Plan2Vec: 基于潜在计划的无监督表示学习（CS LG）---刘持诚

在本文中，我们介绍了 Plan2Vec，一种无监督表示学习方法，它的灵感来自于强化学习。Plan2Vec 使用近邻距离在图像数据集上构建一个加权图，然后通过提炼在计划路径上的路径积分，将这个局部度量推导到全局嵌入。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，这种方法可以在长范围内准确地学习目标条件值估计，而且计算和采样效率都很高。我们在一个模拟和两个具有挑战性的真实世界图像数据集上证明了 Plan2Vec 的有效性。实验结果表明，Plan2Vec 成功地摊销了规划成本，实现了在内存和计算复杂度上是线性的反应式规划，而不是在整个状态空间中的线性规划。

原文题目：Plan2Vec: Unsupervised Representation Learning by Latent Plans

原文：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.

原文作者：Ge Yang, Amy Zhang, Ari S. Morcos, Joelle Pineau, Pieter Abbeel, Roberto Calandra

原文地址：https://arxiv.org/abs/2005.03648