Supporting Root Cause Analysis of Inaccurate Bug Prediction Based on Machine Learning – Lessons Learned When Interweaving Training Data and Source Code
Affiliation Type:
Industry
Keywords:
Machine Learning, Bug Prediction, Verification
Abstract:
How do you verify and debug an implementation based on machine learning (ML)? You got the proof of concept working but now the real implementation is not working. Where do you start? The challenge is that the root cause could be almost anywhere. e.g. bad training data, implementation bugs or a mismatch between data or features during training and inference. This paper describes our lessons learned from a case study. Our implementation is a bug prediction mechanism implemented in software, but we believe these insights could be of interest to anyone working with ML, be it software or hardware.
Track ID:
3.1
Track Name:
Automating the Optimization of Verification Processes
