Papers by Hui Xue

19 papers
From Adversarial Arms Race to Model-centric Evaluation: Motivating a Unified Automatic Robustness Evaluation Framework (2023.findings-acl)

Copied to clipboard

Challenge: Existing models of robustness evaluation are incomprehensive, impractical, and invalid .
Approach: They propose a framework for automatic robustness evaluation that shifts towards model-centric evaluation to further exploit the advantages of adversarial attacks.
Outcome: The proposed framework is based on a model-centric evaluation protocol and a robustness evaluation protocol.
KEHRL: Learning Knowledge-Enhanced Language Representations with Hierarchical Reinforcement Learning (2024.lrec-main)

Copied to clipboard

Challenge: General pre-trained language models (PLMs) leverage relation triples from knowledge graphs (KGs) and integrate external data sources into language models via self-supervised learning.
Approach: They propose to learn Knowledge-Enhanced language representations with Hierarchical Reinforcement Learning (KEHRL) to detect positions for knowledge injection and integrate external knowledge into the model to avoid injecting inaccurate or irrelevant knowledge.
Outcome: The proposed model can detect essential positions in texts for knowledge injection and integrate external knowledge into the model to avoid injecting inaccurate or irrelevant knowledge.
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Existing literature on mechanistic interpretation (MI) treats it as an observational science, leaving practical applications underexplored.
Approach: They propose a survey structured around the pipeline to identify and improve MI models.
Outcome: The proposed framework enables tangible improvements in Alignment, Capability, and Efficiency.
Sparse Black-Box Multimodal Attack for Vision-Language Adversary Generation (2023.findings-emnlp)

Copied to clipboard

Challenge: Existing adversarial attacks using imperceptible perturbations are challenging to simulate . e-commerce product restrictions and hate speech monitoring are examples of such attacks .
Approach: They propose a black-box adversarial attack that leverages sparse perturbations to simulate adversarials exhibited by illegal merchants in the black- box scenario.
Outcome: The proposed method outperforms existing attacks and unimodal attacks by treating images and text in discrete space and outperforming existing models.
TRELM: Towards Robust and Efficient Pre-training for Knowledge-Enhanced Language Models (2024.lrec-main)

Copied to clipboard

Challenge: Existing methods for incorporating external knowledge into language models do not prioritize learning embeddings for entity-related tokens.
Approach: They propose a framework for incorporating external knowledge into pre-training models that utilize entity-related tokens.
Outcome: The proposed framework reduces pre-training time by 50% and outperforms other KEPLMs in knowledge probing tasks and multiple knowledge-aware language understanding tasks.
LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety (2026.acl-long)

Copied to clipboard

Challenge: Large language models (LLMs) have demonstrated better safety performance in high-resource languages than in low-resourced languages.
Approach: They propose language-agnostic semantic alignment (LASA) which anchors safety alignment directly in semantic bottlenecks.
Outcome: The proposed approach significantly improves safety across all languages: average attack success rate drops from 24.7% to 2.8% on LLaMA-3.1-8B-Instruct and remains within 3–4% across Qwen2.5 and Qwend3 Instruct models (7B–32B).
Large Language Models Can be Lazy Learners: Analyze Shortcuts in In-Context Learning (2023.findings-acl)

Copied to clipboard

Challenge: Large language models (LLMs) have shown great potential for in-context learning, but their robustness and performance on downstream tasks remains limited.
Approach: They propose to examine the reliance of LLMs on shortcuts or spurious correlations within prompts for downstream tasks and find larger models are more likely to utilize shortcuts in prompts during inference.
Outcome: The proposed model is “lazy learner” and more likely to use shortcuts in prompts during inference.
SpanMlt: A Span-based Multi-Task Learning Framework for Pair-wise Aspect and Opinion Terms Extraction (2020.acl-main)

Copied to clipboard

Challenge: Aspect terms and opinion terms are key problems of fine-grained aspect-based sentiment analysis.
Approach: They propose a method to extract aspect and opinion terms as pairs from a sentence . they use shared spans to extract the terms under supervision of span boundaries .
Outcome: The proposed method outperforms state-of-the-art methods on both aspects and opinion terms extraction tasks.
Seeing but Not Thinking: Routing Distraction in Multimodal Mixture-of-Experts (2026.acl-long)

Copied to clipboard

Challenge: Existing multimodal Mixture-of-Experts models accurately perceive image content yet fail in subsequent reasoning . Seeing but not thinking phenomenon is a puzzling phenomenon .
Approach: They propose a routing-guided intervention method that enhances domain expert activation.
Outcome: The proposed method achieves consistent improvements on visual reasoning tasks.
START: Self-taught Reasoner with Tools (2025.emnlp-main)

Copied to clipboard

Challenge: Large Reasoning Models (LRMs) have demonstrated remarkable capabilities in complex reasoning through long chain-of-thought, yet they struggle with precise computations and algorithmic operations.
Approach: They propose a training-free approach that activates LRMs’ latent tool-use capabilities through artificial hints and a framework that enables models to learn effective tool utilization through diverse hint patterns and rejection-based data synthesis.
Outcome: Experiments show that START significantly improves state-of-the-art LRMs across challenging benchmarks, including competition-level mathematics (AMC23: 95.0%, AIME24: 75.6%) and graduate-level science questions (GPQA: 64.6%).
Safe-SAIL: Towards a Fine-grained Safety Landscape of Large Language Models via Sparse Autoencoder Interpretation Framework (2026.findings-acl)

Copied to clipboard

Challenge: Existing studies on how SAEs derive most fine-grained latent features for safety remain unexplored.
Approach: They propose a framework for interpreting SAE features in safety-critical domains . they train a suite of SAEs with human-readable explanations and systematic evaluations based on pornography, politics, violence, and terror .
Outcome: The proposed framework reduces interpretation cost by 55% and improves safety-critical features.
The Right Time Matters: Data Arrangement Affects Zero-Shot Generalization in Instruction Tuning (2025.findings-acl)

Copied to clipboard

Challenge: Existing work on instruction tuning has focused on task level, without considering that tasks are artificially defined and, to LLMs, merely consist of tokens and representations.
Approach: They propose a training data arrangement framework that allows for continual learning and loss reduction.
Outcome: The proposed framework promotes continual learning and loss reduction on unseen tasks.
How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities (2026.acl-long)

Copied to clipboard

Challenge: Large language models (LLMs) are increasingly deployed in socially sensitive domains, yet their unpredictable behaviors pose significant risks.
Approach: They propose a hierarchical benchmark for evaluating LLM controllability across three domains: language features, sentiment, and personality.
Outcome: The proposed framework offers a principled and interpretable framework for safe and controllable LLM behavior serving as a foundation for future research.
RoChBert: Towards Robust BERT Fine-tuning for Chinese (2022.findings-emnlp)

Copied to clipboard

Challenge: Pre-trained language models (e.g., BERT) have been proved vulnerable to adversarial texts.
Approach: They propose to fuse Chinese phonetic and glyph features into pre-trained models by using a more comprehensive adversarial graph.
Outcome: The proposed framework outperforms existing methods in significant ways on a wide range of tasks while remaining accurate on benign texts.
Supervised Prototypical Contrastive Learning for Emotion Recognition in Conversation (2022.emnlp-main)

Copied to clipboard

Challenge: Existing methods to capture emotions in conversation (ERC) lack the correlation between emotions and semantics, resulting in many challenges.
Approach: They propose a Supervised Prototypical Contrastive Learning (SPCL) loss for the ERC task . they use a Prototype Network to leverage the supervised contrastive learning approach .
Outcome: The proposed approach outperforms CoG-BART's proposed approach on three widely used benchmarks and shows that it is effective on multiple scenarios.
Multiple Instance Learning for Offensive Language Detection (2022.findings-emnlp)

Copied to clipboard

Challenge: Existing approaches to detect offensive content are expensive and require massive manual effort.
Approach: They propose an approach capable of utilizing the bag-level labeled data for offensive language detection by an annotation-based model.
Outcome: The proposed model can detect offensive language on both bag-level and sentence level.
WenetSpeech-Wu: Datasets, Benchmarks, and Models for a Unified Chinese Wu Dialect Speech Processing Ecosystem (2026.findings-acl)

Copied to clipboard

Challenge: despite its linguistic significance, the Wu dialect of Chinese has long been hindered by the lack of large-scale speech data, standardized evaluation benchmarks, and publicly available models.
Approach: They propose to use WenetSpeech-Wu as a large-scale, multi-dimensionally annotated open-source speech corpus for the Wu dialect of Chinese.
Outcome: The proposed dataset includes 8,000 hours of speech data and strong open-source models . the proposed dataset is competitive and empirically validated .
Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics (2026.acl-long)

Copied to clipboard

Challenge: Methods for controlling large language models (LLMs) are often studied in isolation, obscuring connections and making comparison difficult.
Approach: They propose a preference-utility analysis that separates control effects into preference and utility, and measures both on a shared log-odds scale using polarity-paired contrastive examples.
Outcome: The proposed approach improves preference while preserving utility.
UniPSDA: Unsupervised Pseudo Semantic Data Augmentation for Zero-Shot Cross-Lingual Natural Language Understanding (2024.lrec-main)

Copied to clipboard

Challenge: Existing studies rely on shallow unsupervised data generated by token surface matching regardless of global context-aware semantics of the surrounding text tokens.
Approach: They propose an Unsupervised Pseudo Semantic Data Augmentation mechanism to enrich training data without human intervention.
Outcome: The proposed model improves on general zero-shot cross-lingual understanding tasks on different languages without human intervention.

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