Transfer of learning by composing solutions of elemental sequential tasks (Q1812933)

A class of sequential decision tasks (SDTs) is considered, called composite sequential decision tasks, formed by temporally concatenating a number of elemental sequential decision tasks. Elemental SDTs cannot be decomposed into simpler SDTs. A learning agent has to learn to solve a set of elemental and composite SDTs is examined. It is assumed that the structure of the composite tasks is unknown to the learning agent. The straightforward application of reinforcement learning to multiple tasks requires learning the tasks separately, which can waste computational resources, both memory and time. A new learning algorithm and a modular architecture is described that learns the decomposition of composite SDTs, and achieves transfer of learning by sharing the solutions of elemental SDTs across multiple composite SDTs. The solution of a composite SDT is constructed by computationally inexpensive modifications of the solutions of its constituent elemental SDTs. A proof of one aspect of the learning algorithm is provided.