Semantic Anticipation for Understanding Using Neural Networks

Abstract

This paper presents a new method for incorporating into a recurrent network the anticipated outputs for items in the right-hand context, i.e. for items in the input stream that have not yet been encountered. We rst expose a recurrent system built for a semantic feature tagging task. We then discuss motivations for anticipating outputs within this… (More)

Topics

3 Figures and Tables

Slides referencing similar topics