We consider the design of online no-regret algorithms that are computationally efficient, given access to an offline optimization oracle. We present an algorithm we call Generalized Follow-the-Perturbed-Leader and provide conditions under which it achieves vanishing regret and is oracle-efficient. Our second main contribution is introducing a new adversarial auction-design framework for revenue maximization and applying our oracle-efficient learning results to the adaptive optimization of auctions. Our learning algorithm is a generalization of the FTPL algorithm of Kalai and Vempala that at every step plays the best-performing action subject to some perturbations. Our design uses a shared source of randomness across all actions that can be efficiently implemented using an oracle. Our work extends to oracle-efficient algorithms for contextual learning, learning with Maximal-in-Range approximation algorithms, and no-regret bidding in simultaneous auctions, answering an open problem of Daskalakis and Syrgkanis in the latter case. Our auction-design framework considers an auctioneer learning an optimal auction for a sequence of adversarially selected valuations with the goal of achieving revenue that is almost as good as the optimal auction in hindsight, among a class of auctions. We give oracle-efficient learning results for: (1) VCG auctions with bidder-specific reserves in single-parameter settings, (2) envy-free item pricing in multi-item auctions, and (3) s-level auctions of Morgenstern and Roughgarden for single-item settings. The last result leads to an approximation of the optimal Myerson auction for the stationary distribution of a Markov process, extending prior work that only gave such guarantees for the i.i.d. setting. We also extend our framework to allow the auctioneer to use side information about the bidders in the design of the optimal auction (contextual learning).

本文研究了对手环境下的在线学习算法的设计，提出了广义随机扰动跟随者算法，且证明了在一定条件下它是牛逼优而且可以实现消失的后悔；同时，本文也提出了另一个基于拍卖设计的框架，用于帮助拍卖师在选举获得最佳的拍卖方式方面做出决策，并且得到相应的应用。

高效在线学习和拍卖设计