Xitong Zhang

Senior Engineer

Qualcomm

Biography

I’m Xitong Zhang, currently a machine learning engineer at Qualcomm working on GenAI applications. I obtained my Ph.D. in Computational Mathematics at Michigan State University. I’m honored to be under the guidance of Dr. Wang, Rongrong and Dr. Hirn, Matthew. My academic journey has been deeply rooted in the study of machine learning and its wide-ranging applications.

Interests

GenAI
Computer Vision
Learning on Graphs
AI for Science
Generalization

Education

Ph.D. in Computational Mathematics, 2024

Michigan State University, United States
MSc in Computer Science, 2020

Michigan State University, United States
MSc in Computer Science, 2018

Worcester Polytechnic Institute, United States
BEng in Artificial Intelligence, 2016

Hunan University, China

Experience

Senior Machine Learning Engineer

Qualcomm

July 2024 – Present California, United States

GenAI, Diffusion Models, and Large Language Models.

Graduate Research/Teching Assistant

Michigan State University

August 2018 – June 2024 Michigan, United States

Machine learning Theory and the applications on Directed Graph Neural Networks, Spatiotemporal Data Mining, Generalization and Regularization.

Machine Learning Engineer Intern

Qualcomm

May 2023 – August 2023 California, United States

Supervised and Unsupervised Image Inpainting and Outpainting by Diffusion Models.

Graduate Research Assistant

Los Alamos National Lab

March 2022 – August 2022 New Mexico, United States

Generated large-scale datasets and Uncertainty Quantification for end-to-end Data-driven Full Waveform Inversion (FWI).

Graduate Research Assistant

Los Alamos National Lab

June 2021 – August 2021 New Mexico, United States

Physics-informed Machine Learning for Full Waveform Inversion (FWI).

Graduate Research Assistant

Los Alamos National Lab

June 2020 – August 2020 New Mexico, United States

GNN-based Earthquake Characterization; Squential Seismic Images Generation.

Graduate Research Assistant

Worcester Polytechnic Institute

October 2017 – May 2018 Massachusetts, United States

Image Segmentation on Live cell Images.

Projects

Inpainting and Outpainting with Diffusion Models

During the Qualcomm 2023 summer internship, we initially delved into unsupervised inpainting and outpainting using the RePaint framework. We enhanced this by introducing a smoothing guidance for improved sampling. To speed up the diffusion process, we then investigated both unsupervised and supervised Wavelet Diffusion Models.

OpenFWI Datasets

OpenFWI is a collection of large-scale, multi-structural benchmark datasets for machine learning driven seismic FWI. We release twelve datasets synthesized from different priors, including one 3D dataset. We also provide baseline experimental results with four deep learning methods, InversionNet, VelocityGAN, UPFWI and InversionNet3D. OpenFWI is the first open-source platform to facilitate data-driven FWI research. It will be actively developed and the datasets are expected to evolve.

PyTorch Geometric Signed Directed

PyTorch Geometric Signed Directed is a signed/directed graph neural network extension library for PyTorch Geometric. It builds on open-source deep-learning and graph processing libraries. PyTorch Geometric Signed Directed consists of various signed and directed geometric deep learning, embedding, and clustering methods from a variety of published research papers and selected preprints.

FeedGrain Visualization

The system is designed to visualize the US historical feed grain data and gain insights into the supply and the demand of feed grains in domestic market and international trade of the US with import and export partners.

Featured Publications

Xitong Zhang, Avrajit Ghosh, Guangliang Liu, Rongrong Wang

October, 2023

Unlocking Tuning-free Generalization: Minimizing the PAC-Bayes Bound with Trainable Priors

Our proposed tuning-free PAC-Bayes training framework achieves test performance comparable to that of SGD/Adam, even when the latter are optimized through a complete grid search and supplemented with additional regularization terms.

Xitong Zhang, Will Reichard-Flynn, Miao Zhang, Matthew Hirn, Youzuo Lin

October, 2022 JGR: Solid Earth

Spatiotemporal Graph Convolutional Networks for Earthquake Source Characterization

Earthquake characterization by graph neural networks.

Peng Jin, Xitong Zhang, Yinpeng Chen, Sharon Xiaolei Huang, Zicheng Liu, Youzuo Lin

April, 2022 ICLR

Unsupervised learning of full-waveform inversion: Connecting CNN and partial differential equation in a loop

This paper investigates unsupervised learning of Full-Waveform Inversion (FWI) by integrating PDE and CNN in a loop.

Xitong Zhang, Yixuan He, Nathan Brugnone, Michael Perlmutter, Matthew Hirn

December, 2021 NeurIPS 2021

Magnet: A neural network for directed graphs

We proposed a directed graph neural network based on a complex Hermitian matrix known as the Magnetic Laplacian.

Recent Publications

Quickly discover relevant content by filtering publications.

Avrajit Ghosh, Xitong Zhang, Kenneth K Sun, Qing Qu, Saiprasad Ravishankar, Rongrong Wang (2024). Optimal Eye Surgeon: Finding image priors through sparse generators at initialization. ICML 2024.

Guangliang Liu, Milad Afshari, Xitong Zhang, Zhiyu Xue, Avrajit Ghosh, Bidhan Bashyal, Rongrong Wang, Kristen Johnson (2024). Towards Understanding Task-agnostic Debiasing Through the Lenses of Intrinsic Bias and Forgetfulness. ACL 2024.

Xitong Zhang, Ismail R Alkhouri, Rongrong Wang (2024). Structure-Preserving Network Compression Via Low-Rank Induced Training Through Linear Layers Composition. arXiv.

Holly R Steach, Siddharth Viswanath, Yixuan He, Xitong Zhang, Natalia Ivanova, Matthew Hirn, Michael Perlmutter, Smita Krishnaswamy (2024). Inferring Metabolic States from Single Cell Transcriptomic Data via Geometric Deep Learning. RECOMB 2024.

Guangliang Liu, Zhiyu Xue, Xitong Zhang, Kristen Johnson, Rongrong Wang (2023). PAC-tuning: Fine-tuning Pre-trained Language Models with PAC-driven Perturbed Gradient Descent. EMNLP 2023.

See all publications