韩 路 Lu Han (L. Han) M.Sc. Student, LAMDA Group Department of Computer Science and Technology National Key Laboratory for Novel Software Technology Nanjing University, Nanjing 210023, China Email: hanlu@lamda.nju.edu.cn, hanlu198004@163.com Laboratory: Shaoyifu Building, Xianlin Campus of Nanjing University

Currently I am a Ph.D. candidate of School of Artificial Intelligence in Nanjing University and a member of LAMDA Group( LAMDA Publications), led by professor Zhi-Hua Zhou.

Supervisor

Associate professor De-Chuan Zhan.

Biography

I received my B.Sc. degree from Wuhan University in June 2019. In the same year, I was admitted to study for a M.Sc. degree in Nanjing University without entrance examination.

I received my M.Sc. degree in School of Artificial Intelligence from Nanjing University in June 2022. In the same year, I was admitted to study for a Ph.D. degree in School of Artificial Intelligence from Nanjing University, which is also in the LAMDA Group led by Prof. Zhi-Hua Zhou., under the supervision of Prof. De-Chuan Zhan and Han-Jia Ye.

Research Interests

Lu mainly focuses on machine learning, especially:

Meta-Learning

• Meta-Learning, or learning to learn, aims at extracting meta-knowledge from previous tasks, and reuse them in new tasks. It can be applied to few-shot learning, federated learning, hyper-parameter setting, and other related areas.

Contrastive Learning

• Contrastive Learning methods maximize the agreement between positive pairs and minimize the agreement between negative pairs. It has been the one of the most popular methods in self-supervised learning, representation learning and is the fundamental technique of many pre-trained models like CLIP.

Semi-Supervised Learning

• Semi-supervised learning is a broad category of machine learning that uses labeled data to ground predictions, and unlabeled data to learn the shape of the larger data distribution. Practitioners can achieve strong results with fractions of the labeled data, and as a result, can save valuable time and money.

Publications

Lu Han, Han-Jia Ye, De-Chuan Zhan. Augmentation Component Analysis: Modeling Similarity via the Augmentation Overlaps. International Conference on Learning Representations (ICLR), 2023. [ArXiv] [OpenReview] (Top-1 Artificial Intelligence Publication) Traditional contrastive learning methods pull views of samples together and push different samples away, which utilizes semantic invariance of augmentation but ignores the relationship between samples. In this paper, we propose Augmentation Component Analysis (ACA) which theoretically preserves the similarity of augmentation distribution between samples and helps learn semantically meaningful embeddings.

Lu Han, Han-Jia Ye, De-Chuan Zhan. On Pseudo-Labeling for Class-Mismatch Semi-Supervised Learning. Transactions on Machine Learning Research (TMLR), 2022. [ArXiv] [OpenReview] In this paper, we empirically analyze Pseudo-Labeling (PL) in class-mismatched SSL and find the imbalance problem and show better strategies for pseudo-labeling. Base on these findings, we propose to improve PL in class-mismatched SSL with two components -- Re-balanced Pseudo-Labeling (RPL) and Semantic Exploration Clustering (SEC).

Han-Jia Ye, Lu Han, De-Chuan Zhan. Revisiting Unsupervised Meta-Learning via the Characteristics of Few-Shot Tasks. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2022. [ArXiv] [IEEE Xplore] (学生一作) Unsupervised meta-learning for few-shot classification without any base class labels. We propose strong unsupervised baselines which outperforms some supervised counterparts.

Selected Honors

• HUAWEI Fellowship for Outstanding Contribution, 2021
• First-Class Graduate student talent scholarship, 2022
• First-Class Academic Scholarship of Nanjing University for Ph.D., 2022
• First-Class Academic Scholarship of Nanjing University for M.Sc., 2019 2020 2021
• Citi Cup Innovation and Application Contest Top 5 in China, Chong'qin, 2017

Internship

• 2018.5~2018.9: Technology Engineering Group, Tencent (技术工程事业群，腾讯)
  Software Engineer
  Developing Deep Neural Network Compilers for FPGAs.

Professional Services

• Conference PC member/reviewer: ECAI 2020, IJCAI-PRICAI 2020, CVPR 2021, KDD 2021, CCML 2021, CVPR 2022, PAKDD 2022, UAI 2022, ECCV 2022, KDD 2022, CVPR 2023, SDM 2023, AISTATS 2023
• Journal reviewer: IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI), IEEE Transactions on Neural Networks and Learning Systems (IEEE TNNLS), IEEE Transactions on Intelligent Systems and Technology (TIST), IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), SCIENTIA SINICA Informationis, Chinese Journal of Electronics (CJE)
• Student volunteer: MLA 2019, MLA 2020

Teaching Assistant

• 智能系统设计与应用(詹德川教授;2022年春季)

• 时间序列分析(叶翰嘉讲师;2022年秋季)

Correspondence

Email: hanlu@lamda.nju.edu.cn, hanlu198004@163.com

Office: Room A201, Shaoyifu Building, Xianlin Campus of Nanjing University

Address: Lu Han, National Key Laboratory for Novel Software Technology, Nanjing University, Xianlin Campus Mailbox 603, 163 Xianlin Avenue, Qixia District, Nanjing 210023, China
(南京市栖霞区仙林大道163号, 南京大学仙林校区603信箱, 软件新技术国家重点实验室, 210023.)