A fundamental trait of intelligence is the ability to achieve goals in the
face of novel circumstances, such as making decisions from new action choices.
However, standard reinforcement learning assumes a fixed set of actions and
requires expensive retraining when given a new action se