Unsupervised Domain Adaptation through Iterative Consensus Shift in a Multi-Task Graph.

Babies learn with very little supervision by observing the surrounding world. They synchronize the feedback from all their senses and learn to maintain consistency and stability among their internal states. Such observations inspired recent works in multi-task and multi-modal learning, but existing methods rely on expensive manual supervision. In contrast, our proposed multi-task graph, with consensus shift learning, relies only on pseudo-labels provided by expert models. In our graph, every node represents a task, and every edge learns to transform one input node into another. Once initialized, the graph learns by itself on virtually any novel target domain. An adaptive selection mechanism finds consensus among multiple paths reaching a given node and establishes the pseudo-ground truth at that node. Such pseudo-labels, given by ensemble pathways in the graph, are used during the next learning iteration when single edges distill this distributed knowledge. We validate our key contributions experimentally and demonstrate strong performance on the Replica dataset, superior to the very few published methods on multi-task learning with minimal supervision.