Papers by Xuezhi Wang
Leveraging Dual Process Theory in Language Agent Framework for Real-time Simultaneous Human-AI Collaboration (2025.acl-long)
Copied to clipboard
Shao Zhang, Xihuai Wang, Wenhao Zhang, Chaoran Li, Junru Song, Tingyu Li, Lin Qiu, Xuezhi Cao, Xunliang Cai, Wen Yao, Weinan Zhang, Xinbing Wang, Ying Wen
| Challenge: | Large language models (LLMs) excel in turn-by-turn human-AI collaboration but struggle with simultaneous tasks requiring real-time interaction. |
| Approach: | They propose a language agent framework that integrates *System 1* and *System 2* for efficient real-time simultaneous human-AI collaboration. |
| Outcome: | The proposed framework improves on existing LLM-based agents and human collaborators by integrating Theory of Mind and asynchronous reflection to infer human intentions and perform reasoning-based autonomous decisions. |
AMO-Bench: Large Language Models Still Struggle in High School Math Competitions (2026.findings-acl)
Copied to clipboard
Junlin Liu, Shengnan An, Shuang Zhou, Dan Ma, Yehao Lin, Xinxuan Lv, Xuanlin Wang, Xiaoyu Li, Ziwen Wang, Xuezhi Cao, Xunliang Cai
| Challenge: | Existing benchmarks for mathematical reasoning are becoming less effective due to performance saturation. |
| Approach: | They propose to use a mathematical reasoning benchmark with Olympiad difficulty to evaluate top-tier LLMs. |
| Outcome: | The proposed benchmarks are cross-validated by experts to meet IMO difficulty standards and entirely original problems to prevent performance leakages from data memorization. |
If an LLM Were a Character, Would It Know Its Own Story? Evaluating Lifelong Learning in LLMs (2026.acl-long)
Copied to clipboard
Siqi Fan, Xiusheng Huang, Yiqun Yao, Xuezhi Fang, Kang Liu, Peng Han, Shuo Shang, Aixin Sun, Yequan Wang
| Challenge: | Existing benchmarks for large language models (LLMs) fail to capture these dynamics, focusing on static, open-ended evaluations. |
| Approach: | They propose a benchmark to assess lifelong learning in large language models . they use two episodic datasets rich in narrative structure and character interactions . |
| Outcome: | Experiments on LLMs show that non-parametric methods outperform parametric ones in managing stateful learning. |
FreshLLMs: Refreshing Large Language Models with Search Engine Augmentation (2024.findings-acl)
Copied to clipboard
Tu Vu, Mohit Iyyer, Xuezhi Wang, Noah Constant, Jerry Wei, Jason Wei, Chris Tar, Yun-Hsuan Sung, Denny Zhou, Quoc Le, Thang Luong
| Challenge: | Modern large language models often "hallucinate" plausible but factually incorrect information, which reduces their trustworthiness especially in settings where accurate and up-to-date information is critical. |
| Approach: | They develop a human evaluation procedure to measure correctness and hallucination and use it to benchmark both closed and open-source LLMs. |
| Outcome: | The proposed method outperforms both competing search engine-augmented prompting methods and commercial systems on search-augmented QA. |
Continual Sequence Generation with Adaptive Compositional Modules (2022.acl-long)
Copied to clipboard
| Challenge: | Existing work on continual sequence generation relies on reuse of existing parameters to learn new tasks or blindly adds new parameters for each new task. |
| Approach: | They propose to use continual sequence generation to reuse existing parameters for new tasks . current continuous learning models often forget knowledge of dissimilar tasks if data distributions shift . |
| Outcome: | The proposed framework outperforms state-of-the-art models on various sequences of generation tasks. |
Attribution-Based Analysis and Optimization of Modular Agentic Workflows (2026.findings-acl)
Copied to clipboard
Yingxuan Yang, Bo Huang, Siyuan Qi, Chao Feng, Haoyi Hu, Yuxuan Zhu, Jinbo Hu, Haoran Zhao, Ziyi He, Xiao Liu, ZongYu Wang, Muning Wen, Lin Qiu, Xuezhi Cao, Xunliang Cai, Yong Yu, Weinan Zhang
| Challenge: | Large Language Models (LLMs) have driven the rise of agentic workflows . yet, how can we attribute performance gains to individual upgrades and their interactions? |
| Approach: | They propose a game-theoretic framework that models component upgrades as players and evaluates component coalitions to compute Shapley values. |
| Outcome: | The proposed framework provides interaction-aware attribution and recommendation for model allocation under a fixed workflow structure. |
Continual Learning for Text Classification with Information Disentanglement Based Regularization (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing continual learning methods focus on preserving knowledge from previous tasks . Continual learning is a useful tool for learning over time, but it is not always possible to generalize to new tasks. |
| Approach: | They propose a disentanglement-based regularization method for continual learning on text classification that disentangles text hidden spaces into generic representations and regularizes them differently to constrain knowledge required to generalize. |
| Outcome: | The proposed method disentangles text hidden spaces into representations that are generic to all tasks and representations specific to each individual task. |
Position-Aware Depth Decay Decoding (D3): Boosting Large Language Model Inference Efficiency (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent dynamic computation methods show that not all components are required for inference, enabling a training-free pipeline. |
| Approach: | They propose a token-position aware layer skipping framework to save 1.5x times operations efficiently while maintaining performance. |
| Outcome: | The proposed algorithm achieves 1.5x speedup on large language models with no retraining and with comparable performance on the GSM8K and BBH benchmarks. |
ToTTo: A Controlled Table-To-Text Generation Dataset (2020.emnlp-main)
Copied to clipboard
Ankur Parikh, Xuezhi Wang, Sebastian Gehrmann, Manaal Faruqui, Bhuwan Dhingra, Diyi Yang, Dipanjan Das
| Challenge: | Existing methods for data-to-text generation often hallucinate phrases not supported by the Wikipedia table. |
| Approach: | They propose a controlled task where annotators directly revise existing Wikipedia sentences to generate a one-sentence description. |
| Outcome: | The proposed task produces a one-sentence description from a Wikipedia table and highlighted cells. |
Why Not Act on What You Know? Unleashing Safety Potential of LLMs via Self-Aware Guard Enhancement (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated impressive capabilities across various tasks but are vulnerable to meticulously crafted jailbreak attacks. |
| Approach: | They propose a training-free defense strategy to align LLMs’ strong safety discrimination performance with their relatively weaker safety generation ability. |
| Outcome: | The proposed strategy achieves an average 99% success rate against numerous complex and covert jailbreak methods while maintaining helpfulness on general benchmarks. |
Measure and Improve Robustness in NLP Models: A Survey (2022.naacl-main)
Copied to clipboard
| Challenge: | Despite the performance gains, NLP models are still fragile and brittle to out-of-domain data, adversarial attacks, or small perturbation to the input. |
| Approach: | They propose a survey of how to define, measure and improve robustness in NLP by connecting multiple definitions of robustness and identifying failures. |
| Outcome: | The proposed models are robust against unseen or challenging scenarios, but are still fragile and brittle to out-of-domain data and adversarial attacks. |
CAT-Gen: Improving Robustness in NLP Models via Controlled Adversarial Text Generation (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing adversarial text generation approaches can lead to generation lacking diversity or fluency, whereas perturbing in the intermediate representation space can lead a model to generate generations that are not related to the input. |
| Approach: | They propose to generate adversarial texts through controllable attributes that are known to be invariant to task labels. |
| Outcome: | The proposed model generates more diverse and fluent adversarial examples, compared to existing approaches, and is more robust against model re-training and different model architectures. |
Bounding the Capabilities of Large Language Models in Open Text Generation with Prompt Constraints (2023.findings-eacl)
Copied to clipboard
| Challenge: | Existing and potential applications of open-ended text generation are farreaching, spanning domains such as QA, story generation, open-end dialogue, and ChatGPT 1 . |
| Approach: | They propose a prompt-centric approach to analyzing and bounding the abilities of open-ended generative models by a set of structural and stylistic prompts. |
| Outcome: | The proposed method can be generalized to other large models like BLOOM and OPT. |
Can We Improve Model Robustness through Secondary Attribute Counterfactuals? (2021.emnlp-main)
Copied to clipboard
| Challenge: | Recent research has explored how models rely on spurious correlations and how counterfactual data augmentation (CDA) can mitigate such issues. |
| Approach: | They propose a context-aware methodology which takes into account the impact of secondary attributes on the model’s predictions and increases sensitivity for secondary attributes over reweighted counterfactually augmented data. |
| Outcome: | The proposed approach improves sliced accuracy on the original dataset by 7% compared to existing methods and provides guidelines to extend this to other tasks. |
Investigating Ensemble Methods for Model Robustness Improvement of Text Classifiers (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to reduce model's reliance on bias features ignore the learnability of these features. |
| Approach: | They propose to reduce models' reliance on bias features by first training models with fixed low-capacity models which ignore the learnability of the bias features. |
| Outcome: | The proposed models can perform better on out-of-distribution datasets than baseline models with a more sophisticated model design. |
Meeseeks: A Feedback-Driven, Iterative Self-Correction Benchmark evaluating LLMs’ Instruction Following Capability (2026.findings-acl)
Copied to clipboard
Jiaming Wang, Yunke Zhao, Peng Ding, Jun Kuang, Yibin Shen, Zhe Tang, Yilin Jin, ZongYu Wang, Xiaoyu Li, Xuezhi Cao
| Challenge: | Existing models lack the ability to adhere to instructions, resulting in suboptimal performance. |
| Approach: | They propose an automated iterative instruction-following benchmark with integrated feedback mechanism. |
| Outcome: | The proposed benchmark identifies erroneous components in model responses and provides feedback accurately. |
SOP-Maze: Evaluating Large Language Models on Complicated Business Standard Operating Procedures (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are widely deployed as domain-specific agents, but evaluation of their capabilities in such contexts has not been fully explored. |
| Approach: | They propose a benchmark to evaluate LLMs' ability to follow instructions and make decisions in real-world scenarios. |
| Outcome: | The proposed benchmark is constructed from real-world business data and adapted into 23 complex SOP scenarios. |
Identifying and Mitigating Spurious Correlations for Improving Robustness in NLP Models (2022.findings-naacl)
Copied to clipboard
| Challenge: | Existing work identifies task-specific shortcuts via human priors or error analyses, which requires extensive expertise and efforts. |
| Approach: | They propose to automatically identify spurious correlations in NLP models at scale by using existing interpretability methods to extract tokens that significantly affect model’s decision process. |
| Outcome: | The proposed method can identify spurious correlations in NLP models at scale and mitigate these leads to more robust models in multiple applications. |
Large Language Models Can Self-Improve (2023.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have excellent performance in various tasks, but fine-tuning requires extensive supervision. |
| Approach: | They propose to use a pre-trained Large Language Model to generate rationale-augmented answers for unlabeled questions and fine-tune the LLM using those self-generated solutions as target outputs. |
| Outcome: | The proposed approach improves the general reasoning ability of a 540B-parameter LLM without any ground truth label. |
CoreCodeBench: Decoupling Code Intelligence via Fine-Grained Repository-Level Tasks (2026.acl-long)
Copied to clipboard
Lingyue Fu, Hao Guan, Bolun Zhang, Haowei Yuan, Yaoming Zhu, Lin Qiu, ZongYu Wang, Xuezhi Cao, Xunliang Cai, Weiwen Liu, Weinan Zhang, Yong Yu
| Challenge: | Existing large language models for software engineering rely on coarse-grained pass rates obscuring specific cognitive bottlenecks. |
| Approach: | They propose a repository-level benchmark that dissects coding capabilities through atomized tasks. |
| Outcome: | The proposed framework achieves a 78.55% validity yield, surpassing the 31.7% retention rate of SWE-bench-Verified. |
Improving Classifier Robustness through Active Generative Counterfactual Data Augmentation (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for finding meaningful counterfactuals rely on human annotation or implicit label invariance . a small amount of human-annotated counterf actual data can generate a robust dataset with learned labels. |
| Approach: | They propose a framework that generates counterfactuals by actively sampling from regions of uncertainty and automatically labeling them with a learned auxiliary classifier. |
| Outcome: | The proposed framework generates a large number of diverse counterfactuals and labels them with a learned classifier. |
Transferable and Efficient: Unifying Dynamic Multi-Domain Product Categorization (2023.acl-industry)
Copied to clipboard
| Challenge: | e-commerce platforms are encountering increasingly complex product categorization scenarios . multiple business domains correspond to different category taxonomies, with different depths and distinct literal expressions of category names. |
| Approach: | They propose a taxonomy-agnostic framework that calculates semantic relatedness between product titles and category names in the vector space. |
| Outcome: | The proposed framework outperforms strong baselineson three dynamic multi-domain product categorization tasks. |
Table Fact Verification with Structure-Aware Transformer (2020.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained models cannot be used to encode semi-structured data because of their nature. |
| Approach: | They propose a Structure-Aware Transformer which injects table structural information into mask . method could combine symbolic and linguistic reasoning, they propose . |
| Outcome: | The proposed method outperforms baseline on a large scale table verification dataset. |
CTAP for Chinese:A Linguistic Complexity Feature Automatic Calculation Platform (2022.lrec-1)
Copied to clipboard
| Challenge: | Existing tools to analyze linguistic complexity are limited and different because of different research purposes. |
| Approach: | They propose to integrate Chinese component into CTAP to analyze linguistic complexity . they propose to use 196 linguistic complex indexes to calculate linguistic characteristics . |
| Outcome: | The proposed indexes are compared with three linguistic complexity tools for Chinese . the proposed index sets include four levels of 196 linguistic complex indexe . |