Yao Xie

Spatio-temporal process

Neural spectral marked point processes.
Shixiang Zhu, Haoyun Wang, Xiuyuan Cheng, Yao Xie. Neural Spectral Marked Point Processes.

Spatio-temporal point process simulator.
Shixiang Zhu, Shuang Li, Zhigang Peng, Yao Xie. Imitation Learning of Neural Spatio-Temporal Point Processes. IEEE Transactions on Knowledge and Data Engineering.

Conformal prediction interval

Conformal prediction interval for dynamic time-series.
Chen Xu and Yao Xie.
ICML 2021. (Long presentation)

Checking condition for deterministic matrix completion

An example for how to use ``well-posed'' condition.

This is in paper Matrix completion with deterministic pattern - a geometric perspective. A. Shapiro, Y. Xie, and R. Zhang. IEEE Transactions on Signal Processing.

M-statistics for change-point detection

Download Matlab code here.

M-statistic is a kernel-based statistic for detecting changes in streaming data. The statistic is computationally efficiently and has an false-alarm-rate that can be theoretically well approximated. Hence, the threshold for M-statistic can be chosen easily by evaluating a closed-form expression without requiring the usual onerous simulation. More details in our paper "M-statistic for kernel change-point detection" (a preliminary version appeared in NIPS 2015).

Approximate algorithms for Poisson matrix completion and recovery

Download Matlab code here.

Our PMVSVT (Penalized maximum likelihood singular value threshold) algorithm is tailored to solving maximum likelihood based low-rank matrix recovery, or matrix completion problems. PMLSVT is derived by expanding the likelihood function locally in each iteration, and finding an exact solution to the local approximation problem which results in a simple singular value thresholding procedure. More details in our paper "Poisson matrix recovery and completion", Yang Cao, and Yao Xie, to appear in IEEE Transactions on Signal Processing.

Theoretical false-alarm-rate for multi-sensor change-point detection.

Download Matlab code here.

We characterize theoretical ARL (average run length, a standard performance metric for sequential change-point detection) for the mixture multi-sensor change-point detection procedure. The code evaluate the ARL expression in our paper "Sequential multi-sensor change-point detection", Yao Xie and David Siegmund, Annals of Statistics, Vol. 41, No. 2, pp. 670-692, 2013.