MTL2L: A Context Aware Neural Optimiser
This addresses the problem of input-domain heterogeneity for meta-learning researchers, representing an incremental improvement over previous neural optimizers.
The paper tackles the limitation of neural optimizers being restricted to a single dataset by introducing MTL2L, a context-aware neural optimizer that adapts its rules based on input data, enabling it to update learners for classification on unseen input domains during meta-testing.
Learning to learn (L2L) trains a meta-learner to assist the learning of a task-specific base learner. Previously, it was shown that a meta-learner could learn the direct rules to update learner parameters; and that the learnt neural optimiser updated learners more rapidly than handcrafted gradient-descent methods. However, we demonstrate that previous neural optimisers were limited to update learners on one designated dataset. In order to address input-domain heterogeneity, we introduce Multi-Task Learning to Learn (MTL2L), a context aware neural optimiser which self-modifies its optimisation rules based on input data. We show that MTL2L is capable of updating learners to classify on data of an unseen input-domain at the meta-testing phase.