arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.00403cs.LG
收藏

DROGO: 通过图优化在强化学习中实现默认表示目标

DROGO: Default Representation Objective via Graph Optimization in Reinforcement Learning

Department of Computing Science, University of Alberta(阿尔伯塔大学计算机科学系);Alberta Machine Intelligence Institute (Amii)(阿尔伯塔人工智能研究所);Canada CIFAR AI Chair(加拿大 CIFAR 人工智能主席)

Hon Tik Tse, Marlos C. Machado

AI总结:

DROGO通过图优化在强化学习中直接近似默认表示的主特征向量,用于奖励塑造和多种应用。

摘要:

在计算强化学习中,默认表示(DR)及其主特征向量已被证明在多种应用中有效,包括奖励塑造、基于计数的探索、选项发现和迁移。然而,在先前的研究中,DR的特征向量是通过首先近似DR矩阵,然后进行特征分解来计算的。这一过程计算成本高,无法扩展到高维空间。在本文中,我们推导出一个目标,直接用神经网络近似DR的主特征向量。我们通过多个环境实证展示了该目标的有效性,并将学习到的特征向量应用于奖励塑造。

英文摘要:

In computational reinforcement learning, the default representation (DR) and its principal eigenvector have been shown to be effective for a wide variety of applications, including reward shaping, count-based exploration, option discovery, and transfer. However, in prior investigations, the eigenvectors of the DR were computed by first approximating the DR matrix, and then performing an eigendecomposition. This procedure is computationally expensive and does not scale to high-dimensional spaces. In this paper, we derive an objective for directly approximating the principal eigenvector of the DR with a neural network. We empirically demonstrate the effectiveness of the objective in a number of environments, and apply the learned eigenvectors for reward shaping.

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。

↑