EdgeOpt-Sched-CS: Cold-Start-Aware Dynamic Scheduling for Efficient DNN Inference at the Edge
Yuchang Gu, Diming Zhang, Taiyu Lu · 2026
Read EdgeOpt-Sched-CS: Cold-Start-Aware Dynamic Scheduling for Efficient DNN Inference at the Edge by Yuchang Gu, Diming Zhang, Taiyu Lu (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/pmc-13210435.txt.
Description
Neural networks that run on small edge devices, like cameras and sensors, often stall when a new model first starts up. Yuchang Gu, Diming Zhang, and Taiyu Lu tackle that cold start in a 2026 paper. Their scheduler, EdgeOpt-Sched-CS, decides the order in which a network's parts run on the device, aiming for low delay and low memory use.
Instead of learning that order from scratch, the scheduler borrows knowledge from similar networks it has seen before. Each network is summed up in a compact signature, matched against past ones, and then tuned lightly on the job. Across tests with image networks, transformer models, and small language models, the method cut total cold-start delay by about 11 to 20 percent. It also reached steady behavior faster, with only small added cost.
Catalog record
| CREATOR | Yuchang Gu, Diming Zhang, Taiyu Lu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 83 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |