Papers by Ziwei Zhu
VIGNETTE: Socially Grounded Bias Evaluation for Vision-Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing studies on VLM bias focus on portrait-style images and gender-occupation associations . existing studies ignore broader and more complex social stereotypes and their implied harm . |
| Approach: | They propose a large-scale VQA benchmark for evaluating bias in vision-language models . they use a question-answering framework that spans factuality, perception, stereotyping, and decision making . |
| Outcome: | The proposed framework examines bias in vision-language models using 30M+ images . findings reveal subtle, multifaceted, and surprising stereotypical patterns . |
Toward Inclusive Language Models: Sparsity-Driven Calibration for Systematic and Interpretable Mitigation of Social Biases in LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a new method to mitigate stereotypical bias in large language models is needed . inherent biases from training on vast Internet datasets can amplify harmful stereotypes . |
| Approach: | They propose a method to identify stereotypical bias in decoder-only transformer models . they apply a localization mechanism that correlates internal activations with a new Context Influence score . |
| Outcome: | The proposed method reduces stereotypical biases on BBQ, StereoSet, and CrowS-Pairs while improving reasoning performance on MMLU by 10%. |
Navigating the Shortcut Maze: A Comprehensive Analysis of Shortcut Learning in Text Classification by Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Language models (LMs) often rely on spurious correlations rather than causally relevant features to improve accuracy and generalizability. |
| Approach: | They propose a benchmark that categorizes shortcuts into occurrence, style, and concept . they aim to explore the nuanced ways shortcuts influence the performance of LMs . |
| Outcome: | The proposed benchmark categorizes shortcuts into occurrence, style, and concept . it systematically investigates models’ resilience and susceptibilities to sophisticated shortcuts . |
Talent or Luck? Evaluating Attribution Bias in Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on social biases in large language models focus on surface-level associations or isolated stereotypes. |
| Approach: | They propose a cognitively grounded bias evaluation framework to capture demographic biases across three contexts: single-actor, actor–actor and actor–observer. |
| Outcome: | The proposed framework captures comparative and perspective-driven biases overlooked in previous work. |
ProtoLens: Advancing Prototype Learning for Fine-Grained Interpretability in Text Classification (2025.acl-long)
Copied to clipboard
| Challenge: | ProtoLens provides fine-grained, sub-sentence level interpretability for text classification. |
| Approach: | They propose a prototype-based model that provides fine-grained, sub-sentence level interpretability for text classification. |
| Outcome: | Extensive experiments show that ProtoLens outperforms both prototype-based and non-interpretable baselines on multiple text classification benchmarks. |
HiDe-LLaVA: Hierarchical Decoupling for Continual Instruction Tuning of Multimodal Large Language Model (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods to improve instructionfollowing performance of MLLMs often trade off memory efficiency for performance gains, compromising overall efficiency. |
| Approach: | They propose a task-specific expansion and task-general fusion framework based on variations in Centered Kernel Alignment (CKA) similarity across different model layers when trained on diverse datasets. |
| Outcome: | The proposed framework improves performance compared to existing benchmarks. |
Infusing Disease Knowledge into BERT for Health Question Answering, Medical Inference and Disease Name Recognition (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to augment pre-trained language models with disease knowledge are lacking. |
| Approach: | They propose a method to augment BERT-like pre-trained language models with disease knowledge. |
| Outcome: | The proposed method improves on a suite of BERT models over three tasks. |
Zero-Shot Learners for Natural Language Understanding via a Unified Multiple Choice Perspective (2022.emnlp-main)
Copied to clipboard
Ping Yang, Junjie Wang, Ruyi Gan, Xinyu Zhu, Lin Zhang, Ziwei Wu, Xinyu Gao, Jiaxing Zhang, Tetsuya Sakai
| Challenge: | Existing approaches to zero-shot learning are format-agnostic and can address new learning tasks without additional training. |
| Approach: | They propose a new paradigm for zero-shot learning that is format agnostic and compatible with any format and applicable to a list of language tasks. |
| Outcome: | The proposed model shows state-of-the-art performance on several benchmarks and produces satisfactory results on tasks such as text classification and commonsense reasoning. |
BiasDora: Exploring Hidden Biased Associations in Vision-Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies on social biases focus on a limited set of documented associations, such as gender-profession or race-crime. |
| Approach: | They propose to examine hidden, implicit bias associations across 9 bias dimensions by probing VLMs to uncover hidden, unexamined associations. |
| Outcome: | The proposed methods reveal that biases vary in negativity, toxicity, and extremity. |
Global Gallery: The Fine Art of Painting Culture Portraits through Multilingual Instruction Tuning (2024.naacl-long)
Copied to clipboard
| Challenge: | This study examines the ability of Large Language Models to encapsulate cultural nuances across diverse linguistic landscapes. |
| Approach: | They examine the efficacy of language-specific instruction tuning and the impact of pretraining on dominant language data in Large Language Models. |
| Outcome: | The findings highlight a nuanced landscape, with inconsistencies and biases, particularly in non-Western cultures. |
PromptAttack: Probing Dialogue State Trackers with Adversarial Prompts (2023.findings-acl)
Copied to clipboard
| Challenge: | Toward building more robust and reliable conversational systems, we introduce a prompt-based learning approach to automatically generate effective adversarial examples to probe DST models. |
| Approach: | They propose a prompt-based learning approach to automatically generate effective adversarial examples to probe DST models. |
| Outcome: | The proposed framework leads to the greatest reduction in accuracy and the best attack success rate while maintaining good fluency and a low perturbation ratio. |
Inject to Heal: Alleviating hallucination in LVLMs via Context Embedding Injection (2026.findings-acl)
Copied to clipboard
| Challenge: | a large vision-language model can generate hallucinations inconsistent with visual input . a lightweight method that embeds the last input token as a grounding signal reduces the likelihood of hallucinosity. |
| Approach: | They propose a training-free mitigation strategy that harnesses the hidden state of the last input token as a grounding signal to maintain visual fidelity throughout decoding and curb hallucinations. |
| Outcome: | The proposed method outperforms state-of-the-art methods on CHAIR, AMBER, and MMHal benchmarks. |
Multimodal Emotion Recognition in Conversations: A Survey of Methods, Trends, Challenges and Prospects (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Multimodal Emotion Recognition in Conversations (MERC) is a new way to enhance human-computer interaction. |
| Approach: | This survey offers a systematic overview of Multimodal Emotion Recognition in Conversations . it examines motivations, core tasks, representative methods, and evaluation strategies . |
| Outcome: | The survey examines the effectiveness of MERC and its evaluation strategies. |
Unsupervised Candidate Answer Extraction through Differentiable Masker-Reconstructor Model (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for candidate answer extraction are reliant on linguistic rules or annotated data and face partial annotation issue and challenges in generalization. |
| Approach: | They propose an unsupervised approach that leverages the inherent structure of context passages through a Differentiable Masker-Reconstructor (DMR) Model with the enforcement of self-consistency for picking up salient information tokens. |
| Outcome: | The proposed model outperforms supervised and unsupervised methods in two datasets with exhaustively-annotated answers and shows that it is comparable to supervised methods. |
Spurious Correlations and Beyond: Understanding and Mitigating Shortcut Learning in SDOH Extraction with Large Language Models (2025.acl-short)
Copied to clipboard
| Challenge: | Large language models (LLMs) rely on superficial cues leading to spurious predictions . recent work has highlighted how LLMs exploit spurious patterns rather than learning causal, generalizable features. |
| Approach: | They use a social history annotation corpus dataset to examine drug status extraction . they evaluate prompt engineering and chain-of-thought reasoning to reduce false positives . |
| Outcome: | The proposed model can predict drug use when alcohol or smoking is not present, while uncovering gender disparities in model performance. |
What’s Not Said Still Hurts: A Description-Based Evaluation Framework for Measuring Social Bias in LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing benchmarks evaluate bias by term-based mode, but they fail to capture hidden biases in nuanced settings. |
| Approach: | They propose a dataset to assess bias at the semantic level that bias concepts are hidden within naturalistic, subtly framed contexts in real-world scenarios. |
| Outcome: | The proposed dataset shows that models reduce bias in response at term level, but reinforce bias in nuanced settings. |
FARSS: Fisher-Optimized Adaptive Low-Rank and Singular-Vector Selection for Knowledge-Preserving Fine-Tuning (2026.findings-acl)
Copied to clipboard
Renxing Chen, Ziwei Xiang, Peisong Wang, Hongjian Fang, Meng Li, Fanhu Zeng, Yanan Zhu, Peipei Yang, Xu-Yao Zhang, Jian Cheng
| Challenge: | Low-rank adaptation methods for large language models have limitations in preserving world knowledge and limiting updates to preserve world knowledge. |
| Approach: | They propose a Fisher-optimized adaptive low Rank and Singular-VectorSelection framework for knowledge-preserving fine-tuning that allows efficient and task-sensitive updates. |
| Outcome: | The proposed framework outperforms existing methods for knowledge-preserving fine-tuning. |
Global Voices, Local Biases: Socio-Cultural Prejudices across Languages (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies on human biases are heavily skewed towards Western and European languages . despite growing interest in language models, there are several shortcomings in the literature . |
| Approach: | They scale the Word Embedding Association Test to 24 languages and add culturally relevant information for each language. |
| Outcome: | The proposed language models can reflect and often amplify the effects of bias across linguistic, cultural, and societal borders. |
Co2PT: Mitigating Bias in Pre-trained Language Models through Counterfactual Contrastive Prompt Tuning (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Pre-trained language models can encode unfair social biases from large pre-training corpora and even amplify biase in downstream applications. |
| Approach: | They propose a *debias-while-prompt tuning* method for mitigating biases via counterfactual contrastive prompt tuning on downstream tasks. |
| Outcome: | The proposed method can mitigate biases on three extrinsic bias benchmarks and adapt to existing debiased language models. |
Fighting Spurious Correlations in Text Classification via a Causal Learning Perspective (2025.naacl-long)
Copied to clipboard
| Challenge: | In text classification tasks, models often rely on spurious correlations for predictions, incorrectly associating irrelevant features with the target labels. |
| Approach: | They propose a Causally Calibrated Robust Classifier which integrates a causal feature selection method based on counterfactual reasoning and an unbiased inverse propensity weighting (IPW) loss function. |
| Outcome: | The proposed method achieves state-of-the-art performance among methods without group labels and can compete with the models that utilize group labels. |