This workshop focuses on new machine learning techniques for automatically designing algorithms. Algorithms are central to modern computing, and they have lots of applications in our life. Yet, writing correct, efficient algorithms is a time-consuming and difficult task. It also often requires intuition and expertise to tailor algorithmic choices to specific instances that arise in particular applications. However, there have been a number of recent advancements that have allowed algorithms to be selected or designed from specific algorithmic families automatically, often leading to either state-of-the-art empirical performance or provable performance guarantees on observed instance distributions. In this workshop, we take a broad view of the problem and seek to bring together researchers with different viewpoints and approaches to the general challenge.