4 Systems Perspectives into Human-Centered Machine Learning

Abstract

Machine learning (ML) has had a tremendous impact in across the world over the last decade. As we think about ML solving complex tasks, sometimes at super-human levels, it is easy to forget that there is no machine learning without humans in the loop. Humans define tasks and metrics, develop and program algorithms, collect and label data, debug and optimize systems, and are (usually) ultimately the users of the ML-based applications we are developing. In this talk, we will cover 4 human-centered perspectives in the ML development process, along with methods and systems, to empower humans to maximize the ultimate impact of their ML-based applications. In particular, we will cover: 1. Developer tools for ML that allow a wider range of people to create intelligent applications?, focusing on mobile devices. 2. Learning to optimize the performance and power of ML models on a wide range of hardware backends and mobile devices. 3. Closing the gap between the loss function we optimize in ML and the product metrics we really want to optimize. 4. Helping humans understand why ML models make each prediction, when these models will break, and how to improve them.