Ziteng Pang

ZitengPang2027@u.northwestern.edu

I’m a first-year Ph.D. student in the Statistics department at Northwestern University. Prior to this, I was a master’s student in the Applied Statistics department at the University of Michigan, Ann Arbor, where I was involved with two projects:

Machine olfaction (supervised by Professor Ambuj Tewari): similar to computer vision, but instead of trying to teach computers to see, it is about the smell with input space being molecules and outputting a subset of the possible target labels instead of just one element of the label set.
Bayesian Light Source Separator (supervised by Professor Jeffrey Regier): simultaneously identifying a large number of light sources in a crowded field and constructing a probabilistic astronomical cataloging based on telescope images using variational inference.

Besides the school and research work, I have found my interests in lots of things over time including photography, rock climbing (mainly bouldering roughly at a v5 level), motorsports(mostly karting but sometimes also track day). Because of the COVID and that my attention shifted a lot more towards academic work, I haven’t spent much time on my hobbies but hopefully I can find a nice balance later on.

news

Jul 1, 2022	Our paper “Scalable Bayesian Inference for Detection and Deblending in Astronomical Image” was accepted to ICML 2022 in Workshop Machine Learning for Astrophysics.

recent publications

ICML
Scalable Bayesian Inference for Detection and Deblending in Astronomical Images

Hansen, Derek, Mendoza, Ismael, Liu, Runjing, Pang, Ziteng, Zhao, Zhe, Avestruz, Camille, and Regier, Jeffrey

2022

Abs Bib HTML PDF Code

We present a new probabilistic method for detecting, deblending, and cataloging astronomical objects called the Bayesian Light Source Separator (BLISS). BLISS is based on deep generative models, which embed neural networks within a Bayesian model. For posterior inference, BLISS uses a new form of variational inference known as Forward Amortized Variational Inference (FAVI). FAVI has scaling advantages over Markov chain Monte Carlo and achieves improved fidelity of the posterior approximation compared with traditional variational inference in our application. The BLISS inference routine is fast, requiring a single forward pass of the encoder networks on a GPU once the encoder networks are trained. BLISS can perform fully Bayesian inference on megapixel images in seconds, and produces highly accurate catalogs than traditional methods do. BLISS is highly extensible, and has the potential to directly answer downstream scientific questions in addition to producing probabilistic catalogs.
doi = {10.48550/ARXIV.2207.05642}, journal = {Internation Conference on Machine Learning}, author = {Hansen, Derek and Mendoza, Ismael and Liu, Runjing and Pang, Ziteng and Zhao, Zhe and Avestruz, Camille and Regier, Jeffrey}, keywords = {Instrumentation and Methods for Astrophysics (astro-ph.IM), Applications (stat.AP), Machine Learning (stat.ML), FOS: Physical sciences, FOS: Physical sciences, FOS: Computer and information sciences, FOS: Computer and information sciences}, title = {Scalable Bayesian Inference for Detection and Deblending in Astronomical Images}, publisher = {arXiv}, year = {2022}, copyright = {arXiv.org perpetual, non-exclusive license}, }