Task Weighting in Meta-learning with Trajectory Optimisation

Developing meta-learning algorithms that are un-biased toward a subset of training tasks often requires hand-designed criteria to weight tasks, potentially resulting in sub-optimal solutions. In this paper, we introduce a new principled and fully-automated task-weighting algorithm for meta-learning methods. Specifically, we frame the task-weighting problem as a trajectory optimization problem, where the weights of tasks within a mini-batch are treated as an action, and the meta-parameter of interest is viewed as the system state. Such a modelling allows us to employ the iterative linear quadratic regulator to determine the optimal task weights. We theoretically show that the proposed algorithm converges to an ϵ₀-stationary point, and empirically demonstrate that the proposed approach out-performs common hand-engineering weighting methods on two few-shot learning benchmarks.