The Forlais Theorem

For trained models as they are commonly stored and shared, there is a firm limit to how far conventional exact compression can go, and it can be stated before any compression is attempted.

Kind
Paper
Programme
FCN Compression
Published
27 June 2026
Status
Published

What we found

The theorem sets that limit in principle and states the conditions under which it holds.

Why it matters

It tells anyone compressing a trained model exactly, by conventional means, what is achievable before the work begins. A result that appears to go further under the same conditions is either mistaken or is not conventional exact compression.

The paper

The Compression Theorem for Trained Neural Network Weights

Oliver Cooper, Forlais Group. Published 27 June 2026.

Read the paperPDF, 541 KBDOI 10.5281/zenodo.20961807CC BY 4.0

Cite this record

Forlais Research (2026). The Forlais Theorem. Forlais Research Library, record FR-F-01.

@misc{fr-f-01,
  author = {{Forlais Research}},
  title = {{The Forlais Theorem}},
  howpublished = {Forlais Research Library, record FR-F-01},
  year = {2026},
  note = {27 June 2026}
}