arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2209.01127cs.LGcs.AI
收藏

线性化潜在状态空间多步预测的表示学习

Multi-Step Prediction in Linearized Latent State Spaces for Representation Learning

A. Tytarenko

摘要:

在本文中,我们推导出一种新的方法,作为对LCE(如E2 C)的推广。该方法发展了学习局部线性状态空间的思想,通过添加多步预测,从而允许对曲率进行更明确的控制。我们表明,该方法优于E2 C,而无需剧烈的模型变化,这与其他工作,如PCC和P3 C。我们讨论了E2 C和所提出的方法之间的关系,并推导出更新方程。我们提供了经验证据,这表明,通过考虑多步预测,我们的方法-ms-E2 C-允许学习更好的潜在状态空间的曲率和下一个状态的可预测性。最后,我们还讨论了我们在多步预测中遇到的某些稳定性挑战以及缓解这些挑战的方法。

英文摘要:

In this paper, we derive a novel method as a generalization over LCEs such as E2C. The method develops the idea of learning a locally linear state space, by adding a multi-step prediction, thus allowing for more explicit control over the curvature. We show, that the method outperforms E2C without drastic model changes which come with other works, such as PCC and P3C. We discuss the relation between E2C and the presented method and derived update equations. We provide empirical evidence, which suggests that by considering the multi-step prediction our method - ms-E2C - allows to learn much better latent state spaces in terms of curvature and next state predictability. Finally, we also discuss certain stability challenges we encounter with multi-step predictions and the ways to mitigate them.

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。

↑