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

We demonstrate a reinforcement learning agent which uses a compositionalrecurrent neural network that takes as input an LTL formula and determinessatisfying actions. The input LTL formulas have never been seen before, yet thenetwork performs zero-shot generalization to satisfy them. This is a novel formof multi-task learning for RL agents where agents learn from one diverse set oftasks and generalize to a new set of diverse tasks. The formulation of thenetwork enables this capacity to generalize. We demonstrate this ability in twodomains. In a symbolic domain, the agent finds a sequence of letters that isaccepted. In a Minecraft-like environment, the agent finds a sequence ofactions that conform to the formula. While prior work could learn to executeone formula reliably given examples of that formula, we demonstrate how toencode all formulas reliably. This could form the basis of new multitask agentsthat discover sub-tasks and execute them without any additional training, aswell as the agents which follow more complex linguistic commands. Thestructures required for this generalization are specific to LTL formulas, whichopens up an interesting theoretical question: what structures are required inneural networks for zero-shot generalization to different logics?

Encoding formulas as deep networks: Reinforcement learning for zero-shot execution of LTL formulas