Qitian Wu  

		I am a PhD student at Department of Computer Science and Engineering from Shanghai Jiao Tong University (SJTU), advised by Prof. Junchi Yan. I achieved the Bachelor (Microelectronics, Mathematics) and Master (Computer Science) degrees from SJTU. My general research interests lie in out-of-distribution generalization, structured learning and inference, and generative models, with their applications in recommender systems, drug discovery and advertisement. My current research focus on building scalable/expressive structured representation models and generalizable/reliable learning algorithms over semi-structured data (graphs, sequences, user behaviors, etc.). I endeavor on developing the next generation of learning approaches for dynamic, open and uncertain environments.

Recent News

• [2022.11] Two papers were selected as Spotlight Presentation by NeurIPS'22.
• [2022.10] I was awarded with National Scholarship.
• [2022.09] Papers on out-of-distribution generalization/detection, graph structure learning, temporal causal learning and molecular representation were accepted to NeurIPS'22.
• [2022.05] Papers on weakly-supervised graph neural networks and adversarial robustness were accepted to SIGKDD'22.
• [2022.04] One paper on negative sampling in recsys was accepted to IJCAI'22.
• [2022.01] One paper on out-of-distritbuion generalization on graphs was accepted to ICLR'22.
• [2021.11] I was awarded with Baidu PhD Scholarship.
• [2021.10] I was awarded with Microsoft Research PhD Fellowship.
• [2021.09] Papers on feature extrapolation, self-supervised GNNs and deep generative models were accepted to NeurIPS'21.
• [2021.08] One paper on sequential behavior modeling was accepted to CIKM'21 as spotlight.
• [2021.04] I was elected as Global Top 100 AI Rising Start by Baidu.

Selected Publications (see a complete list in Google Scholar)

• NodeFormer: A Scalable Graph Structure Learning Transformer for Node Classification
  Qitian Wu, Wentao Zhao, Zenan Li, David Wipf and Junchi Yan
  Advances in Neural Information Processing Systems (NeurIPS 2022, Spotlight, top 5%)
  
  • Summary: We propose a scalable graph structure learning model with efficient all-pair message passing on latent topology, empowering a pioneering Transformer-style model for large node classification graphs with linear complexity.
  • Materials: Slides | Codes
• Handling Distribution Shifts on Graphs: An Invariance Perspective
  Qitian Wu, Hengrui Zhang, Junchi Yan and David Wipf
  International Conference on Learning Representations (ICLR 2022)
  
  • Summary: We formulate out-of-distribution generalization problem on graphs and propose a provably effective invariant learning approach. As a piorneering work, we systematically discuss how to leverage invariance principle for handling node-level distribution shifts on graphs.
  • Materials: Slides | Poster | Codes
• Towards Open-World Feature Extrapolation: An Inductive Graph Learning Approach
  Qitian Wu, Chenxiao Yang and Junchi Yan
  Advances in Neural Information Processing Systems (NeurIPS 2021)
  
  • Summary: We study a new problem, i.e., feature space expansion from training data to testing data, and propose a graph-enhanced representation learning approach for handling tabular data with variable-length features.
  • Materials: Slides | Poster | Codes
• Towards Open-World Recommendation: An Inductive Model-based Collaborative Filtering Approach
  Qitian Wu, Hengrui Zhang, Xiaofeng Gao, Junchi Yan and Hongyuan Zha
  In International Conference on Machine Learning (ICML 2021)
  
  • Summary: We consider an inductive collaborative filtering setting where testing users are unappeared in the training phase, and propose to leverage representations of existing users to compute the new ones' through latent graph structures.
  • Materials: Slides | Poster | Codes
• Learning Substructure Invariance for Out-of-Distribution Molecular Representations
  Nianzu Yang, Kaipeng Zeng, Qitian Wu, Xiaosong Jia and Junchi Yan
  Advances in Neural Information Processing Systems (NeurIPS 2022, Spotlight, top 5%)
  
  • Summary: We build a substructure-aware invariant learning approach for out-of-distribution generalization in molecular representation. The new method outperforms the strongest baselines on OGB molecular property prediction and DrugOOD benchmarks by 5.9% and 3.9% in terms of ROC-AUC.
  • Materials: Codes
• GraphDE: A Generative Framework for Debiased Learning and Out-of-Distribution Detection on Graphs
  Zenan Li, Qitian Wu, Fan Nie and Junchi Yan
  Advances in Neural Information Processing Systems (NeurIPS 2022)
  
  • Keywords: Graph Representation Learning, Out-of-Distribution Detection, Debiased Learning
  • Summary: We introduce a probabilistic generative framework for out-of-distribution detection and debiased learning on graph data. The key insight is that the two problems can be unified under a single framework.
  • Materials: Codes
• Towards Out-of-Distribution Sequential Event Prediction: A Causal Treatment
  Chenxiao Yang, Qitian Wu, Qingsong Wen, Zhiqiang Zhou, Liang Sun and Junchi Yan
  Advances in Neural Information Processing Systems (NeurIPS 2022)
  
  • Summary: We use a causal perspective to analyze sequential event prediction under temporal distribution shifts and propose variational context adjustment as an effective treatment.
  • Materials: Coming soon
• Geometric Knowledge Distillation: Topology Compression for Graph Neural Networks
  Chenxiao Yang, Qitian Wu and Junchi Yan
  Advances in Neural Information Processing Systems (NeurIPS 2022)
  
  • Summary: We propose a geometric knowledge distillation model that can generalize the topological knowledge from larger GNNs to smaller one. The key insight is to capture the underlying geometric diffusion process within GNNs that are invariant to graph sizes.
  • Materials: Coming soon
• Variational Inference for Training Graph Neural Networks in Low-Data Regime through Joint Structure-Label Estimation
  Danning Lao, Xinyu Yang, Qitian Wu, Junchi Yan
  ACM SIGKDD Conference on Knowledge Discovery and Data Mining (SIGKDD 2022)
  
  • Summary: We propose a generative framework for training graph neural networks with incompletely observed data where labels and structures are jointly missing. The new approach outperforms GCN by 75.6% when only 1 label per class and 1% edges are observed on Cora.
  • Materials: Codes
• DICE: Domain-attack Invariant Causal Learning for Improved Data Privacy Protection and Adversarial Robustness
  Qibing Ren, Yiting Chen, Yichuan Mo, Qitian Wu, Junchi Yan
  ACM SIGKDD Conference on Knowledge Discovery and Data Mining (SIGKDD 2022)
  
  • Summary: We adopt a data-generative causal perspective to unify the formulation of privacy protection and adversarial robustness from a bottom-up perspective. Based on this, a new principled approach is proposed for performance boosting.
  • Materials: Codes
• Bridging Explicit and Implicit Deep Generative Models via Neural Stein Estimators
  Qitian Wu, Han Gao and Hongyuan Zha
  Advances in Neural Information Processing Systems (NeurIPS 2021)
  
  • Summary: We leverage the Stein's distance as a novel tool to construct a theoretically grounded bridging term between two families of deep generative models.
  • Materials: Slides | Poster | Codes
• From Canonical Correlation Analysis to Self-supervised Graph Neural Networks
  Hengrui Zhang, Qitian Wu, Junchi Yan, David Wipf and Philip S. Yu
  Advances in Neural Information Processing Systems (NeurIPS 2021)
  
  • Summary: We introduce a simple yet effective contrastive objective for self-supervised node representation learning over graphs. Compared with other works, our approach requires none of the parameterized mutual information estimator, additional projector, asymmetric structures, and most importantly, negative samples which can be costly.
  • Materials: Codes
• Learning Latent Process from High-Dimensional Event Sequences via Efficient Sampling
  Qitian Wu, Zixuan Zhang, Xiaofeng Gao, Junchi Yan and Guihai Chen
  In Advances in Neural Information Processing Systems (NeurIPS 2019)
  
  • Summary: We target the modeling and generation of temporal dynamics over a large interaction latent graph, and propose a scalable and efficient sampling approach for extremely high-dimensional sequences.
  • Materials: Poster | Codes
• Dual Sequential Prediction Models Linking Sequential Recommendation and Information Dissemination
  Qitian Wu, Yirui Gao, Xiaofeng Gao, Paul Weng and Guihai Chen
  In ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD 2019)
  
  • Summary: We propose a dual learning model that resorts to mutual reinforcement between two sequential learning models via a teacher-student distillation loss.
  • Materials: Poster | Codes
• Dual Graph Attention Networks for Deep Latent Representation of Multifaceted Social Effects in Recommender Systems
  Qitian Wu, Hengrui Zhang, Xiaofeng Gao, Peng He, Paul Weng, Han Gao and Guihai Chen
  In World Wide Web Conference (WWW 2019)
  
  • Summary: We design a dual graph attention architecture that simultaneously learns to capture user-user relations and item-item relations in a fully data-driven manner.
  • Materials: Slides | Poster | Codes

Selected Awards

• National PhD Scholarship
• Baidu PhD Fellowship (only 10 from worldwide PhD students)
• Microsoft Research PhD Fellowship (only 11 in Asia)
• Baidu Global Top 100 Rising Stars in Artificial Intelligence
• National Scholarship (2016, 2017, top 1 in the department)
• The 1st-Class Academic Excellence Scholarship (2016, 2017, top 1 in the department)
• Lixin Tang Scholarship (2017, 2018)
• Yuanqing Yang Scholarship (2019)
• Outstanding Winner, INFORMS Awards, Mathematical Contest in Modeling, Data Insights Problem (top 3 out of 4748 teams, the INFORMS Awards selects one team among all the participants) [paper]
  • Summary: We build an interpretable gaussian process regression model for energy consumption prediction and a principled multi-objective optimization model for social policy improvement.
• National Second Award, China Undergraduate Mathematical Contest in Modeling (top 5.8% in 28046 teams)
• First Award, Physics Competition of Chinese College Students (2015)
• Outstanding Graduate of Shanghai (2018)
• Excellent Thesis of Undergraduates (2018)

Academic Service

I will serve or served as the reviewer or (senior) PC member for TKDE, TNNLS, AAAI'21, CVPR'21, IJCAI'21, ICML'21, ICCV'21, NeurIPS'21, ICLR'22, AAAI'22, CVPR'22, IJCAI'22, ICML'22, NeurIPS'22, AAAI'23, ICLR'23, LoG'23.