorous, cognitively diverse problems . |
| Approach: | They propose a cognitive atom-based framework for synthesizing mathematically rigorous problems. |
| Outcome: | The proposed framework outperforms existing methods in accuracy, reasoning depth and diversity while exceeding the difficulty of AIME. |
ASTRA: An Automated Framework for Strategy Discovery, Retrieval, and Evolution for Jailbreaking LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods lack the capability for continuous learning and self-evolution from interactions, limiting the diversity and adaptability of attack strategies. |
| Approach: | They propose an automated framework capable of discovering, retrieving, and evolving attack strategies. |
| Outcome: | The proposed framework outperforms existing baselines in a black-box setting. |