Timeline for Why Gradient methods work in finding the parameters in Neural Networks?
Current License: CC BY-SA 4.0
3 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 5, 2018 at 17:44 | comment | added | induction601 | However, I am not sure how the node-switching invariant can be interpreted as finding a very good minimizer. I mean, how do we meausre "goodness" of minimizer, unless it can be quantified by $\|\theta^* - \theta_{local}\| \le \delta$ or $\|\mathcal{L}(\theta^*) - \mathcal{L}(\theta_{local})\| \le \epsilon$? | |
| Oct 5, 2018 at 17:42 | comment | added | induction601 | Thanks for your answer. I am aware that the gradient method can find the global minimizer if the cost function is convex and its gradient satisfies a certain condition, e.g., Lipschitz continuous or/and uniformly bounded gradient. Also, I am aware that the gradient method eventually (if it converges) find a local minimum under the certain conditions. | |
| Oct 5, 2018 at 16:43 | history | answered | Adrian Keister | CC BY-SA 4.0 |