Parallelizable Neural Turing Machines Instituto de Ciências Matemáticas e de Computação (ICMC), University of São Paulo.
Gabriel Faria, Arnaldo Candido Junior · 2026
Read Parallelizable Neural Turing Machines Instituto de Ciências Matemáticas e de Computação (ICMC), University of São Paulo. by Gabriel Faria, Arnaldo Candido Junior (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-2602-18508.txt.
Description
Gabriel Faria and Arnaldo Candido Junior make Neural Turing Machines parallel. Their 2026 paper introduces P-NTM, a simpler version changing the core steps of the original design for efficient scan-based execution.
The team tests the model on a synthetic benchmark of algorithm problems covering state tracking, memorization and arithmetic, solved with autoregressive decoding. They compare it against a revisited stable build of standard NTM, plus conventional recurrent and attention models. Despite simplifying, the model works across lengths like the original, solving all problems, including unseen lengths, with perfect accuracy. Parallel execution of P-NTM also improves training speed, running up to ten times faster than the standard NTM. The work contributes toward efficient neural models able to express broad algorithms.
Catalog record
| CREATOR | Gabriel Faria, Arnaldo Candido Junior |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-18508