Enhancement of Deep Neural Network Recognition on MPSoC with Single Event Upset
Weitao Yang, Wuqing Song, Yaxin Guo, Yonghong Li, Chaohui He, Longsheng Wu, Bin Wang, Huan Liu, Guang Shi · 2023
Read Enhancement of Deep Neural Network Recognition on MPSoC with Single Event Upset by Weitao Yang, Wuqing Song, Yaxin Guo, Yonghong Li, Chaohui He, Longsheng Wu, Bin Wang, Huan Liu, Guang Shi (2023) 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/pmc-10745775.txt.
Description
Weitao Yang and co-authors report a surprise: a radiation fault can make a neural network better at its task. Single event upsets flip bits in chip memory and are usually classed as pure harm. The team tests deep networks running on a multiprocessor system-on-chip built in a modern process.
Their 2023 study injects faults through dynamic reconfiguration and watches recognition scores. One upset in the configuration memory lifted accuracy by a clear margin. The likely cause is tiny shifts in weights or biases, which wake neurons to better levels and sharpen the final call. The authors float the odd idea of using such flips, in a tamed form, as a cheap performance lever.
Catalog record
| CREATOR | Weitao Yang, Wuqing Song, Yaxin Guo, Yonghong Li, Chaohui He, Longsheng Wu, Bin Wang, Huan Liu, Guang Shi |
|---|---|
| YEAR | 2023 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |