arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2005.03648cs.LGcs.AIstat.ML
收藏

Plan2Vec:基于潜在计划的无监督表示学习

Plan2Vec: Unsupervised Representation Learning by Latent Plans

Facebook AI Research(脸书人工智能研究院);McGill University(麦吉尔大学);UC Berkeley(加州大学伯克利分校)

Ge Yang, Amy Zhang, Ari S. Morcos, Joelle Pineau, Pieter Abbeel, Roberto Calandra

摘要:

在本文中,我们介绍了plan2vec,一种受强化学习启发的无监督表示学习方法。Plan2vec使用近邻距离在图像数据集上构建加权图,然后通过提取规划路径上的路径积分将此局部度量外推到全局嵌入。当应用于控制时,plan2vec提供了一种学习目标条件值估计的方法,该方法在长时间范围内都是准确的,计算和采样效率都很高。我们证明了plan2vec在一个模拟和两个具有挑战性的现实世界的图像数据集上的有效性。实验结果表明,plan2vec成功地摊销规划成本,使反应式规划,是线性的内存和计算复杂性,而不是在整个状态空间的穷举。

英文摘要:

In this paper we introduce plan2vec, an unsupervised representation learning approach that is inspired by reinforcement learning. Plan2vec constructs a weighted graph on an image dataset using near-neighbor distances, and then extrapolates this local metric to a global embedding by distilling path-integral over planned path. When applied to control, plan2vec offers a way to learn goal-conditioned value estimates that are accurate over long horizons that is both compute and sample efficient. We demonstrate the effectiveness of plan2vec on one simulated and two challenging real-world image datasets. Experimental results show that plan2vec successfully amortizes the planning cost, enabling reactive planning that is linear in memory and computation complexity rather than exhaustive over the entire state space.

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。

↑