Learning to learn by solving many tasks is called meta-learning. The idea behind meta-learning for neural networks is to train a system on many tasks with the goal to solve new tasks and adapt to new environments quickly; by observing how different machine learning approaches perform on a wide range of learning tasks. Current artificial intelligence systems excel at constructing a single approach for a singular task. However, these systems fail to generalize and perform adequately when the task or the approach is altered. Several promising meta-learning approaches and success stories of meta-learning will be surveyed in this literature review.
Meta-learning