arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1011.0686cs.LGcs.AIstat.ML
收藏

模仿学习和结构化预测到无遗憾在线学习的约化

A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning

Carnegie Mellon University(卡内基梅隆大学)

Stephane Ross, Geoffrey J. Gordon, J. Andrew Bagnell

摘要:

序列预测问题,如模仿学习,未来的观察取决于以前的预测(行动),违反了共同的独立同分布。统计学习中的假设。这在理论上和实践中往往导致性能不佳。最近的一些方法在这种情况下提供了更强的保证,但仍然有些不令人满意,因为它们训练非平稳或随机的政策,并需要大量的迭代。在本文中,我们提出了一个新的迭代算法,它训练一个固定的确定性政策,可以被看作是一个没有遗憾的算法在在线学习设置。我们表明,任何这样的无遗憾算法,结合额外的减少假设,必须找到一个政策下,它诱导在这样的顺序设置的观测分布具有良好的性能。我们证明了这种新的方法优于以前的方法在两个具有挑战性的模仿学习问题和基准序列标记问题。

英文摘要:

Sequential prediction problems such as imitation learning, where future observations depend on previous predictions (actions), violate the common i.i.d. assumptions made in statistical learning. This leads to poor performance in theory and often in practice. Some recent approaches provide stronger guarantees in this setting, but remain somewhat unsatisfactory as they train either non-stationary or stochastic policies and require a large number of iterations. In this paper, we propose a new iterative algorithm, which trains a stationary deterministic policy, that can be seen as a no regret algorithm in an online learning setting. We show that any such no regret algorithm, combined with additional reduction assumptions, must find a policy with good performance under the distribution of observations it induces in such sequential settings. We demonstrate that this new approach outperforms previous approaches on two challenging imitation learning problems and a benchmark sequence labeling problem.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑