Model-agnostic meta-learning (MAML) effectively meta-learns an initialization of model parameters for few-shot learning where all learning problems share the same format of model parameters -- congruous meta-learning. However, there are few-shot learning scenarios, such as adversarial attack design, where different yet related few-shot learning problems may not share any optimizee variables, necessitating incongruous meta-learning. We extend MAML to this setting -- a Learned Fine Tuner (LFT) is used to replace hand-designed optimizers (such as SGD) for the task-specific fine-tuning. Here, MAML instead meta-learns the parameters of this LFT across incongruous tasks leveraging the learning-to-optimize (L2O) framework such that models fine-tuned with LFT (even from random initializations) adapt quickly to new tasks. As novel contributions, we show that the use of LFT within MAML (i) offers the capability to tackle few-shot learning tasks by meta-learning across incongruous yet related problems and (ii) can efficiently work with first-order and derivative-free few-shot learning problems. Theoretically, we quantify the difference between LFT (for MAML) and L2O. Empirically, we demonstrate the effectiveness of LFT through a novel application of generating universal adversarial attacks across different image sources and sizes in the few-shot learning regime.
Download