Jul 7, 2018 · We analyze the behavior of mirror descent (MD) in a class of non-monotone problems whose solutions coincide with those of a naturally associated variational ...
Dec 20, 2018 · We show how the inclusion of an extra-gradient step in first-order GAN training methods can improve stability and lead to improved convergence results.
We then show that this deficiency is mitigated by optimism: by taking an “extra-gradient” step, optimistic mirror descent (OMD) converges in all coherent ...
This work analyzes the behavior of mirror descent in a class of non-monotone problems whose solutions coincide with those of a naturally associated ...
Oct 10, 2018 · I tried to implement the " Optimistic mirror descent in saddle-point problems: Going the extra(-gradient) mile" algorithm, from here, built on top of the Adam ...
We then show that this deficiency is mitigated by optimism: by taking an "extra-gradient" step, optimistic mirror descent (OMD) converges in all coherent ...
is mitigated by optimism: by taking an “extra-gradient” step, optimistic mirror ... The extragradient method for finding saddle points and other problems.
Jul 25, 2019 · Bibliographic details on Optimistic mirror descent in saddle-point problems: Going the extra (gradient) mile.
People also ask
What is saddle point problem of gradient descent?
Can gradient descent stuck at saddle point?
Jul 20, 2018 · To mitigate this deficiency, we add on an "extra-gradient" step which we show stabilizes MD methods by looking ahead and using a "future ...
Oct 27, 2020 · As these papers show, OMD works better than vanilla MD for solving saddle-point problems. These saddle-point problems arise in adversarial ...