Interpretable agent communication from scratch (with a generic visual processor emerging on the side)
Abstract
As deep networks begin to be deployed as autonomous agents, the issue of how they can communicate with each other becomes important. Here, we train two deep nets from scratch to perform large-scale referent identification through unsupervised emergent communication. We show that the partially interpretable emergent protocol allows the nets to successfully communicate even about object classes they did not see at training time. The visual representations induced as a by-product of our training regime, moreover, when re-used as generic visual features, show comparable quality to a recent self-supervised learning model. Our results provide concrete evidence of the