THE PARALLEL TRANSFER OF TASK KNOWLEDGE USING DYNAMIC LEARNING RATES
by Daniel L. Silver and Robert E. Mercer
With a distinction made between two forms of task knowledge transfer, representational and functional, etaMTL, a modified version of the MTL method of functional (parallel) transfer, is introduced. The etaMTL method employs a separate learning rate, eta-k, for each task output node k. eta-k varies as a function of a measure of relatedness, R-k, between the k-th task and the primary task of interest. Results of experiments demonstrate the ability of etaMTL to dynamically select the most related source task(s) for the functional transfer of prior domain knowledge. The etaMTL method of learning is nearly equivalent to standard MTL when all parallel tasks are sufficiently related to the primary task, and is similar to single task learning when none of the parallel tasks are related to the primary task.