Chao Zhang

I work in the intersection of data mining, machine learning, and natural language processing. Currently, my research focuses on developing more label-efficient, robust, and interpretable models that can better understand text data. Towards this goal, I and my awesome students are currently working on the following topics:

• Knowledge representation, extraction, and reasoning: I am interested in how can machines represent knowledge and use it for machine intelligence. As much of such knowledge is encoded in text, we study how to extract structured knowledge from text data, as well as how to integrate knowledge into machine learning models for text understanding.
  • STEAM: Self-Supervised Taxonomy Expansion with Mini-Paths, KDD 2020
  • BOND: Bert-Assisted Open-Domain Named Entity Recognition with Distant Supervision, KDD 2020
  • TaxoGen: Unsupervised Topic Taxonomy Construction by Adaptive Term Embedding and Clustering, KDD 2018
  • TrioVecEvent: Embedding-Based Online Local Event Detection in Geo-Tagged Tweet Streams, KDD 2017
  • GeoBurst: Real-Time Local Event Detection in Geo-Tagged Tweet Streams, SIGIR 2016
• Out-of-distribution generalization: Current deep learning models typically assume the training and test distributions are i.i.d., which is rarely the case in practice. As a result, deep learning models can fail miserably in open-world settings. We want machines to go beyond the training distribution: How can we quantify uncertainties for deep learning models and enable them say "I don't know" for out-of-distribution samples? How can we make them to generalize to different domains and make robust predictions? How can we make them fast adapt to tasks that were never seen before?
  • Calibrated Language Model Fine-Tuning for In- and Out-of-Distribution Data, EMNLP 2020
  • SDE-Net: Equipping Deep Neural Networks with Uncertainty Estimates, ICML 2020
  • RDeepSense: Reliable Deep Mobile Computing Models with Uncertainty Estimations, UbiComp 2018
• Learning from weak/noisy supervision: Training deep learning models often requires tons of labeled data for a specific task, which can be labor-intensive and costly. In many applications, weakly labeled data can be easily obtained from domain knowledge. Can we train deep learning models using easy-to-provide but noisy supervision signals? Our recent works marry pre-trained language models with self-training as a generic approach to learning from weak/noisy supervision.
  • Text Classification Using Label Names Only: A Language Model Self-Training Approach, EMNLP 2020
  • Denoising Multi-Source Weak Supervision for Neural Text Classification, EMNLP 2020
  • SeqMix: Augmenting Active Sequence Labeling via Sequence Mixup, EMNLP 2020
  • BOND: Bert-Assisted Open-Domain Named Entity Recognition with Distant Supervision, KDD 2020
  • Weakly-Supervised Neural Text Classification, CIKM 2018
• Composable and interpretable models: When we know the structure of a problem, can we inject such knowledge into deep learning to design more interpretable and effective models? Can we design compositional models where some modules are shared and reused across tasks?
  • State-Sharing Sparse Hidden Markov Models for Personalized Sequences, KDD 2019
  • GMove: Group-Level Mobility Modeling Using Geo-Tagged Social Media, KDD 2016

• We have 4 paper accepted by EMNLP'20, discussing model calibration, sequence labeling, and zero-shot text classification.
• We have 1 paper accepted by ICML'20. We introduce a new method for quantifying uncertainties for neural networks based on the connection between NN and linear systems.
• We have 4 papers accepted by KDD'20, discussing self-supervised taxonomy construction, low-resource NER, hierarchical topic mining, and tensor factorization.
• Congrats to my student Wendi Ren for winning the Marshall D. Williamson Fellowship!
• Honored to receive the 2020 Google Faculty Research Award!
• Two papers accepted by the Web Conference 2020.
• Honored to receive the ACM SIGKDD 2019 Dissertation Runner-up Award!
• Our monograph Multidimensional Mining of Massive Text Data is published by Morgan & Claypool!

• 2020 Amazon AWS Machine Learning Research Award
• 2020 Google Faculty Research Award
• 2019 ACM SIGKDD Dissertation Award Runner-up
• 2018 ACM IMWUT Distinguished Paper Award
• 2015 ECML/PKDD Best Student Paper Runner-up Award
• 2013 Chiang Chen Overseas Graduate Fellowship

(* denotes equal contribution)

• SDE-Net: Efficient uncertainty estimation for deep neural networks
• BOND: Distantly-supervised named entity recognition
• STEAM: Automatic taxonomy expansion
• TaxoGen: Unsupervised topic taxonomy construction from text corpus
• WestClass: Weakly-supervised text classification
• GeoBurst: Unsupervised spatiotemporal event detection from social media streams

• 2020 Fall: CSE8803-DLT: Deep Learning for Text Data
• 2020 Spring: CX4240: Introduction to Computational Data Analysis
• 2019 Fall: CSE8803-DLT: Deep Learning for Text Data
• 2019 Spring: CX4240: Introduction to Computational Data Analysis

• Ph.D. Students:
  • Lingkai Kong
  • Rui Feng
  • Yue Yu
  • Yinghao Li
• Master Students:
  • Rongzhi Zhang
  • Yuchen Zhuang
  • Isaac Rehg
  • Ruijia Wang
• Alumni:
  • Wendi Ren
  • Yi Rong