RI: Mediuim: Abstraction, Equilibrium Finding, Safe Opponent Exploitation, and Robust Strategies for Imperfect-Information Games

点击翻译按钮获取中文摘要

While game theory has traditionally been used as a thinking tool to analyze small stylized versions of real situations, this project is developing operational tools for analyzing real large-scale settings in full detail. To ensure true scalability, the techniques will be benchmarked on very-large-scale games of negotiation, auctions, and poker. This work is on a path toward a vision where software agents conduct commerce on behalf of humans, or at least advise them to act more intelligently. This will lead to increased social welfare through better decision making. This will also enable broader and fairer access because it will help put less experienced/educated people/companies on an equal footing with expert market participants. Broader access will, in turn, increase the benefits of (electronic) commerce further, and the benefits will be distributed more fairly across segments of society. The paradigm will also yield a form of increased security because strategies generated by the proposed techniques cannot be taken advantage of (at least not by much and not easily). The projects particular contributions to computational game theory include the development of (lossless and lossy) game-abstraction methods to allow scale-up to larger 'games' of societal import, techniques for settings with more than two agents, techniques for dealing with less-than-rational opponents robustly, and derivation of worst-case guarantees on approximate solutions via nemesis computation.