The Significance of Interfaces for Machine Learning
Interfaces for machine learning (ML) play a crucial role in building robust and responsible ML systems. They provide practitioners with the necessary tools to access information, visualize models and data, and develop effective ML models. However, despite their numerous benefits, recent studies have shown that ML interfaces have limited adoption in practice.
Challenges with Existing ML Interfaces
While existing ML interfaces are effective for specific tasks, they lack the ability to be reused, explored, and shared by multiple stakeholders in cross-functional teams. This limitation hinders collaboration and restricts the analysis and communication between different ML practitioners.
Introducing Symphony: A Framework for Interactive ML Interfaces
To address these challenges, we have designed and implemented Symphony, a framework that allows for the creation of interactive ML interfaces. Symphony consists of task-specific, data-driven components that can be used across different platforms, including computational notebooks and web dashboards.
Development and Deployment of Symphony
We developed Symphony through participatory design sessions with 10 teams, involving a total of 31 participants. This collaborative approach ensured that Symphony catered to the needs and preferences of the ML community. We then deployed Symphony to three production ML projects at Apple.
The Impact of Symphony
Thanks to Symphony, ML practitioners were able to identify previously unknown issues such as data duplicates and blind spots in their models. The framework also facilitated the sharing of insights and findings with other stakeholders, improving collaboration and enhancing the overall ML workflow.