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

Active inference is a Bayesian framework for understanding biologicalintelligence. The underlying theory brings together perception and action underone single imperative: minimizing free energy. However, despite its theoreticalutility in explaining intelligence, computational implementations have beenrestricted to low-dimensional and idealized situations. In this paper, wepresent a neural architecture for building deep active inference agentsoperating in complex, continuous state-spaces using multiple forms ofMonte-Carlo (MC) sampling. For this, we introduce a number of techniques, novelto active inference. These include: i) selecting free-energy-optimal policiesvia MC tree search, ii) approximating this optimal policy distribution via afeed-forward `habitual' network, iii) predicting future parameter beliefupdates using MC dropouts and, finally, iv) optimizing state transitionprecision (a high-end form of attention). Our approach enables agents to learnenvironmental dynamics efficiently, while maintaining task performance, inrelation to reward-based counterparts. We illustrate this in a new toyenvironment, based on the dSprites data-set, and demonstrate that activeinference agents automatically create disentangled representations that are aptfor modeling state transitions. In a more complex Animal-AI environment, ouragents (using the same neural architecture) are able to simulate future statetransitions and actions (i.e., plan), to evince reward-directed navigation -despite temporary suspension of visual input. These results show that deepactive inference - equipped with MC methods - provides a flexible framework todevelop biologically-inspired intelligent agents, with applications in bothmachine learning and cognitive science.

Deep active inference agents using Monte-Carlo methods