English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Reinforcement learning (RL) has garnered significant attention for developingdecision-making agents that aim to maximize rewards, specified by an externalsupervisor, within fully observable environments. However, many real-worldproblems involve partial observations, formulated as partially observableMarkov decision processes (POMDPs). Previous studies have tackled RL in POMDPsby either incorporating the memory of past actions and observations or byinferring the true state of the environment from observed data. However,aggregating observed data over time becomes impractical in continuous spaces.Moreover, inference-based RL approaches often require many samples to performwell, as they focus solely on reward maximization and neglect uncertainty inthe inferred state. Active inference (AIF) is a framework formulated in POMDPsand directs agents to select actions by minimizing a function called expectedfree energy (EFE). This supplies reward-maximizing (exploitative) behaviour, asin RL, with information-seeking (exploratory) behaviour. Despite thisexploratory behaviour of AIF, its usage is limited to discrete spaces due tothe computational challenges associated with EFE. In this paper, we propose aunified principle that establishes a theoretical connection between AIF and RL,enabling seamless integration of these two approaches and overcoming theiraforementioned limitations in continuous space POMDP settings. We substantiateour findings with theoretical analysis, providing novel perspectives forutilizing AIF in the design of artificial agents. Experimental resultsdemonstrate the superior learning capabilities of our method in solvingcontinuous space partially observable tasks. Notably, our approach harnessesinformation-seeking exploration, enabling it to effectively solve reward-freeproblems and rendering explicit task reward design by an external supervisoroptional.

Active Inference and Reinforcement Learning: A unified inference on continuous state and action spaces under partially observability