Blog
Short essays on language model research, efficient AI systems, and the engineering choices behind them.
Learning from mistakes with critique-guided distillation
Teacher feedback can supervise refinement during training while the deployed student answers from the prompt alone. A visual guide to CGD and its evaluation limits.
Uncertainty Quantification (UQ) and Sensitivity Analysis: Part I
Uncertainty quantification asks how uncertain a prediction is, while sensitivity analysis identifies which uncertain inputs contribute most to that result.
Physics-Informed Machine Learning (PIML): Application to Additive Manufacturing
Physics-informed machine learning can make limited experimental data more useful by adding physical constraints, simulation outputs, or simulation-based pretraining.
Multi-objective Optimization Under Uncertainty
Real design problems balance conflicting goals, and the best deterministic trade-off may be fragile once input and model uncertainty are included.