Hi, I'm Yikai Zhang👋

Visitors

I’m a statistical Ph.D. candidate with 10+ years of experience in statistics and machine learning:

Developed innovative algorithms, including the Finite Smoothing Algorithm and Generalized Takeuchi’s Information Criteria, transforming SVM and large-margin classifiers.
Proficient in Python, R, SQL, and Fortran, delivering high-performance tools and open-source packages; integrate Fortran/CUDA kernels.
Published author with research showcased at ICML.
Successfully collaborated across diverse industries, including insurance and chemical manufacturing, to deliver impactful tools.

Now I’m open to work!

Data Science Intern — UFG Insurance (Summer 2024)

Refitted a bodily injury (BI) cost model in commercial auto (CA) insurance using XGBoost; leveraged large language models (LLMs) to extract, validate, and interpret multi-source data, improving prediction accuracy by 15% and significantly enhancing model robustness.
Built a Python Shiny tool integrating SQL and LLMs to estimate insurable replacement value.
Automated fraud detection from police reports with an LLM-based backend which significantly boosts claim flagging efficiency.

Data Science Intern — Dow Inc. (Summer 2023)

Developed a DOE simulation app (R + Shiny) that improved design efficiency by 50% and computation speed by 35%.
Enhanced usability for statisticians and engineers, achieving 90%+ user satisfaction.

Graduate Researcher — University of Iowa (2019–Present)

Research on large-scale kernel SVMs, kernel logistics regression, GPU acceleration, and insurance risk modeling.
Built open-source packages (TorchSVM, hdsvm, SAFE, GTIC) in PyTorch and R.
Integrated Fortran/CUDA kernels for high-performance computing.

Developed TorchSVM, a PyTorch-based library that efficiently trains kernel SVMs and computes exact leave-one-out cross-validation (LOOCV) errors with GPU acceleration.
Achieved scalable and fast computation, reducing the cost of LOOCV to that of training a single SVM, making large-scale kernel classification practical.
Benchmarked superior performance against existing solvers (e.g., LibSVM, scikit-learn), with significant improvements in speed and scalability.
Repo: torchsvm
Docs: torchsvm

Introduced a finite smoothing algorithm (FSA), a novel approach to tackle computational challenges in applying support vector machines (SVM) and quantile regression to high-dimensional data.
Implemented FSA using the coordinate descent method and demonstrated that FSA significantly outpaces its competitors in speed, often by orders of magnitude, while maintaining or improving precision.
Developed twoopen-source R packages: hdsvm for high-dimensional SVM and hdqr for high-dimensional quantile regression.
Repo: FSA
Paper: FSA

More projects → All repositories

Tang, Q.∗, Zhang, Y.∗, and Wang, B. Finite Smoothing Algorithm for High-Dimensional Support Vector Machines and Quantile Regression. International Conference on Machine Learning (ICML), 2024.
Zhang,Y.,Jia, G., andWang, B.TorchKSVM:EfficientKernelLarge-MarginClassifierswithExactCross Validation Error Computation in PyTorch. To be submitted.

GPU-accelerated kernel SVM with exact LOOCV and scalable model selection.

pip install torchsvm

CRAN Downloads

Implements an efficient algorithm for fitting the entire regularization path of support vector machine models with elastic-net penalties using a generalized coordinate descent scheme. The framework also supports SCAD and MCP penalties. It is designed for high-dimensional datasets and emphasizes numerical accuracy and computational efficiency.
Docs: hdsvm

install.packages("hdsvm")

Prospectus: On Advanced and Modern Computational Frameworks for Support Vector Machines. University of Iowa.

I’m open to research collabs, industry partnerships (ML/Stats/Insurance). Email: yikai-zhang@uiowa.edu