5,337
edits
Deep Learning: Difference between revisions
→Convex-concave min-max
| Line 1,206: | Line 1,206: | ||
* f-> convex in x/concave in y | * f-> convex in x/concave in y | ||
* S-> convex/compact/bounded | * S-> convex/compact/bounded | ||
* <math>\eta \approx \frac{1}{\sqrt{T}</math> | * <math>\eta \approx \frac{1}{\sqrt{T}}</math> | ||
Then for Sim GDA: <math>f(\bar{x}, \bar{y}) \to f(x^*, y^*)</math> with order <math>O(1/\sqrt{T})</math> | Then for Sim GDA: <math>f(\bar{x}, \bar{y}) \to f(x^*, y^*)</math> with order <math>O(1/\sqrt{T})</math> | ||
No guarantees exist for the last iteration. | No guarantees exist for the last iteration. | ||