Changheng Wang
Research Assistant at Association Analysis of Rare Variants and Diseases using UK Biological Databases
New Haven, Connecticut, United States
Summary
🤩
Rockstar🎓
Top SchoolChangheng Wang is a research-focused ML engineer with 8 years of hands-on experience building scalable data pipelines and model optimization tools across academia and industry. Based in New Haven, he automated high-throughput genomic and phenotypic analyses on Yale's HPC (SLURM) to process 50K+ samples and 17M+ loci, and has applied Transformer-based architectures to boost medical image reconstruction performance. His open-source contributions center on model compression and inference acceleration—notably enhancing Intel Neural Compressor integrations and quantization support in Hugging Face’s optimum-intel project. He also brings quantitative finance experience, engineering factor pipelines and LightGBM ensembles that drove robust backtest results and improved transaction-cost estimators. Comfortable across Python, R, Linux, and SLURM, he blends rigorous biostatistics training (Yale MPH) with practical ML engineering to turn complex datasets into reproducible, production-ready workflows. A less obvious strength is his track record of improving both low-level inference performance (bf16/IPEX static quantization) and end-to-end research productivity through automation.
8 years of coding experience
1 year of employment as a software developer
Master, Public Health in Biostatistics, Master, Public Health in Biostatistics at Yale UniversityBachelor of Science, Biotechnology, Bachelor of Science, Biotechnology at Shanghai Jiao Tong University
Github Skills (15)
transformers10
auto-tuning10
tokenize10
quantization10
pytorch10
machine-learning10
tokenizer10
large-language-models10
python10
testing10
question-answering9
inference8
nlp8
optimization8
onnx7
Programming languages (3)
Github contributions (5)
intel/neural-compressor
Jun 2021 - Jan 2023
SOTA low-bit LLM quantization (INT8/FP8/INT4/FP4/NF4) & sparsity; leading model compression techniques on TensorFlow, PyTorch, and ONNX Runtime
Role in this project:
ML Engineer & Test Automation Engineer
ML Engineer & Test Automation EngineerContributions:85 reviews, 125 commits, 77 PRs in 1 year 6 months
Contributions summary:Changheng implemented tokenization and evaluation utilities for a question-answering task, including a result checker, by modifying test files. The user also integrated SigOpt for hyperparameter optimization, indicating a focus on model training and optimization. Furthermore, the user contributed to fixing data loading issues within the dataset, demonstrating experience in data preprocessing and model training processes within the context of a low-bit LLM quantization project.
knowledge-distillationauto-tuningcompressorsparsityintel
huggingface/optimum-intel
Dec 2022 - Jan 2023
🤗 Optimum Intel: Accelerate inference with Intel optimization tools
Role in this project:ML Engineer
ML EngineerContributions:25 reviews, 6 commits, 16 PRs in 1 month
Contributions summary:Changheng primarily contributed to the Intel Neural Compressor (INC) integration within the `optimum-intel` repository, focusing on enabling and improving quantization techniques for PyTorch models, particularly using the IPEX backend. Their commits involved adding examples for question-answering tasks, incorporating bf16 support, and fixing issues related to arguments and model loading. The user also worked on tests to ensure the correct functionality of IPEX static quantization.
pytorchtransformersdeep-learningpruningoptimization
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.