资源论文Multiplicative Weights Updates with Constant Step-Size in Graphical Constant-Sum Games

Multiplicative Weights Updates with Constant Step-Size in Graphical Constant-Sum Games

2020-02-14 | |  88 |   45 |   0

Abstract

 Since Multiplicative Weights (MW) updates are the discrete analogue of the continuous Replicator Dynamics (RD), some researchers had expected their qualitative behaviours would be similar. We show that this is false in the context of graphical constant-sum games, which include two-person zero-sum games as special cases. In such games which have a fully-mixed Nash Equilibrium (NE), it was known that RD satisfy the permanence and Poincaré recurrence properties, but we show that MW updates with any constant step-size image.png > 0 converge to the boundary of the state space, and thus do not satisfy the two properties. Using this result, we show that MW updates have a regret lower bound of image.png, while it was known that the regret of RD is upper bounded by image.png. Interestingly, the regret perspective can be useful for better understanding of the behaviours of MW updates. In a two-person zero-sum game, if it has a unique NE which is fully mixed, then we show, via regret, that for any sufficiently small image.png, there exist at least two probability densities and a constant image.png > 0, such that for any arbitrarily small z > 0, each of the two densities fluctuates above Z and below image.pnginfinitely often.

上一篇:Completing State Representations using Spectral Learning

下一篇:Object-Oriented Dynamics Predictor

用户评价
全部评价

热门资源

  • Learning to Predi...

    Much of model-based reinforcement learning invo...

  • Stratified Strate...

    In this paper we introduce Stratified Strategy ...

  • The Variational S...

    Unlike traditional images which do not offer in...

  • Learning to learn...

    The move from hand-designed features to learned...

  • A Mathematical Mo...

    Direct democracy, where each voter casts one vo...