Anqi Liu

This page is not updated anymore. Here is my new site.

I am a DOLCIT postdoctoral fellow in Department of Computing and Mathematical Sciences (CMS) at California Institute of Technology (Caltech), working with Prof. Yisong Yue and Prof. Anima Anandkumar.

I received my Ph.D. in Department of Computer Science, University of Illinois at Chicago. I was fortunate to have Prof. Brian Ziebart as my advisor.

My research interest lies in machine learning and its applications. I am currently working on robust learning under covariate shift, interactive machine learning.

News

Nov. 2017 -Our paper will be presented in NIPS Workshop 2017: Aligned AI and WIML Workshop 2017.

Nov. 2016 -Our paper will be presented in NIPS Workshop 2016: Reliable Machine Learning in the Wild.

Aug. 2016 -Our paper got accepted in NIPS 2016.

Dec. 2015 -Our paper got accepted in AISTATS 2016.

Nov. 2015 -I will present my thesis proposal in the 21st AAAI/SIGART Doctoral Consortium at AAAI-16 in Phoenix.

Sep. 2015 -Our paper got accepted in the International Conference on the Theory of Information Retrieval.

July. 2015 -Workshop paper got presented at ICML Active Learning Workshop 2015, lille France.

Research

I am generally interested in machine learning theory and applications. I currently focus on covariate shift settings and investigate robust classification method from different perspectives.

Recent projects are related with active learning, model selection, kernel methods, statistical learning theory and information retrieval. Below is a list of past projects.

Covariate Shift: Robust Classification/
Regression/Structure Prediction

In many important machine learning applications, the source distribution used to estimate a probabilistic classifier differs from the target distribution on which the classifier will be used to make predictions. We develop a framework for learning robust bias-aware (RBA) probabilistic models that adapts to sample selection biases using a minimax estimation formulation.

Shift Pessimistic Active Learning

Most active learning techniques assume that unlabeled datapoints’ labels are distributed according to the model learned from labeled data. This assumption is usually very strong and will lead to inefficient label solicitation. We view active learning as a special case of covariate shift and develop methods and label solicitation strategies that are pessimistic of the inherent data shift and therefore, robust to the bias.

Predictive IOC with Linear Quadratic Regulator model

To facilitate interaction with people, robots must infer a person’s intentions and future behavior. We address two challenges, high dimensionality and uncertainty in this problem, by employing predictive inverse optimal control methods to estimate a probabilistic model of human motion trajectories. Our inverse optimal control formulation estimates quadratic cost functions that best rationalize observed trajectories framed as solutions to linear-quadratic regularization problems.

Publications

Preprints

Anqi Liu and Brian D. Ziebart “Robust Covariate Shift Classification with General Losses and Feature Views ” In Submission, 2017.

Anqi Liu, Rizal Fathony, Brian D. Ziebart “Kernel Robust Bias-Aware Prediction under Covariate Shift ” Manuscript, 2016.

Conference Paper

Rizal Fathony, Anqi Liu, Kaiser Asif, Brian D. Ziebart “Adversarial Multiclass Classification: A Risk Minimization Perspective ” In Advances in Neural Information Processing Systems 29 (NIPS), 2016.

Xiangli Chen, Mathew Monfort, Anqi Liu, Brian D. Ziebart “Robust Covariate Shift Regression ” In the Proceedings of the 19th International Conference of Artificial Intelligence and Statistics (AISTATS),2016.

Hong Wang, Anqi Liu, Jing Wang, Brian D. Ziebart, Clement T. Yu, Warren Shen “Context Retrieval for Web Tables,” In the Proceedings of ACM SIGIR International Conference on the Theory of Information Retrieval, 2015.

Mathew Monfort, Anqi Liu and Brian D. Ziebart “Trajectory Forecasting and Intent Recognition via Predictive Inverse Linear-Quadratic Regulation,” In the Proceedings of the 29th AAAI Conference on Artificial Intelligence (AAAI), 2015. Accept Rate: 26.67%.

Anqi Liu, Lev Reyzin and Brian D. Ziebart “Pessimistic Active Learning using Robust Bias-Aware Prediction,” In the Proceedings of the 29th AAAI Conference on Artificial Intelligence (AAAI), 2015. Accept Rate: 26.67%. [PDF]

Anqi Liu and Brian D. Ziebart “Robust Classification under Sample Selection Bias,” In Advances in Neural Information Processing Systems (NIPS), 2014. Spotlight. [PDF]

Workshop Paper

Anqi Liu and Brian D. Ziebart “Robust Covairate Shift Classification with Exact Loss Functions,” In NIPS Workshop: Aligned AI, 2017. Contributed Talk. [PDF]

Anqi Liu, Hong Wang, Brian D. Ziebart “Robust Covairate Shift Classification using Multiple Feature Views,” In NIPS Workshop: Reliable Machine Learning in the Wild, 2016. Contributed Talk. [PDF]

Anqi Liu, Kaiser Asif, Wei Xing, Sima Behpour, Brian Ziebart, Lev Reyzin “Addressing Covariate Shift in Active Learning with Adversarial Prediction,” In ICML Active Learning Workshop, 2015. Contributed Talk.

Mathew Monfort, Anqi Liu and Brian D. Ziebart “Trajectory Forecasting and Intent Recognition via Predictive Inverse Linear-Quadratic Regulation,” In IROS Workshop on Assistance and Service Robotics in a Human Environment, 2014.

Other Talks

Anqi LiuRobust Classification under Covariate Shift with Application to Active Learning, ” In Doctoral Consortium of the 30th AAAI Conference on Artificial Intelligence (AAAI),2016.