Papers by Pengyu Zhu
LECO: Improving Early Exiting via Learned Exits and Comparison-based Exiting Mechanism (2023.acl-srw)
Copied to clipboard
| Challenge: | Recent work on dynamic early exiting has neglected the intermediate exits’ architectural designs. |
| Approach: | They propose a framework for learning exits and COmparison-based early exiting to improve PTMs’ early exit performance. |
| Outcome: | The proposed framework achieves the SOTA performance on multi-exit BERT training and dynamic early exiting on pre-trained models. |
Adaptive Spatial and Temporal Redundancy Optimization for Efficient Reasoning in Large Language Models (2026.acl-long)
Copied to clipboard
Tianle Chen, Pengyu Cheng, Qiyuan Zhu, Jiacheng Wang, Bei Liu, Hao Gu, Ruijie Shen, Xiaofeng Hou, Sirui Han, Jiacheng Liu
| Challenge: | Existing research to improve CoT efficiency falls into three categories, each with distinct limitations. |
| Approach: | They propose a training-free framework that addresses both dimensions of CoT reasoning by applying a progressive precision reduction strategy coupled with an entropy-based confidence mechanism for adaptive termination. |
| Outcome: | Empirical results show that the proposed framework achieves 11.3 efficiency gain without compromising accuracy. |
Unsupervised Morphological Tree Tokenizer (2025.findings-acl)
Copied to clipboard
| Challenge: | Conventional statistical tokenizers often disrupt constituent boundaries within words, thereby corrupting semantic information. |
| Approach: | They propose a method that uses morphological structure guidance to induce character-level structures of words by training a deep model. |
| Outcome: | Empirical results show that the proposed method retains complete morphemes and outperforms existing methods on morphological segmentation and language modeling tasks. |
Sparsity-Accelerated Training for Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated proficiency across various NLP tasks but often require additional training, such as continual pre-training and supervised fine-tuning. |
| Approach: | They propose to leverage sparsity in pre-trained LLMs to accelerate training by disregarding computations for unimportant neurons. |
| Outcome: | The proposed framework achieves comparable or superior performance to standard training while significantly accelerating the process. |
DemonAgent: Dynamically Encrypted Multi-Backdoor Implantation Attack on LLM-based Agent (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a new method for detecting advanced backdoors is proposed to bypass safety audits. |
| Approach: | They propose a backdoor implantation strategy that introduces dynamic encryption to bypass safety audits. |
| Outcome: | The proposed method achieves an attack success rate approaching 100% while maintaining a detection rate of 0%. |
WildGraphBench: Benchmarking GraphRAG with Wild-Source Corpora (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks for Graph-based Retrieval-Augmented Generation (GraphRAG) rely on short, curated passages as external knowledge, failing to adequately evaluate systems in realistic settings involving long contexts and large-scale heterogeneous documents. |
| Approach: | They propose a benchmark to assess GraphRAG performance in the wild using Wikipedia's unique structure where cohesive narratives are grounded in long and heterogeneous external reference documents. |
| Outcome: | Experiments with articles across 12 top-level topics show that GraphRAG performs better in the wild than existing methods. |
Adversarial Preference Learning for Robust LLM Alignment (2025.findings-acl)
Copied to clipboard
Yuanfu Wang, Pengyu Wang, Chenyang Xi, Bo Tang, Junyi Zhu, Wenqiang Wei, Chen Chen, Chao Yang, Jingfeng Zhang, Chaochao Lu, Yijun Niu, Keming Mao, Zhiyu Li, Feiyu Xiong, Jie Hu, Mingchuan Yang
| Challenge: | Modern language models rely on Reinforcement Learning from Human Feedback (RLHF) to encourage safe behaviors, but they remain vulnerable to adversarial attacks due to three key limitations: (1) the inefficiency and high cost of human annotation; (2) the vast diversity of potential adversarials; and (3) the risk of feedback bias and reward hacking. |
| Approach: | They propose an iterative adversarial training method that incorporates three key innovations to address these challenges. |
| Outcome: | Experiments on Mistral-7B-Instruct-v0.3 show that the proposed method significantly enhances robustness and reduces harmful outputs from 5.88% to 0.43%. |
DecIF: Improving Instruction-Following through Decomposition (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to obtain high-quality instruction-following data rely heavily on existing documents and existing methods. |
| Approach: | They propose a data synthesis framework, DecIF, which automatically generates accurate and diverse instruction-following data from scratch for supervised fine-tuning and reinforcement learning. |
| Outcome: | Extensive experiments show that the proposed framework can synthesize accurate instruction-following data for both SFT and RL paradigms compared to baselines. |
Generative Pretrained Structured Transformers: Unsupervised Syntactic Language Models at Scale (2024.acl-long)
Copied to clipboard
| Challenge: | Existing syntactic language models require a gold tree and sequential training to generate sentences. |
| Approach: | They propose an unsupervised syntactic language model that incrementally generates a sentence with its syntaktic tree in a left-to-right manner. |
| Outcome: | The proposed model outperforms existing models on grammar induction and comprehension tasks while holding a substantial acceleration on training. |