Skip to content
← All research

Current capstone · In progress

Sharpness-aware optimization.

I’m exploring sharpness-aware optimization for machine-learning training. This is the space for the capstone’s experiments and findings as the work develops.

Illustrative SGD and SAM solutions on an uneven loss curve: a sharp minimum versus a broad basin, with equal-size parameter neighborhoods.
A synthetic landscape—not a capstone result or a guarantee of either optimizer’s outcome. The shaded bands show equal-size parameter neighborhoods. Based on Foret et al., Sharpness-Aware Minimization (2021).

What I’m working on

My capstone focuses on sharpness-aware optimization, often referred to as SAM, in machine-learning training. The project is ongoing.

The basic idea is to look beyond the loss at one set of model parameters. SAM considers nearby parameter settings too, seeking a neighborhood where the loss stays low rather than a single low point surrounded by steep slopes.

Progress

I’m still working on the capstone and haven’t published results yet. The experimental setup and findings will be added here when they are ready to share.

← Back to the homepageGet in touch ↗