Meta-Learning Introduction

We hear a lot about new advances in deep learning every day. We want to make our models better and better with the minimum risk of choosing appropriate base learners in a safe decision-making process. This article is a brief introduction to meta-learning. There is no doubt that working on sequential datasets such as time series, natural language, etc. demands a huge work with a high level of uncertainties caused by external sources. Meta-learning is learning to learn where we use it to learn better learning algorithms, for example, parameter initialization, optimization strategy, network architecture, etc…