Codes
Hyperspectral Image Anomaly Detection using RPCA
Python implementaion of Robust Principal Component Analysis (RPCA)-based image anomaly detection algorithm described in Xu, Kong, and Asgharzadeh (2021). The algorithm is particularly designed for wildfire detection using GOES-16 imagery. The brightness temperature (BT) difference between short-wave and long-wave infrared bands is sensitive to active fire. Image matrix of this measurement can be decomposed into low-rank and sparse components, where the latter represent wildfire pixels. The sparse component is post-processed by cloud and noise filters to eliminate false positives.
Bayesian Nonparametric Quantile Process Regression
R implementaion of the QUINN model described in Xu and Reich (2021+). The conditional density function is expanded by a convex sum of monotonic spline basis functions, and the spline coefficients are modeled as feature-dependent functions using 3-layer neural network. The network weights are assigned prior distributions, and posterior distribution of the model is approximated by No-U-Turn Sampler (NUTS).
Bayesian Quantile Regression using Neural Network
A simple Python implementation of Bayesian quantile regression where the quantile is modeled by a 4-layer neural network, and the data is assumed to follow an asymmetric Laplace distribution. Posterior sampling is carried out by Hamilton Monte Carlo with Dual Averaging.