Stochastic first-order methods such as Stochastic Extragradient (SEG) or Stochastic Gradient Descent-Ascent (SGDA) for solving smooth minimax problems and, more generally, variational inequality problems (VIP) have been gaining a lot of attention in recent years due to the growing popularity of adversarial formulations in machine learning. However, while high-probability convergence bounds are known to reflect the actual behavior of stochastic methods more accurately, most convergence results are provided in expectation. Moreover, the only known high-probability complexity results have been derived under restrictive sub-Gaussian (light-tailed) noise and bounded domain assumption [Juditsky et al., 2011]. In this work, we prove the first high-probability complexity results with logarithmic dependence on the confidence level for stochastic methods for solving monotone and structured non-monotone VIPs with non-sub-Gaussian (heavy-tailed) noise and unbounded domains. In the monotone case, our results match the best-known ones in the light-tails case [Juditsky et al., 2011], and are novel for structured non-monotone problems such as negative comonotone, quasi-strongly monotone, and/or star-cocoercive ones. We achieve these results by studying SEG and SGDA with clipping. In addition, we numerically validate that the gradient noise of many practical GAN formulations is heavy-tailed and show that clipping improves the performance of SEG/SGDA.
翻译:由于机器学习中对抗性配方越来越受人欢迎,近年来,由于在机器学习中日益流行的对抗性配方(VIP)越来越受人关注,在Stochastistic Extradient(SEG)或Stochastetic Gladient-Emprole-Accent(SGDA)(SGDA)(SSGDA)(SGDA)解决平滑的小型问题和更广泛的变异性不平等问题(VIP)方面,第一个高概率的第一阶方法(SVIP)(SVIP)(SEDA)(SEGE) (SEGE) (SEGE) (SEGE) (SEG) (SE) (SE) (SB) (SOI(SO) (SO) (SOI) (SOI) (SOI) (SO) (SO) (SO) (SODI) (SO) (SO) (O(O-I) (O-O(OI-OD) (O) (ODI-I) (O(SI) (O) (O) (O) (O(SI) (O(SI) (O) (O) (O) (O) (SI) (SI) (SI) (SI) (O) (S) (S) (SLI) (O) (S) (S) (S) (S) (S) (S) AL AL) (S) ((S) AL) AL AL AL AL AL) ((S) AL AL AL) AL AL AL) ((S) (S) (S) AL) (的) AL) AL AL AL AL AL) AL ) (S) AL) ((S) AL AL AL AL) AL) ((S) (O) AL AL ) ) AL AL AL ) ) ) ) ) AL AL AL ty ) ) ty ) ) (O) ((O) ((O) AL) AL)