| |
|---|
| Omfang
| |
|---|
| Opplysninger
| Årstall på tittelblad: 2025. - Avhandling (Ph.D) - Universitetet i Oslo, 2026. - This thesis explores two key topics in high-dimensional optimization. The first part examines the implicit regularization of overparameterized models, demonstrating that gradient descent from small initializations converges to minimal solutions, such as sparse or low-rank solutions, without explicit regularization. The second part presents a practical method for constructing minimal convex environmental contours by discretizing infinite-dimensional problems into linear programs, achieving near-optimal solutions.
|
|---|
| Emner
| |
|---|
| Sjanger
| |
|---|
| Dewey
| |
|---|
