A rigorous uncertainty-aware quantification framework is essential for reproducible and replicable machine learning workflows

Abstract

The ability to replicate predictions by machine learning (ML) or artificial intelligence (AI) models and results in scientific workflows that incorporate such ML/AI predictions is driven by numerous factors. An...

Artificial Intelligence Must Be Made More Scientific

“…This is reminiscent of molecular dynamics where one-off simulations are not reproducible. 11,12 There are other challenges to reproducibility, 13 which require access to the underlying data and ML algorithms employed, which may be kept confidential, 14,15 and sometimes access to substantial computational power.…”

“…Generative methods suffer from similar problems but are more strongly dependent on random number generators, so a fortiori , they produce different answers each time the code is run. This is reminiscent of molecular dynamics where one-off simulations are not reproducible. , There are other challenges to reproducibility, which require access to the underlying data and ML algorithms employed, which may be kept confidential, , and sometimes access to substantial computational power.…”