期刊
ENTROPY
卷 24, 期 3, 页码 -出版社
MDPI
DOI: 10.3390/e24030401
关键词
autonomy; reinforcement learning; information theory; partial information decomposition; non-trivial informational closure; deep learning
资金
- Estonian Centre of Excellence in IT (EXCITE) [TK148]
- European Union [952060]
- European Social Fund via IT Academy Programme [SLTAT18311]
- Estonian Research Council [PRG1604]
- Niedersachsisches Vorab (of the VolkswagenStiftung under the program `Big Data in den Lebenswissenschaften') [ZN3326, ZN3371]
In this study, we introduce an algorithm for computing the level of autonomy of an agent using an information-theoretic formulation. We use the partial information decomposition framework to monitor the autonomy level and environment internalization of reinforcement learning agents. Our experiments show strong correlations between specific PID terms and the obtained reward, as well as the agent's behavior in response to perturbations in the observations.
Intuitively, the level of autonomy of an agent is related to the degree to which the agent's goals and behaviour are decoupled from the immediate control by the environment. Here, we capitalise on a recent information-theoretic formulation of autonomy and introduce an algorithm for calculating autonomy in a limiting process of time step approaching infinity. We tackle the question of how the autonomy level of an agent changes during training. In particular, in this work, we use the partial information decomposition (PID) framework to monitor the levels of autonomy and environment internalisation of reinforcement-learning (RL) agents. We performed experiments on two environments: a grid world, in which the agent has to collect food, and a repeating-pattern environment, in which the agent has to learn to imitate a sequence of actions by memorising the sequence. PID also allows us to answer how much the agent relies on its internal memory (versus how much it relies on the observations) when transitioning to its next internal state. The experiments show that specific terms of PID strongly correlate with the obtained reward and with the agent's behaviour against perturbations in the observations.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据