Papers by Mu Yang

15 papers
Cross-layer Attention Sharing for Pre-trained Large Language Models (2026.tacl-1)

Copied to clipboard

Challenge: Existing studies focus on compressing the Key-Value cache or grouping attention heads, while overlooking redundancy between layers.
Approach: They propose a lightweight substitute for self-attention in well-trained LLMs that uses feed-forward networks to align attention heads between adjacent layers and low-rank matrices to approximate differences in layer-wise attention weights.
Outcome: The proposed model reduces redundancy by sharing weights across layers while maintaining high response quality while reducing redundant calculations within 53% 84% of the total layers.
Controllable Contrastive Generation for Multilingual Biomedical Entity Linking (2023.emnlp-main)

Copied to clipboard

Challenge: Multilingual biomedical entity linking (MBEL) aims to map language-specific mentions in biomedically text to standardized concepts in a multilingual knowledge base (KB).
Approach: They propose a prompt-based controllable contrastive generation framework for MBEL which summarizes multidimensional information of the UMLS concept mentioned in biomedical text into a natural sentence following a predefined template.
Outcome: The proposed framework matches against UMLS concepts in as many languages and types as possible, thus facilitating cross-information disambiguation.
Biomedical Event Extraction with Hierarchical Knowledge Graphs (2020.findings-emnlp)

Copied to clipboard

Challenge: Biomedical event extraction is critical in understanding biomolecular interactions described in scientific corpus.
Approach: They propose to integrate domain knowledge from Unified Medical Language System (UMLS) to a pre-trained language model using Graph Edge-conditioned Attention Networks and hierarchical graph representation.
Outcome: The proposed approach achieves 1.41% F1 and 3.19% F1 improvements on the BioNLP 2011 GENIA Event Extraction task.
Beyond Static Evaluation: A Dynamic Approach to Assessing AI Assistants’ API Invocation Capabilities (2024.lrec-main)

Copied to clipboard

Challenge: Existing evaluation methods for human-machine interactions are static and can be misleading.
Approach: They propose to use a LLM-based user agent to assess an assistant's API call capability without human involvement.
Outcome: The proposed method mirrors real human conversation patterns in human-machine interactions, and shows that it aligns more closely with human assessment.
Stealthy Jailbreak Attacks on Large Language Models via Benign Data Mirroring (2025.naacl-long)

Copied to clipboard

Challenge: Existing black-box jailbreak methods often rely on model feedback . existing methods may be intercepted by content moderators during the search process .
Approach: They propose a method that guides malicious prompt construction by local training a mirror model of the target black-box model through benign data distillation.
Outcome: The proposed method achieves a 92% attack success rate and 80% stealth rate on a subset of AdvBench.
Training Language Model to Critique for Better Refinement (2025.findings-acl)

Copied to clipboard

Challenge: Large language models (LLMs) have remarkable evaluation and critique capabilities, providing insightful feedback and identifying flaws in various tasks.
Approach: They propose a framework to train critic models using refinement signals to generate feedback loops where critiques guide the model in refining its responses.
Outcome: The proposed framework outperforms traditional methods and open-source models in terms of critique quality and refinement outcomes.
Headword-Oriented Entity Linking: A Special Entity Linking Task with Dataset and Baseline (2020.lrec-1)

Copied to clipboard

Challenge: specialized entity linking problem involves linking only headwords of entities to knowledge bases . full product names are rarely written in context, instead abbreviated to shorter, irregular versions .
Approach: They propose a specialized entity linking problem where only the headwords of entities are to be linked to knowledge bases.
Outcome: The proposed model provides a strong benchmark performance on the special task.
EventPlus: A Temporal Event Understanding Pipeline (2021.naacl-demos)

Copied to clipboard

Challenge: Event information is a type of common sense knowledge that helps people understand how stories evolve and provides predictive hints for future events.
Approach: They propose a temporal event understanding pipeline that integrates state-of-the-art components.
Outcome: The proposed pipeline can be easily adapted to other domains, including biomedical domains.
Concise and Precise Context Compression for Tool-Using Language Models (2024.findings-acl)

Copied to clipboard

Challenge: Existing methods suffer from key information loss and difficulty in adjusting the length of compressed sequences based on documentation lengths.
Approach: They propose two strategies for compressing tool documentation into concise and precise summary sequences for tool-using language models.
Outcome: The proposed approach achieves comparable performance to the upper-bound baseline under 16x compression ratio.
A Cheaper and Better Diffusion Language Model with Soft-Masked Noise (2023.emnlp-main)

Copied to clipboard

Challenge: Existing diffusion models have limitations in modeling discrete data, e.g., languages . we present a novel diffusion model for language modeling inspired by linguistic features in languages based on iterative denoising .
Approach: They propose a method that iteratively denoises and adds corruptions to the textual data through soft-masking to better noise it.
Outcome: The proposed model achieves better generation quality and lower training cost than current models with better performance.
Self-Constructed Context Decompilation with Fined-grained Alignment Enhancement (2024.findings-emnlp)

Copied to clipboard

Challenge: Decompilation is the process of converting compiled code back into a high-level programming language for analysis when source code is unavailable.
Approach: They propose two methods to improve decompilation performance without fine-tuning and fine-grained alignment enhancement to achieve further improvements.
Outcome: The proposed methods achieved a Re-Executability performance improvement of approximately 3.90% on the Decompile-Eval benchmark, establishing a new state-of-the-art performance of 52.41%.
Vulnerability of Text-to-Image Models to Prompt Template Stealing: A Differential Evolution Approach (2025.findings-acl)

Copied to clipboard

Challenge: Prompt trading has emerged as a significant intellectual property concern in recent years, where vendors entice users by showcasing sample images before selling prompt templates that can generate similar images.
Approach: They propose a prompt-stealing benchmark consisting of 50 templates and 450 images organized into Easy and Hard difficulty levels.
Outcome: The proposed method outperforms baseline methods with an average improvement of over 10%.
DenseLoRA: Dense Low-Rank Adaptation of Large Language Models (2025.acl-long)

Copied to clipboard

Challenge: Low-rank adaptation (LoRA) is an efficient approach for adapting large language models (LLMs) but many of the weights in these matrices are redundant, leading to inefficiencies in parameter utilization.
Approach: They propose a low-rank adaptation approach that fine-tunes two low-ranked matrices and adapts them through a dense low-Rank matrix, improving parameter utilization and adaptation efficiency.
Outcome: The proposed approach achieves 83.8% accuracy with only 0.01% of trainable parameters compared to LoRA's 80.8% with 0.70% of trainability parameters on LLaMA3-8B.
Self-Reflective Generation at Test Time (2026.acl-long)

Copied to clipboard

Challenge: Existing self-reflection mechanisms are reactive and inefficient for large language models . a fundamental tension persists between the ability to execute complex multi-step reasoning and the ability of the model to generate coherent traces.
Approach: They propose a test-time framework that reflects before generating at uncertain points . SRGen utilizes dynamic entropy thresholding to identify high-uncertainty tokens .
Outcome: The proposed framework can significantly strengthen large language models' reasoning process.
LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for enhancing multi-step reasoning have not fully translated to multilingual contexts.
Approach: They propose a framework that leverages language-conditioned hints to guide exploration in non-English reasoning tasks.
Outcome: Empirical results show that the proposed framework improves reasoning performance without compromising language consistency.

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