Papers by Xiangyu Xu

27 papers
MSEarth: A Multimodal Benchmark for Earth Science Phenomenon Discovery with MLLMs (2026.acl-long)

Copied to clipboard

Challenge: Existing datasets often rely on synthetic data or figure-caption pairs, failing to capture the depth and complexity of geoscientific reasoning.
Approach: They propose a multimodal scientific dataset and benchmark curated from open-access publications.
Outcome: MSEarth features over 289K figures with captions enriched by contextual discussions and reasoning from original papers.
Beyond Demographics: Enhancing Cultural Value Survey Simulation with Multi-Stage Personality-Driven Cognitive Reasoning (2025.emnlp-main)

Copied to clipboard

Challenge: Introducing **MARK**, a framework for cultural value survey simulation . based on type dynamics theory, it improves accuracy and interpretation of models .
Approach: They propose a framework that integrates psychological theory into cultural value survey simulations.
Outcome: The proposed framework outperforms baseline models on the World Values Survey by 10% accuracy and reduces divergence between model predictions and human preferences.
AdaTooler-V: Adaptive Tool-Use for Images and Videos (2026.findings-acl)

Copied to clipboard

Challenge: Existing models exhibit blind tool-use reasoning patterns, which significantly increases inference overhead and degrades model performance.
Approach: They propose an MLLM that performs adaptive tool-use by determining whether a visual problem truly requires tools.
Outcome: The proposed model outperforms existing methods in visual reasoning tasks.
Solve-Detect-Verify: Inference-Time Scaling with Flexible Generative Verifier (2026.acl-long)

Copied to clipboard

Challenge: Recent advances in Large Language Models (LLMs) have enhanced capabilities in complex reasoning through step-by-step trace generation.
Approach: They propose a generative verifier that dynamically allocates compute between rapid fast thinking and deliberative slow thinking.
Outcome: The proposed solution outperforms GenPRM-32B on ProcessBench while requiring 2.3x fewer TFLOPS and 15x less training data.
Sanitizing Large Language Models in Bug Detection with Data-Flow (2024.findings-emnlp)

Copied to clipboard

Challenge: Large language models (LLMs) have potential in code reasoning tasks but the hallucination effect can compromise the reliability of bug reports.
Approach: They propose a new schema of bug detection that enforces LLMs to emit data-flow paths in few-shot chain-of-thought prompting and validates them via the program-property decomposition.
Outcome: The proposed approach achieves 91.03% precision and 74.00% recall upon synthetic benchmarks and boosts precision by 21.99% with the sanitization.
Bridging Relevance and Reasoning: Rationale Distillation in Retrieval-Augmented Generation (2025.findings-acl)

Copied to clipboard

Challenge: Existing approaches to rerank and align documents based on reasoning capabilities of large language models (LLMs) . prior work shows that LLMs have exceptional reasoning and text generation capabilities .
Approach: They propose a rationale extraction method that leverages reasoning capabilities of large language models to extract the rationales necessary for answering a query.
Outcome: The proposed method is compared with baseline methods on two tasks across three datasets.
Mitigating Hallucinations of Large Language Models in Medical Information Extraction via Contrastive Decoding (2024.findings-emnlp)

Copied to clipboard

Challenge: Medical Information Extraction (MIE) tasks are a fundamental component of medical NLP.
Approach: They propose an alternative adaptive constraint strategy to adjust the scale and scope of contrastive tokens.
Outcome: The proposed approach selectively enhances the identification and classification capabilities while minimizing the influence of other inherent abilities in LLMs.
Revisiting the Self-Consistency Challenges in Multi-Choice Question Formats for Large Language Model Evaluation (2024.lrec-main)

Copied to clipboard

Challenge: Multi-choice questions (MCQs) are a common method for assessing the world knowledge of large language models.
Approach: They propose three knowledge-equivalent question variants to assess LLMs' world knowledge . they propose option position shuffle, option label replacement, and conversion to a True/False format .
Outcome: The proposed questions are shuffle, label replacement, and True/False format.
LoRA-drop: Efficient LoRA Parameter Pruning based on Output Evaluation (2025.coling-main)

Copied to clipboard

Challenge: Low-Rank Adaptation (LoRA) is currently the most commonly used PEFT method for fine-tuning models with billions of parameters.
Approach: They propose to use low-rank Adaptation to evaluate LoRA parameter features and then retain LoRA for important layers and the other layers share the same LoRA.
Outcome: The proposed method achieves comparable performance to full fine-tuning and LoRA while retaining 50% of the LoRA parameters on average.
AdapTime: Enabling Adaptive Temporal Reasoning in Large Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Existing methods for temporal reasoning are limited and apply a fixed pipeline to all questions.
Approach: They propose an adaptive temporal reasoning method that dynamically executes reasoning steps based on context and task requirements.
Outcome: Experiments on two temporal QA benchmarks show the proposed method works.
When LLMs Meets Acoustic Landmarks: An Efficient Approach to Integrate Speech into Large Language Models for Depression Detection (2024.emnlp-main)

Copied to clipboard

Challenge: Large language models (LLMs) are used for depression detection but their application remains unexplored.
Approach: They propose to integrate acoustic speech information into LLMs for depression detection by integrating aural landmarks into the framework.
Outcome: The proposed method adds critical dimensions to speech transcripts and provides insights into the unique speech patterns of individuals.
Learning How and What to Memorize: Cognition-Inspired Two-Stage Optimization for Evolving Memory (2026.acl-long)

Copied to clipboard

Challenge: Existing memory systems rely on static, hand-crafted update rules for personalization, but sparse outcome rewards provide weak supervision, resulting in unstable long-horizon optimization.
Approach: They propose a memory guideline optimization framework that learns how memory should be organized and what information to update.
Outcome: The proposed framework learns how memory should be organized and what information to update.
PsychEthicsBench: Evaluating Large Language Models Against Australian Mental Health Ethics (2026.findings-acl)

Copied to clipboard

Challenge: Mental disorders affect nearly one in seven people worldwide, yet the vast majority do not receive adequate care.
Approach: They propose a framework to evaluate LLMs' ethical knowledge and behavioral responses through multiple-choice and open-ended tasks with fine-grained ethicality annotations.
Outcome: Empirical results across 14 models reveal that refusal rates are poor indicators of ethical behavior, revealing a significant divergence between safety triggers and clinical appropriateness.
PRIME: A Process-Outcome Alignment Benchmark for Verifiable Reasoning in Mathematics and Engineering (2026.acl-long)

Copied to clipboard

Challenge: Current outcome-centric verification paradigms neglect potential errors in the derivation process.
Approach: They propose a process-aware RLVR training paradigm utilizing verifiers selected via **PRIME**.
Outcome: The proposed approach outperforms the baseline verification paradigm on AIME24, AIME25, and Beyond-AIME models.
A Multi-Expert Structural-Semantic Hybrid Framework for Unveiling Historical Patterns in Temporal Knowledge Graphs (2025.findings-acl)

Copied to clipboard

Challenge: Existing methods focus on graph structure learning or semantic reasoning, lacking the capability to capture the inherent differences between historical and non-historical events.
Approach: They propose a temporal knowledge graph reasoning framework that integrates both structural and semantic information to guide the reasoning process for different events.
Outcome: The proposed framework integrates structural and semantic information to predict future events . it can provide evidence for many downstream tasks, including situation analysis and political decision making .
TP-Link: Fine-grained Pre-Training for Text-to-SQL Parsing with Linking Information (2024.lrec-main)

Copied to clipboard

Challenge: Existing pre-training frameworks for text-to-SQL parsing have shown inherent differences in distributions between tables and plain text.
Approach: They propose a framework to improve context-dependent Text-to-SQL parsing by leveraging Linking information.
Outcome: The proposed framework achieves state-of-the-art performance on two leading downstream benchmarks.
LPO: Towards Accurate GUI Agent Interaction via Location Preference Optimization (2026.findings-acl)

Copied to clipboard

Challenge: Existing strategies for spatial localization are limited due to their limited capacity to perceive positional data.
Approach: They propose a location-based approach that leverages locational data to optimize interaction preferences.
Outcome: The proposed approach achieves SOTA results across offline benchmarks and real-world evaluations.
QuantAgents: Towards Multi-agent Financial System via Simulated Trading (2025.findings-emnlp)

Copied to clipboard

Challenge: Existing LLM-based agent models exhibit significant deviations from real-world fund companies.
Approach: They propose a multi-agent financial system that incorporates simulated trading . they propose simulated trades are evaluated without assuming actual risks .
Outcome: The proposed system evaluates various investment strategies without assuming actual risks without involving real-world investors.
When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Life (2026.findings-acl)

Copied to clipboard

Challenge: Safety impact of Multimodal Large Language Models (MLLMs) on human behavior is evaluated in this study.
Approach: They propose a safety-warning-based evaluation framework that encourages models to provide clear and informative safety warnings, rather than generic refusals.
Outcome: The proposed safety-warning-based evaluation framework encourages models to provide clear and informative safety warnings, rather than generic refusals.
MultiDx: A Multi-Source Knowledge Integration Framework towards Diagnostic Reasoning (2026.findings-acl)

Copied to clipboard

Challenge: Existing approaches focus on diagnostic reasoning based on internal model knowledge or static knowledge bases.
Approach: They propose a two-stage diagnostic reasoning framework that integrates multi-perspective evidence to generate a diagnostic prediction.
Outcome: The proposed method generates suspected diagnoses and reasoning traces from web search, SOAP-formatted case, and clinical case database.
SemAttack: Natural Textual Attacks via Different Semantic Spaces (2022.findings-naacl)

Copied to clipboard

Challenge: Existing approaches to attack pre-trained language models suffer from low success rates or fail to search efficiently in the exponentially large perturbation space.
Approach: They propose an efficient framework to generate natural adversarial text by constructing different semantic perturbation functions.
Outcome: The proposed framework generates natural adversarial texts for different languages with high success rates.
BalanceSFT: Improving LLM Function Calling with Balanced Training Signals and Data Hardness (2026.findings-acl)

Copied to clipboard

Challenge: Currently, Supervised Fine-Tuning (SFT) is the prevailing method for equipping Large Language Models (LLMs) with function calling capabilities, but its effectiveness is often compromised by two challenges: 1) lengthy Chain-of-Thought (CoT) reasoning tokens dominate training signals over concise function calls in the learning objective; 2) scarcity of hard training examples.
Approach: They propose a framework that uses a self-adjusted signal balancing loss and a hard data re-sampling strategy to selectively generate new, high-quality complex data guided by model errors.
Outcome: The proposed framework surpasses state-of-the-art models like GPT-5 in function calling performance.
MirrorCAPTCHA: Wild CAPTCHA, Wild Distribution, Wild Web-based Platform Meet Multimodal LLM Agents (2026.acl-long)

Copied to clipboard

Challenge: Existing agent benchmarks fail to evaluate an agent's real-world capacity to handle CAPTCHA . Existing benchmarks ignore this practical challenge, failing to evaluate agents' ability to handle complex visual CAPTchas.
Approach: They propose a benchmark annotated with Weighted Pass Rate and a new metric to measure agent's ability to handle CAPTCHA.
Outcome: The proposed benchmark outperforms current state-of-the-art closed-source models on mirrorCAPTCHA and achieves 9.4% higher average weighted pass rate and 2.13% higher average Completion degree.
Think in Safety: Unveiling and Mitigating Safety Alignment Collapse in Multimodal Large Reasoning Model (2025.emnlp-main)

Copied to clipboard

Challenge: Existing Large Reasoning Models have demonstrated broad application potential, yet their safety and reliability remain critical concerns.
Approach: They conduct a safety evaluation of 13 MLRMs across 5 benchmarks and examine their safety performance.
Outcome: The proposed model improves safety on jailbreak and safety-awareness benchmarks.
Dyve: Thinking Fast and Slow for Dynamic Process Verification (2025.emnlp-main)

Copied to clipboard

Challenge: Existing process verification methods struggle with reliably assessing incomplete reasoning traces and are limited by the cost of high-quality human annotations or the inherent noise in automatically generated labels.
Approach: They propose a dynamic process verifier that integrates fast and slow thinking to enhance reasoning error detection in large language models.
Outcome: The proposed system outperforms existing process-based verifiers and maintains computational efficiency while maintaining high performance.
Guideline Compliance in Task-Oriented Dialogue: The Chained Prior Approach (2025.findings-naacl)

Copied to clipboard

Challenge: Existing solutions based on large language models cannot achieve strict guideline compliance . a novel TOD system is being developed to improve guideline adherence .
Approach: They propose a task-oriented dialogue system that explicitly considers domain-specific guidelines by integrating a policy module.
Outcome: The proposed system achieves 20% better guideline compliance than state-of-the-art solutions.
Multi-perspective Improvement of Knowledge Graph Completion with Large Language Models (2024.lrec-main)

Copied to clipboard

Challenge: Knowledge graph completion (KGC) is a widely used method to tackle incompleteness in knowledge graphs (KGs).
Approach: They propose a general framework to compensate for the deficiency of contextualized knowledge by querying large language models from various perspectives.
Outcome: The proposed framework improves knowledge graph completion (KGC) by querying large language models from various perspectives.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations