Semi-sorted batching with variable length input for efficient training
US11615782B2 · kind B2 · utility
0Cited by
9References
20Claims
0Family size
Assignee
Inventors
Key dates
| Filing date | Nov 30, 2020 |
| Grant date | Mar 28, 2023 |
| Priority date | — |
| Expiry date | Jul 10, 2041 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG10L15/06
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
Techniques are described for training neural networks on variable length datasets. The numeric representation of the length of each training sample is randomly perturbed to yield a pseudo-length, and the samples sorted by pseudo-length to achieve lower zero padding rate (ZPR) than completely randomized batching (thus saving computation time) yet higher randomness than strictly sorted batching (thus achieving better model performance than strictly sorted batching).
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.