Lowering the barrier to applying machine learning

Machine learning algorithms are key components in many cutting edge applications of computation. However, the full potential of machine learning has not been realized because using machine learning is hard, even for otherwise tech-savvy developers. This is because developing with machine learning is different than normal programming. My thesis is that developers applying machine learning need new general-purpose tools that provide structure for common processes and common pipelines while remaining flexible to account for variability in problems. In this paper, I describe my efforts to understanding the difficulties that developers face when applying machine learning. I then describe Gestalt, a general-purpose integrated development environment designed the application of machine learning. Finally, I describe work on developing a pattern language for building machine learning systems and creating new techniques that help developers understand the interaction between their data and learning algorithms.