Planning

cosmos 17th July 2017 at 3:25pm
Reinforcement learning

Used in different ways in different fields/contexts

In reinforcement learning

methods that given a model of the environment, it tries to find optimal policies (depends on definition of model..)

Solving the RL (prediction/control) problem in the context of a fully-known of MDP. It also refers to approximate solution of the problem. See Model-based reinforcement learning

Neural Mechanisms of Hierarchical Planning in a Virtual Subway Network


See Sutton-Barto, chapter 8.

–> Model-free reinforcement learning methods can be used for planning, when trained using simulated experience

State-space planning

Plan-space planning


Background planning

When planning occurs in parallel and in a sense independently of other learning/acting/decisions processes.

Decision-time planning

Planning which occurs when going to decide which action to take every time a new state is visited. The backups performed focus forward from states which most affect the current state (i.e. those that are likely to occur in the future).