Near-Universal Multiplicative Updates for Nonnegative Einsum Factorization
John Hood, Aaron Schein · 2026
Read Near-Universal Multiplicative Updates for Nonnegative Einsum Factorization by John Hood, Aaron Schein (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02759.txt.
Description
John Hood and 1 co-author introduce a near-universal algorithm for fitting custom nonnegative tensor factorizations, in a 2026 paper. Multiway data appears across science, yet fitting non-standard factorization models is painful: gradient methods struggle under nonnegativity constraints, mature implementations cover few models, and hand-rolling an algorithm is slow work.
Their method, NNEinFact, uses multiplicative updates built on tensor contractions, the einsum formalism. A researcher specifies the model with a short string and picks from many loss functions, including a two-parameter divergence family. The algorithm provably converges to a stationary point, handles missing data, and fits tensors with hundreds of millions of entries in seconds. On real prediction tasks, custom models beat standard ones by over a third, with under half the test loss of gradient methods and far faster convergence.
Catalog record
| CREATOR | John Hood, Aaron Schein |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 81 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-02759