Deep Learning: Difference between revisions
| Line 123: | Line 123: | ||
}} | }} | ||
;Theorem 4. | ;Theorem 4.2 (Local PL* implies existence of a solution + fast convergence) | ||
Assume <math>L(w)</math> is <math display="inline">\beta</math>-smooth and satisfies <math display="inline">\mu</math>-PL condition around a ball <math>B(w_0, R)</math> with <math>R = \frac{2\sqrt{w\beta L(w_0)}{\mu}</math>. | Assume <math>L(w)</math> is <math display="inline">\beta</math>-smooth and satisfies <math display="inline">\mu</math>-PL condition around a ball <math>B(w_0, R)</math> with <math>R = \frac{2\sqrt{w\beta L(w_0)}{\mu}</math>. | ||
Then: | Then: | ||