Reinforcement Learning: An Introduction
A technical reference for learning through interaction with an environment.
THE RESEARCH GARDEN
Latest posts
Papers, evolving concepts and pieces of thinking about agentic systems.
7 postsRetaining context does not explain why activity continues
A system might remember an earlier task while requiring a new instruction before acting again.
Permission can outlast an opportunity to intervene
A useful research question is whether an authorisation remains active after meaningful human intervention becomes difficult.
Action space
A starting explanation of the set of actions available in a specified model or system.
When can a human still change what an agentic system does?
An emerging chapter question about intervention across a sequence of actions.
Observability: what can a human reconstruct?
A record of activity may still leave the reasons for a consequential action difficult to reconstruct.
Persistence: what continues, and for how long?
Persistence needs an object: an activity, a stored state, an authorisation or something else.
No posts match this search. Try another term or category.