Chaoshun Hu

About Chaoshun Hu

Hi, I’m Chaoshun. My recently experience includes using DeepSpeed-Chat and LLaMa Factory for LLM finetuning (with or without Lora) and alignment with RLHF PPO, DPO, ORPO; using Lanchain for Prompt Engineering and RAG through different projects, Also able to use transformer based vision large model for geophysical signal restoration, and GAI GAN/Diffusion for resolution enhancement and inverse problem. I have an ongoing research collaboration project about using offline policy for language model alignment.I also completed a full stack webapp deployment with React and Django which enables the Q/A with language model and also the deep learning model for damage detection.Before August, 2022,I was a Staff Data Scientist who loves building AI solutions for IT and energy especially oil & gas industry. I had a data scientist master degree and also Geophysics PhD degree. I have created major AI projects in Chevron including Deep learning as deep preconditioned for LSRTM and FWI, ML based petrophysics and unconventional reservoir characterization, geosteering using Gaussian processes path planning and Least squares Gaussian beam real-time imaging and Deep reinforcement learning for oil field development optimization. I also pioneered cloud computing in Chevron for seismic imaging and data analysis based on AWS and Azure Platform using docker, container and kubernetes and distributed computing on Ray platform. I’d love to combine my passion for learning and teaching with my data science skills and domain knowledge from physics to continue building personalized learning products for business needs.Specialties: Machine Learning and Deep Learning( scikit, CNN/RNN/MTL, keras,tensorflow, prtorch, dask, xarray)+ Reinforcement Learning(one of two partner with Ray Cluster on Azure, DQN/PPO), Gaussian processes path planning, LLM and GAI, DeepSpeed-Chat, LLaMa Factory, SFT+RW+RLHF, Diffusion model for inverse problemCloud Computing( delivered product in AWS and Azure platform, also completed some SMU class project in GCP) and cloud databaseComputational geophysics (HPC);Seismic imaging with GBmig, RTM, Elastic RTM;Salt inversion using levelset FWI;Interactive Imaging and Velocity Analysis

Experience

1. Staff Geoscience Data Scientist

   Chevron

   Feb 2017 — Present · Houston, TX, US

   My current work is focused on the R&D of the combination of machine learning algorithm and data science skills to help geophysics and petrophysics to extract more accurate and more useful subsurface information. I invented a facies constraint petrophysics attribute machine learning estimation algorithm, a multitask CNN algorithm for subsurface data and also recently a new well logging depth series data alignment using a new dynamic time warping algorithm; I implemented a meta learning based FWI and Deep reinforcement learning based drilling automation and collaborated with CVX RPE and Microsoft in Deep reinforcement learning for oilfield development optimization, I also used python to develop machine learning or deep learning algorithm in seismic imaging and data analysis on private and public cloud like Azure and AWS.

Education

• The University of Texas at Austin

  PhD, Geophysics

  2003 — 2008

• Southern Methodist University

  Master's degree, Data Science

  2018 — 2020