Near-optimal no-regret learning for correlated equilibria in multi-player general-sum games
Near-optimal no-regret learning for correlated equilibria in multi-player general-sum games
复制标题
多人一般和博弈中相关均衡的近乎最优无悔学习
DOI:
10.1145/3519935.3520031
复制
发表时间:
2022
期刊:
影响因子:
--
通讯作者:
Sandholm, Tuomas
中科院分区:
文献类型:
--
作者:
Anagnostides, Ioannis;Daskalakis, Constantinos;Farina, Gabriele;Fishelson, Maxwell;Golowich, Noah;Sandholm, Tuomas
Recently, Daskalakis, Fishelson, and Golowich (DFG) (NeurIPS ‘21) showed that if all agents in a multi-player general-sum normal-form game employ Optimistic Multiplicative Weights Update (OMWU), the external regret of every player isO(polylog(T)) afterTrepetitions of the game. In this paper we extend their result from external regret to internal and swap regret, thereby establishing uncoupled learning dynamics that converge to an approximate correlated equilibrium at the rate ofO(T−1). This substantially improves over the prior best rate of convergence ofO(T−3/4) due to Chen and Peng (NeurIPS ‘20), and it is optimal up to polylogarithmic factors.To obtain these results, we develop new techniques for establishing higher-order smoothness for learning dynamics involving fixed point operations. Specifically, we first establish that the no-internal-regret learning dynamics of Stoltz and Lugosi (Mach Learn ‘05) are equivalently simulated by no-external-regret dynamics on a combinatorial space. This allows us to trade the computation of the stationary distribution on a polynomial-sized Markov chain for a (much more well-behaved) linear transformation on an exponential-sized set, enabling us to leverage similar techniques as DGF to near-optimally bound the internal regret.Moreover, we establish anO(polylog(T)) no-swap-regret bound for the classic algorithm of Blum and Mansour (BM) (JMLR ‘07). We do so by introducing a technique based on the Cauchy Integral Formula that circumvents the more limited combinatorial arguments of DFG. In addition to shedding clarity on the near-optimal regret guarantees of BM, our arguments provide insights into the various ways in which the techniques by DFG can be extended and leveraged in the analysis of more involved learning algorithms.
登录
查看更多内容
DOI:
10.1016/j.geb.2014.01.003
发表时间:
2011
期刊:
ArXiv
影响因子:
--
作者:
C. Daskalakis;Alan Deckelbaum;A. Kim
通讯作者:
A. Kim
DOI:
--
发表时间:
2012-08
期刊:
ArXiv
影响因子:
--
作者:
A. Rakhlin;Karthik Sridharan
通讯作者:
A. Rakhlin;Karthik Sridharan
DOI:
10.4230/lipics.itcs.2019.27
发表时间:
2018-07
期刊:
--
影响因子:
--
作者:
C. Daskalakis;Ioannis Panageas
通讯作者:
C. Daskalakis;Ioannis Panageas
DOI:
--
发表时间:
2010
期刊:
Annual Conference Computational Learning Theory
影响因子:
--
作者:
Jacob D. Abernethy;P. Bartlett;Elad Hazan
通讯作者:
Elad Hazan
DOI:
--
发表时间:
2018-07
期刊:
--
影响因子:
--
作者:
C. Daskalakis;Ioannis Panageas
通讯作者:
C. Daskalakis;Ioannis Panageas