Freenet routing uses isotonic regression and empirical Bayes, not neural networks

Sanity · x · 2026-11-20

The author notes that the right machine learning solution often isn't a neural network. For routing in Freenet, they use isotonic regression, empirical Bayes, and other classical statistical techniques.

The advantage: existing domain knowledge can be used to constrain the model, enabling useful predictions even with limited data — a pragmatic counterpoint to defaulting to deep learning everywhere.

Related event: Freenet Drops Neural Nets for Routing in Favor of Classic Statistics(3 posts)→

Original post →

More from Research

Research channel →