Papers by Zhaofeng He

Interpretable Safety Alignment via SAE-Constructed Low-Rank Subspace Adaptation (2026.acl-long)

Copied to clipboard

Dianyun Wang, Qingsen Ma, Yuhu Shang, Zhifeng Lu, Zhenbo Xu, Lechen Ning, Huijia Wu, Zhaofeng He

Explainability & Interpretability in NLP

Challenge:	Prior work has shown that safety behaviors are governed by low-rank structures . Low-Rank Adaptation (LoRA) consistently underperforms full fine-tuning and reinforcement learning on safety benchmarks .
Approach:	They propose a safety alignment system that disentangles safety-relevant directions into monosemantic features and constructs an interpretable safety subspace from SAE directions.
Outcome:	Empirically, the proposed model achieves 99.6% safety rates across multiple model families and scales . low-rank Adaptation consistently underperforms full fine-tuning and reinforcement learning on safety benchmarks compared with previous methods .

Select-Then-Decompose: From Empirical Analysis to Adaptive Selection Strategy for Task Decomposition in Large Language Models (2025.emnlp-main)

Copied to clipboard

Shuodi Liu, Yingzhuo Liu, Zi Wang, Yusheng Wang, Huijia Wu, Liuyu Xiang, Zhaofeng He

Language Models

Challenge:	Existing task decomposition methods focus on memory, tool usage, and feedback mechanisms, but they often overlook the trade-off between performance and cost.
Approach:	They propose a strategy that selects the most suitable decomposition approach based on task characteristics and enhances the reliability of the results through a verification module.
Outcome:	The proposed strategy is based on categories of approaches, characteristics of tasks, and configuration of decomposition and execution models.

ChatEdit: Towards Multi-turn Interactive Facial Image Editing via Dialogue (2023.emnlp-main)

Copied to clipboard

Xing Cui, Zekun Li, Pei Li, Yibo Hu, Hailin Shi, Chunshui Cao, Zhaofeng He

Visual Data in NLP

Challenge:	Existing approaches to interactive facial image editing treat multi-turn editing as a sequence of successive single-turn edits, leading to attribute forgetting and error accumulation.
Approach:	They propose a framework for interactive facial image editing through dialogues based on the CelebA-HQ dataset and a benchmark dataset to evaluate this.
Outcome:	The proposed framework outperforms existing methods and improves existing ones.

SCVQ: Sparse-Compensated Vector Quantization for Large Language Models (2026.acl-long)

Copied to clipboard

Zixuan Zhou, Yujun Diao, Zicheng Kong, Dehua Ma, Zhenbo Xu, Pei Pei Li, Zhaofeng He

Semantic Text Processing

Challenge:	Existing vector quantization methods incur inference overhead due to massive codebook storage and intensive index lookups.
Approach:	They propose a framework for vector quantization that incorporates a salience-aware weighted K-means clustering scheme with symmetry constraints to reduce codebook size and indexing costs.
Outcome:	The proposed framework achieves a perplexity of 5.78 on WikiText-2 for LLaMA-2-7B at 2-bit quantization while delivering a 1.4 speedup over existing baselines.

HyperMoE: Towards Better Mixture of Experts via Transferring Among Experts (2024.acl-long)

Copied to clipboard

Hao Zhao, Zihan Qiu, Huijia Wu, Zili Wang, Zhaofeng He, Jie Fu

Language Models

Challenge:	Existing methods for enhancing performance through increased use of expert knowledge often result in diminishing sparsity during expert selection.
Approach:	They propose a framework that integrates the computational processes of MoE with the concept of knowledge transferring in multi-task learning.
Outcome:	The proposed framework outperforms existing methods under identical conditions concerning the number of experts.

Translation or Recitation? Calibrating Evaluation Scores for Machine Translation of Extremely Low-Resource Languages (2026.acl-short)

Copied to clipboard

Danlu Chen, Ka Sing He, Jiahe Tian, Chenghao Xiao, Zhaofeng Wu, Taylor Berg-Kirkpatrick, Freda Shi

Multilinguality

Challenge:	Existing studies show that performance across low-resource settings is variable, resulting in a significant barrier for the MT community.
Approach:	They propose to use FRED Difficulty Metrics to contextualize reported performance across different language pairs to determine whether breakthroughs reported in other contexts are artifacts of benchmark collection.
Outcome:	The proposed metrics explain a significant portion of result variability rather than model capability.

An Adaptive Prompt Generation Framework for Task-oriented Dialogue System (2023.findings-emnlp)

Copied to clipboard

Jun Gao, Liuyu Xiang, Huijia Wu, Han Zhao, Yiqi Tong, Zhaofeng He

Language Models

Challenge:	Existing black-box large language models (LLMs) have excellent performance in task-oriented dialogue (TOD) tasks, but obtaining suitable prompts for specific tasks is challenging.
Approach:	They propose a black-box large language model that generates domain and slot information in the belief state, which serves as prior knowledge for subsequent prompt generation.
Outcome:	The proposed framework outperforms existing prompting methods on the MultiWOZ 2.0 dataset.

Prototype-based HyperAdapter for Sample-Efficient Multi-task Tuning (2023.emnlp-main)

Copied to clipboard

Hao Zhao, Jie Fu, Zhaofeng He

Responsible & Trustworthy NLP

Challenge:	Existing methods for fine-tuning pre-trained language models are limited to low-data regimes and require learning different modules to adapt to diverse tasks.
Approach:	They propose a framework for parameter-efficient fine-tuning that trains modules per task . they use an instance-dense retriever and a prototypical hypernetwork to generate conditional modules .
Outcome:	The proposed framework outperforms existing methods on multi-task learning and few-shot transfer learning.

LLMArena: Assessing Capabilities of Large Language Models in Dynamic Multi-Agent Environments (2024.acl-long)

Copied to clipboard

Junzhe Chen, Xuming Hu, Shuodi Liu, Shiyu Huang, Wei-Wei Tu, Zhaofeng He, Lijie Wen

Language Models

Challenge:	Existing benchmarks for evaluating large language models use static datasets, leading to data leakage or overlooking the complexities of multi-agent interactions.
Approach:	They propose a framework that evaluates the diverse capabilities of LLM agents in multi-agent dynamic environments.
Outcome:	The proposed framework assesses the diverse capabilities of LLM agents in multi-agent dynamic environments.

SecDecoding: Steerable Decoding for Safer LLM Generation (2025.findings-emnlp)

Copied to clipboard

Jiayou Wang, Rundong Liu, Yue Hu, Huijia Wu, Zhaofeng He

Robustness in NLP

Challenge:	Existing decoding-time defense methods suffer from limited generalization, high computational overhead, or significant utility degradation.
Approach:	They propose a decoding-time defense framework that leverages a pair of small contrastive models to estimate token-level safety signals by measuring divergence in their output distributions.
Outcome:	The proposed framework achieves near-zero attack success rates against a wide spectrum of advanced jailbreak attacks while maintaining the model’s helpfulness with minimal degradation.

MORE-3S:Multimodal-based Offline Reinforcement Learning with Shared Semantic Spaces (2024.lrec-main)

Copied to clipboard

Tianyu Zheng, Ge Zhang, Xingwei Qu, Ming Kuang, Wenhao Huang, Zhaofeng He

Multimodality

Challenge:	Existing approaches to offline reinforcement learning (RL) focus on learning value functions or policy gradients, but they view it as a sequence modeling task.
Approach:	They propose a method that integrates multimodal and pre-trained language models to transform offline reinforcement learning into a supervised learning task by integrating state information derived from images and action-related data obtained from text.
Outcome:	The proposed approach outperforms baselines on Atari and OpenAI Gym environments while promoting long-term strategic thinking.