On Human Intellect and Machine Failures: Troubleshooting Integrative Machine Learning Systems
This addresses the difficulty of diagnosing and fixing failures in complex ML pipelines for system designers, though it is incremental as it builds on existing human-in-the-loop approaches.
The paper tackles the problem of troubleshooting errors in integrative machine learning systems with multiple components, proposing a human-in-the-loop methodology that simulates fixes to measure expected improvements, and demonstrates its effectiveness on an automated image captioning system in real-world use.
We study the problem of troubleshooting machine learning systems that rely on analytical pipelines of distinct components. Understanding and fixing errors that arise in such integrative systems is difficult as failures can occur at multiple points in the execution workflow. Moreover, errors can propagate, become amplified or be suppressed, making blame assignment difficult. We propose a human-in-the-loop methodology which leverages human intellect for troubleshooting system failures. The approach simulates potential component fixes through human computation tasks and measures the expected improvements in the holistic behavior of the system. The method provides guidance to designers about how they can best improve the system. We demonstrate the effectiveness of the approach on an automated image captioning system that has been pressed into real-world use.