A few ML projects from labs and industry.
Hard nonconvex optimization with neural nets
Difference-of-convex neural optimization
Models that use DC programming inside the network, with custom PyTorch autograd and better anchor selection.
Training setup works and early results look good. Working toward a conference submission.
See related experience
Cell segmentation and tracking in noisy video
Microscopy and microplastics
Cellpose and TrackMate pipeline, fine-tuned on our labels, plus Python tools for cytotoxicity analysis.
False-positive microplastic detections dropped by up to 21%. OUR Distinction List and an AAAR submission.
See related experience
Demand forecasts by product line
Product demand forecasting
XGBoost and Random Forest models on 15 product lines, plus Plotly and Dash dashboards.
About 90% forecast accuracy. Saved roughly five hours of reporting each week.
See related experienceSupported by Purdue OUR and SURF. Email me if you want to work together.