Comparing PRNGs in PyTorch, TensorFlow, and NumPy against reference C implementations with TestU01 BigCrush reveals implementation-specific test failures, though most vanish under a strict p-value filter.
Replicability is not reproducibility: Nor is it good science,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.OH 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Statistical Quality and Reproducibility of Pseudorandom Number Generators in Machine Learning technologies
Comparing PRNGs in PyTorch, TensorFlow, and NumPy against reference C implementations with TestU01 BigCrush reveals implementation-specific test failures, though most vanish under a strict p-value filter.