Papers by Di Sun

25 papers
Union-of-Experts: Neurons in Mixture-of-Experts are Secretly Routers (2026.acl-long)

Copied to clipboard

Challenge: Mixture-of-Experts (MoE) models rely on an external router to assign tokens to experts, resulting in suboptimal performance.
Approach: They propose an MoE variant that performs "expert-autonomous routing" by pre-designating a fraction of neurons within each expert as "routing neurons" they pre-train UoE models with up to 3B parameters and show they outperform traditional MoEs with matched efficiency.
Outcome: The proposed model outperforms existing models with 3B parameters and provides valuable insights into expert-autonomous selection and the broader routing mechanisms of MoE models.
Label Effects: Shared Heuristic Reliance in Trust Assessment by Humans and LLM-as-a-Judge (2026.acl-long)

Copied to clipboard

Challenge: Large language models (LLMs) are increasingly used as automated evaluators . et al., 2024: strong labels can foster trust but also undermine it .
Approach: They show that LLMs' source labels bias trust judgments by humans . they use eye-tracking data to analyze LLM internal states during judgment .
Outcome: The proposed model is biased by disclosed source labels, the authors show . eye-tracking data show humans rely heavily on source labels for judgments .
ShieldLM: Empowering LLMs as Aligned, Customizable and Explainable Safety Detectors (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing tools for detecting safety issues in LLMs are expensive and inefficient.
Approach: They propose an LLM-based safety detector which annotates the safety of queries and provides explanations for its decisions.
Outcome: The proposed detector outperforms baselines on four sets of query-response pairs and is effective as a safety evaluator for advanced LLMs.
ConKI: Contrastive Knowledge Injection for Multimodal Sentiment Analysis (2023.findings-acl)

Copied to clipboard

Challenge: Existing methods for multimodal sentiment analysis focus on general knowledge, which is inadequate to identify specific sentiments across modalities.
Approach: They propose a method where specific-knowledge representations for each modality can be learned together with general knowledge representations via knowledge injection based on an adapter architecture.
Outcome: The proposed method outperforms all prior methods on three popular benchmarks on multimodal sentiment analysis metrics.
D-Artemis: A Deliberative Cognitive Framework for Mobile GUI Multi-Agents (2026.findings-acl)

Copied to clipboard

Challenge: Graphical User Interface (GUI) agents aim to automate a wide spectrum of human tasks by emulating user interaction.
Approach: They propose a deliberative framework that leverages a fine-grained tip retrieval mechanism to inform its decision-making process.
Outcome: The proposed framework achieves SOTA among open-source general models on AndroidWorld and ScreenSpot-V2 . it leverages a fine-grained, app-specific tip retrieval mechanism to inform its decision-making process .
RoleMRC: A Fine-Grained Composite Benchmark for Role-Playing and Instruction-Following (2025.findings-acl)

Copied to clipboard

Challenge: Existing role-playing datasets mostly contribute to controlling role style and knowledge boundaries, but overlook role-following in instruction-follower scenarios.
Approach: They propose a fine-grained role-playing and instruction-following composite benchmark, named RoleMRC, which includes multi-turn dialogues between ideal roles and humans, including free chats or discussions upon given passages .
Outcome: The proposed model improves instruction-following without compromising general role-playing and reasoning capabilities.
Tell Me What You Don’t Know: Enhancing Refusal Capabilities of Role-Playing Agents via Representation Space Analysis and Editing (2025.findings-acl)

Copied to clipboard

Challenge: Role-playing Agents (RPAs) struggle to recognize and respond to hard queries that conflict with their role-play knowledge.
Approach: They propose a lightweight representation editing approach that conveniently shifts conflicting requests to the rejection region, thereby enhancing the model’s refusal accuracy.
Outcome: The proposed model improves RPAs’ refusal ability of conflicting requests while maintaining their general role-playing capabilities.
Prompting ChatGPT in MNER: Enhanced Multimodal Named Entity Recognition with Auxiliary Refined Knowledge (2023.findings-emnlp)

Copied to clipboard

Challenge: Existing methods to enhance textual entity prediction neglect the need for external knowledge or encounter high redundancy in the retrieved knowledge.
Approach: They propose a framework that leverages ChatGPT as an implicit knowledge base and heuristically generates auxiliary knowledge for more efficient entity prediction.
Outcome: The proposed framework outperforms state-of-the-art methods on two classic datasets and exhibits a stronger robustness and generalization capability.
VisAidMath: Benchmarking Visual-Aided Mathematical Reasoning (2026.acl-long)

Copied to clipboard

Challenge: Existing Large Multi-modal Models lack a robust visual processing capability that is often masked by evaluation metrics that prioritize final-answer accuracy.
Approach: They propose a three-layer evaluation framework that scrutinizes the generation of valid visual aids and the soundness of subsequent reasoning steps.
Outcome: The proposed framework examines the generation of valid visual aids and the soundness of subsequent reasoning steps on state-of-the-art models.
Eliminating Biased Length Reliance of Direct Preference Optimization via Down-Sampled KL Divergence (2024.emnlp-main)

Copied to clipboard

Challenge: Existing studies attributed verbosity to biased labels, but new research shows that DPO can be effective in mitigating verboses.
Approach: They propose to use a method to reduce the amount of verbosity in LLMs by using a downsampling approach.
Outcome: The proposed approach overcomes the problem of verbosity by reducing the length reliance of the proposed algorithm.
FIPO: Free-form Instruction-oriented Prompt Optimization with Preference Dataset and Modular Fine-tuning Schema (2025.coling-main)

Copied to clipboard

Challenge: naive prompts can enhance the task performance of large language models, but they are resource-intensive.
Approach: They propose an automatic prompt optimization method that refines naive prompts according to task outputs from in-box testing models.
Outcome: The proposed method is based on a large-scale dataset and performed fairly across multiple models.
HMoE: Heterogeneous Mixture of Experts for Language Modeling (2025.emnlp-main)

Copied to clipboard

Challenge: Mixture of Experts (MoE) models use homogeneous experts with diverse capacities, resulting in a lack of expert specialization and parameter utilization.
Approach: They propose a framework where experts differ in size and possess diverse capacities . they propose HMoE to encourage frequent activation of smaller experts .
Outcome: The proposed framework outperforms homogeneous homogenous MoE models on evaluation benchmarks and achieves lower loss rate with fewer activated parameters.
VP-MEL: Visual Prompts Guided Multimodal Entity Linking (2025.findings-acl)

Copied to clipboard

Challenge: Existing methods for multimodal entity linking rely on mention words as retrieval cues, which limits their ability to effectively utilize information from both images and text.
Approach: They propose a visual prompt-guided multimodal entity linking task for a text-image pair . they propose VPWiki to facilitate this task and a framework to capture latent information.
Outcome: The proposed framework outperforms baseline methods on a VPWiki dataset.
RolePlot: A Systematic Framework for Evaluating and Enhancing the Plot-Progression Capabilities of Role-Playing Agents (2025.acl-long)

Copied to clipboard

Challenge: Existing research has focused on role-playing agents’ ability to portray specified characters, but their ability to advance the plot requires substantial improvements to deliver more engaging interaction.
Approach: They propose a role-playing framework to evaluate and enhance the plot-progression capabilities of role-players.
Outcome: The proposed framework improves RPAs’ ability to time plot developments and yields a significant increase in conversation turns and sustained higher arousal levels.
Direct Preference Optimization of Video Large Multimodal Models from Language Model Reward (2025.naacl-long)

Copied to clipboard

Challenge: Existing studies have demonstrated that direct preference optimization (DPO) can be effective in generalizing large language models, but its effectiveness in video domain remains limited.
Approach: They propose a framework that utilizes detailed video captions as a proxy of video content to enable language models to incorporate this information as supporting evidence for scoring video Question Answering (QA) predictions.
Outcome: The proposed framework shows that it can be used to align language models with video content and improves performance on open-ended video QA tasks.
Sequential-NIAH: A Needle-In-A-Haystack Benchmark for Extracting Sequential Needles from Long Contexts (2025.emnlp-main)

Copied to clipboard

Challenge: Recent models have extended Corresponding Author. context lengths to millions of tokens while maintaining reasoning and comprehension capabilities.
Approach: They propose a benchmark to evaluate the ability of large language models to extract sequential information items from long contexts.
Outcome: The proposed model achieves maximum accuracy of 63.50% on six well-known LLMs.
KMatrix-2: A Comprehensive Heterogeneous Knowledge Collaborative Enhancement Toolkit for Large Language Model (2025.emnlp-demos)

Copied to clipboard

Challenge: Existing studies on K-LLMs systems focus on declarative knowledge and procedural knowledge (rules) .
Approach: They propose to build a toolkit that supports comprehensive heterogeneous knowledge collaborative enhancement for Large Language Models (LLMs).
Outcome: The proposed toolkit provides unified knowledge integration and joint knowledge retrieval methods to achieve more comprehensive heterogeneous knowledge collaborative enhancement.
The Security Threat of Compressed Projectors in Large Vision-Language Models (2025.findings-emnlp)

Copied to clipboard

Challenge: Mainstream VLPs have significant security implications, but their security implications have not been thoroughly examined.
Approach: a study evaluates the security of visual language projectors by comparing them to uncompressed projector.
Outcome: The evaluation reveals significant differences in security profiles between compressed and uncompressed projectors.
Rethinking the Video Sampling and Reasoning Strategies for Temporal Sentence Grounding (2022.findings-emnlp)

Copied to clipboard

Challenge: Existing methods for temporal sentence grounding ignore two crucial issues . 1) Boundary-bias: the video downsampling process may lose these two frames . 2) Reasoning-biases: such incorrect new boundary frames lead to the reasoning bias .
Approach: They propose a siamese sampling mechanism to generate additional contextual frames . they use a reasoning strategy to learn the inter-relationship among these frames a .
Outcome: Extensive experiments demonstrate the effectiveness of a new siamese sampling network on three challenging datasets.
LLMs as Bridges: Reformulating Grounded Multimodal Named Entity Recognition (2024.findings-acl)

Copied to clipboard

Challenge: Existing methods for Grounded Multimodal Named Entity Recognition (GMNER) lack a strong correlation between image-text pairs and is ungroundable.
Approach: They propose a framework that reformulates GMNER into a joint MNER-VE-VG task by leveraging large language models as a connecting bridge.
Outcome: The proposed framework outperforms state-of-the-art methods on the existing GMNER dataset and achieves absolute leads of 10.65%, 6.21%, and 8.83% in all three subtasks.
Ted-Tok: Maintaining an Evolving Vocabulary for Lifelong Learning (2026.acl-long)

Copied to clipboard

Challenge: a static tokenizer fragments newly emerging lexical items as language evolves . as language grows, a dynamic tokenizer reduces compression efficiency and performance .
Approach: They propose a Temporal Drift Tokenizer that maintains an evolving vocabulary that adapts to emerging linguistic patterns over time.
Outcome: The proposed tokenizer maintains an evolving vocabulary that adapts to emerging linguistic patterns over time.
Decoding at the Speed of Thought: Harnessing Parallel Decoding of Lexical Units for LLMs (2024.lrec-main)

Copied to clipboard

Challenge: Large language models have demonstrated exceptional capability in natural language understanding and generation, but their generation speed is limited by the inherently sequential nature of their decoding process.
Approach: They propose a method that accelerates decoding process without sacrificing quality . they propose lexical unit decoding, which can be integrated with other methods .
Outcome: The proposed method significantly reduces decoding time while maintaining quality while maintaining output quality.
MAC-SQL: A Multi-Agent Collaborative Framework for Text-to-SQL (2025.coling-main)

Copied to clipboard

Challenge: Recent LLM-based Text-to-SQL methods suffer from performance degradation on “huge” databases and complex user questions that require multi-step reasoning.
Approach: They propose a framework that integrates a decomposer agent and auxiliary agents to generate SQL queries from natural language text.
Outcome: The proposed framework achieves comparable execution accuracy on SQL-Llama tasks compared to the baseline model.
How Numerical Precision Affects Arithmetical Reasoning Capabilities of LLMs (2025.findings-acl)

Copied to clipboard

Challenge: Despite the success of transformer-based large language models, understanding and enhancing their mathematical capabilities remains a significant challenge.
Approach: They propose to use numerical precision as a key factor that influences LLMs' effectiveness in arithmetical tasks to determine their effectiveness.
Outcome: The proposed models perform better in arithmetic tasks than transformer-based models with standard numerical precision.
Parrot: Enhancing Multi-Turn Instruction Following for Large Language Models (2024.acl-long)

Copied to clipboard

Challenge: Existing studies overlook the multi-turn instruction following ability of large language models (LLMs) Extensive experiments show that Parrot improves current LLMs by up to 7.2% in multi- turn instruction following.
Approach: They propose a method for collecting multi-turn instructions that feature human-like queries, such as anaphora and ellipsis, and a context-aware preference optimization strategy to further enhance LLMs for complex queries.
Outcome: The proposed method improves existing LLMs by up to 7.2% in multi-turn instruction following.

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