Paper abstract

Modeling Transfer Relationships Between Learning Tasks for Improved Inductive Transfer

Eric Eaton - University of Maryland Baltimore County, USA
Marie desJardins - University of Maryland Baltimore County, USA
Terran Lane - University of New Mexico, USA

Session: Transfer Learning
Springer Link: http://dx.doi.org/10.1007/978-3-540-87479-9_39

In this paper, we propose a novel graph-based method for knowledge transfer. We model the transfer relationships between source tasks by embedding the set of learned source models in a graph using transferability as the metric. Transfer to a new problem proceeds by mapping the problem into the graph, then learning a function on this graph that automatically determines the parameters to transfer to the new learning task. This method is analogous to inductive transfer along a manifold that captures the transfer relationships between the tasks. We demonstrate improved transfer performance using this method against existing approaches in several real-world domains.