274e6fcf4a583de4a81c6376f17673e7-AuthorFeedback.pdf

How can I adapt (R2) or extend (R5) IMAGINE to other RL approaches? Our descriptive setup was studied in33 isolation from instruction-feedback, but the two can be combined. Two features ofIMAGINE could be added to34 traditional RL approaches (R2,R5). In new environments,40 reusing learned patterns islikelytoboostfurther exploration: thissuggests interesting future work. In our approach, sentences are goals.