Assistant Professor, USC Computer Science
Research Lead, USC Information Sciences Institute
Director, USC INK Research Lab
Information Director, SIGKDD
Forbes' Asia 30 Under 30

xiangren [at] usc.edu
Machine Learning, Natural Language Processing


Curriculum Vitae  |  INK@GitHub  |  Google Scholar

I'm an Assistant Professor of Computer Science at USC, with affiliated appointment at USC Information Sciences Institute (ISI). At USC, I'm the director of the Intelligence and Knowledge Discovery (INK) Research Lab, and member of the USC Machine Learning Center, NLP Community@USC, and ISI Center on Knowledge Graphs. In 2018, I served as a part-time Data Science Advisor at Snapchat. Prior to USC, I was a visiting researcher at Stanford University and received my PhD in Computer Science from UIUC.

A summary of my previous work on label-efficient information extraction can be found in the book "Mining Structures of Factual Knowledge from Text: An Effort-Light Approach". They are also covered in several tutorials in major conferences. I have been serving as area chairs and workshop co-chairs in ACL, EMNLP, ICLR, KDD and AAAI, and received awards including ACM SIGKDD Dissertation Award, WWW Best Poster runner-up, David J. Kuck Outstanding Thesis Award, and Google PhD fellowship. My research is funded by NSF, DARPA, IARPA and receives faculty awards from industry partners including Google, Amazon, JP Morgan, Adobe and Snapchat. I was named Forbes' Asia 30 Under 30 in 2019.


Research  

Please check out the INK Lab website for more information.

I work on new algorithms and datasets in natural language processing and machine learning, with limited labeled data. My research group at USC focuses on developing label-efficient, prior-informed models that extract machine-actionable knowledge from natural language data and perform neural-symbolic knowledge reasoning for question answering. I'm particularly excited about problems in the space of modeling language data with weak supervision and prior knowledge. This includes neural-symbolic learning, learning from high-level supervision, learning with noisy data, and zero/few-shot learning.

Learning from distant, high-level human supervision. State-of-the-art neural models are still quite data hungry to build. Training these models towards a specific task may require hundreds of thousands of labeled examples. Going beyond the standard instance-label training data design, we are developing new training paradigms for building NLP systems, including translating high-level human supervisions into machine-executable, modularized programs (ICLR19, ACL19 Demo, TheWebConf18), and referencing pre-existing knowledge resources for automatic data annotation (EMNLP18, WSDM18, EMNLP17).

Learning with structural inductive biases. Deep neural networks have demonstrated strong capability in fitting large dataset towards mastering a task, but at the same time also showing poor generalization ability in terms of task/domain transferability. One main reason is because the common mechanisms shared across the tasks, such as model components and constraints, are not explicitly specified in the models (i.e., inductive biases). We are exploring various ways of designing structural inductive biases that are task-general and human-readable (EMNLP19a, ICLR19, EMNLP19b), and developing novel models and learning algorithms to impose such inductive biases (EMNLP19a, TheWebConf18).

- Blog posts: Information Extraction with Indirection Supervision and Heterogeneous Supervision, Dynamic Network Embedding.

- Acknowledgement: We are currently supported by DARPA awards on machine commonsense (MCS), learning for open-world novelty (SAIL-ON) and Systematizing Confidence in Open Research (SCORE), IARPA award on Better Extraction from Text Towards Enhanced Retrieval (BETTER), and NSF award on machine reading for understanding careers of scientific ideas. We are also supported by generous gifts from Schmidt Family Foundation and industry partners including Google, Amazon, JP Morgan, Adobe, Sony and Snapchat.


News  cv

Aug, 2020 - Will serve as Area Chair for AAAI 2021.
Jul, 2020 - Our ACL work on examining and reducing biases in hate speech detection algorithms is featured by Digital Trends, ScienceDaily, EurekAlert, Unite.AI, and USCViterbi.
Jul, 2020 - Will serve as area chair in ICLR 2021.
May, 2020 - INK lab members Ryan Moreno and Lily Cao won Undegraduate Outstanding Student Award from USC. Congratulations!
May, 2020 - Our paper "NERO: A Neural Rule Grounding Framework for Label Efficient Relation Extraction" received Best Paper Award Runner-up at The Web Conference 2020!
Apr, 2020 - INK Lab has four papers accepted at ACL 2020.
Apr, 2020 - Our LEAN-LIFE system for label-efficient, explanation-based annotation has been accepted to ACL 2020 demo track.
Mar, 2020 - Excited to receive a Sony Faculty Research Award to support our work on learning from natural language explanations.
Mar, 2020 - Will serve as area chair for ML on NLP in EMNLP 2020.
Feb, 2020 - Give invited talk about "Fast Learning with Explanation and Prior Knowledge" at CMU LTI colloquium and UT Austin.
Dec, 2019 - INK lab has two papers (spotlight & poster) accepted at ICLR 2020.
Nov, 2019 - Invited talk at CMU LTI Colloquium in Feb, 2020.
Sep, 2019 - Excited to receive a data science research award from Adobe Research to work on neural symbolic learning for recommendation.
Aug, 2019 - INK lab members have 10 papers accepted at EMNLP 2019. Congratulations!
June, 2019 - We're excited to receive a gift award from Snapchat to work on modular neural networks for interpretable NLP!
June, 2019 - We received a DARPA GAILA grant to work on building AI to mimic children language learning.
May, 2019 - Serve as area chair for EMNLP 2019, ACL 2019; as senior PC for AAAI 2020.
Mar, 2019 - Excited to receive a Google Faculty Award for supporting our research on explainable recommendation.
Mar, 2019 - Our research on interpretable knowledge reasoning is funded by JP Morgan AI Research Award.
Feb, 2019 - As part of the USC/ISI team, we received DARPA award to work on Machine Commonsense and Learning with Less Data.
Jan, 2019 - Our research on neural-symbolic deep learning for NLP is funded by an Amazon Research Award.
Dec, 2018 - Organizing the ICLR 2019 LLD Workshop on learning from limited labeled data.
Dec, 2018 - Organizing the RepL4NLP Workshop at ACL 2019 on representation Learning for NLP. We're soliciting submissions.
Nov, 2018 - Organizing the DeepLo Workshop at EMNLP 2019 on deep learning for low-resource NLP.