Academic Profile 学术主页

Mengyang Li

李蒙阳
Lecturer & Master's Supervisor 讲师 · 硕士生导师
College of Electronic & Communication Engineering, Tianjin Normal University 天津师范大学 · 电子与通信工程学院

I received my Ph.D. from Tianjin University (2021–2024) under Prof. Ou Wu, focusing on deep learning and data mining. My current research spans data optimization, preference optimization (RLHF/DPO for LLMs), and embodied intelligence. I have published papers at top venues including KDD, ICML, CVPR, ACL, ICLR, AAAI, IEEE TIP, and TNNLS.

博士毕业于天津大学（2021–2024），师从吴偶教授，研究方向为深度学习与数据挖掘。目前主要从事数据优化、偏好优化（面向大语言模型的RLHF/DPO）及具身智能研究。研究成果发表于KDD、ICML、CVPR、ACL、ICLR、AAAI、IEEE TIP、TNNLS等顶级期刊与会议。

Email Me发送邮件 Google Scholar GitHub ORCID
Mengyang Li
limengyang@tjnu.edu.cn
Tianjin, China中国·天津
Ph.D., Tianjin University博士·天津大学
01

Research Interests研究方向

⚡
Data Optimization数据优化
Training dynamics, sample weighting, curriculum learning, and data valuation for deep neural networks.
深度神经网络的训练动态、样本赋权、课程学习与数据估值。
🧠
Preference Optimization偏好优化
RLHF, DPO and beyond — aligning large language models with human preferences efficiently and robustly.
RLHF、DPO及大语言模型偏好对齐，提升对齐的效率与鲁棒性。
🤖
Embodied Intelligence具身智能
Perception–action loops, world models, and multi-modal reasoning for physically grounded agents.
感知-动作闭环、世界模型与多模态推理，面向物理环境中的智能体。
📊
Imbalanced Learning不均衡学习
Label noise, class imbalance, and local differential privacy for robust machine learning at scale.
标签噪声、类别不均衡与局部差分隐私，面向大规模鲁棒机器学习。
02

Experience主要经历

2025.02 — Present
Lecturer & Master's Supervisor讲师 · 硕士生导师
College of Electronic & Communication Engineering, Tianjin Normal University天津师范大学 · 电子与通信工程学院
2021.09 — 2024.11
Ph.D. — Deep Learning & Data Mining博士 · 深度学习与数据挖掘
Tianjin University, supervised by Prof. Ou Wu天津大学，师从吴偶教授
03

Publications学术论文

[1]
Dual-Difficulty Curriculum Learning for Direct Preference Optimization
Mengyang Li, Haozhan Geng, Zhong Zhang, Shuang Liu
KDD 2026 CCF-A
[2]
Layer-wise Gradient Disentanglement: Decoupling Semantics and Preferences in Direct Preference Optimization
Mengyang Li, Shuang Liu, Zhong Zhang
ICML 2026 CCF-A
[3]
Task-Aware Preference Calibration for Direct Preference Optimization
Mengyang Li, Zhong Zhang, Pinlong Zhao
ICML 2026 CCF-A
[4]
DABO: Difficulty-aware Bayesian Optimization with Diffusion-learned Priors
Mengyang Li, Pinlong Zhao
CVPR 2026 CCF-A Highlight
[5]
What Do LLMs Learn First? Asymmetric Learning Dynamics of Input Complexity and Output Ambiguity in Preference Alignment
Mengyang Li, Jingwen Wang, Pinlong Zhao
ACL 2026 CCF-A
[6]
Aligner, Diagnose Thyself: A Meta-learning Paradigm for Fusing Intrinsic Feedback in Preference Alignment
Mengyang Li, Pinlong Zhao
ICLR 2026 CCF-A
[7]
Difficulty-Aware Learning Curve Extrapolation
Mengyang Li, Pinlong Zhao
AAAI 2026 CCF-A
[8]
Delving into the Training Dynamics for Image Classification
Mengyang Li, Xiaoling Zhou, Ou Wu
IEEE TIP 2025 CCF-A CAS Q1 TOP · IF 13.7
[9]
Toward Learnable and Interpretable Data Shapley Valuation for Deep Learning
Mengyang Li, Weiyao Zhu, Ou Wu
KBS 2025 CAS Q1 TOP · IF 7.2
[10]
Class-level Logit Perturbation
Mengyang Li, Fengguang Su, Ou Wu, Ji Zhang
IEEE TNNLS 2024 CAS Q1 TOP · IF 8.9
[11]
Logit Perturbation
Mengyang Li, Fengguang Su, Ou Wu
AAAI 2022 CCF-A
[12]
Learning Temporally-Aware Sample Weights for Preference Optimization
Mengyang Li, Xudong Zhou, Pinlong Zhao
ACL Findings 2026 CCF-A
[13]
ECOC-IL: Robust and Efficient Label LDP for Imbalanced Learning
Mengyang Li, Ou Wu
CVPR Findings 2026 CCF-A
[14]
What Tokens Truly Matter? The Logit Conflation Problem in LLM Sampling
Pinlong Zhao, Huijun Tang, Pengfei Jiao, Mengyang Li (corresponding)
ACL Findings 2026 CCF-A
[15]
Revisiting the Effective Number Theory for Imbalanced Learning
Ou Wu, Mengyang Li
IEEE TKDE 2024 CCF-A · IF 8.9
[16]
Unequal Vulnerability: The Differential Impact of Label Flipping Attacks Across Classes
Pinlong Zhao, Mengyang Li, Pengfei Jiao, Huijun Tang, Ou Wu
WWW 2026 CCF-A
[17]
Investigating the Sample Weighting Mechanism Using an Interpretable Weighting Framework
Xiaolin Zhou, Ou Wu, Mengyang Li
IEEE TKDE 2023 CCF-A · IF 8.9
[18]
Two-Level LSTM for Sentiment Analysis With Lexicon Embedding and Polar Flipping
Ou Wu, Tao Yang, Mengyang Li, Ming Li
IEEE Trans. Cybernetics 2022 CAS Q1 TOP · IF 11.8
[19]
Investigating Annotation Noise for Named Entity Recognition
Yu Zhu, Yingchun Ye, Mengyang Li, Ji Zhang, Ou Wu
2023 · IF 6.0
04

Patents专利

📋
A Method and Apparatus for Layout Parsing of Receipt-type Images 一种票据类图像版面解析方法及装置 Granted发明授权
📋
A Hard Sample Mining Method and Apparatus for Content Safety 一种面向内容安全的困难样本挖掘方法及装置 Granted发明授权
📄
An Interpretable Deep Learning Sample Weighting Method 一种可解释性的深度学习样本赋权方法 Published发明公开
📄
A Subject Information Extraction Method for Financial Statement Images 一种面向财务报表图像的科目信息提取方法 Published发明公开
📄
A Structured Information Extraction Method for Financial Statement Images 一种面向财务报表图像的结构化信息提取方法 Published发明公开
05

Graduate Recruitment研究生招生

🎓 Accepting Master's Students for 2026🎓 2026级硕士研究生招生

I am currently planning to recruit 2 master's students. Research topics center on data optimization, large language models, and embodied intelligence.

I welcome applicants who are diligent and rigorous, with a solid mathematical foundation and programming skills (Python preferred). If interested, please contact me by email with the subject line: "Master Application + Name + Undergraduate School/Major".

目前计划招收 2名硕士研究生，研究方向围绕数据优化、大语言模型、具身智能展开。

欢迎踏实认真、具备一定数学基础和编程能力（Python优先）的同学报考。有意者请发送邮件，邮件主题注明："硕士报考+姓名+本科院校/专业"。

limengyang@tjnu.edu.cn
06

Contact联系方式

📧
Email邮箱
limengyang@tjnu.edu.cn
📍
Address地址
Room C424, Mingli Building, Tianjin Normal University, 393 Binshui West Rd, Xiqing, Tianjin
天津师范大学明理楼C区424，天津市西青区宾水西道393号
🎓
Google Scholar
SJYlWVgAAAAJ
🔬
ORCID
0000-0002-8958-3163