File:Proxylessnas-1.png
Size of this preview: 798 × 74 pixels. Other resolution: 1,574 × 146 pixels.
Original file (1,574 × 146 pixels, file size: 35 KB, MIME type: image/png)
Summary
| Description | English: In order to retrieve the output feature maps of all N paths, they are computed and subsequently saved in the memory. Nevertheless, training a compact model necessitates only a solitary path. This implies that executing the One-Shot and DARTS techniques consume roughly N times the amount of GPU memory and GPU hours needed to train a compact model. This can pose difficulties when working with vast datasets since it can readily go beyond the memory limits of hardware that has a large design space. |
| Date | 1 September 2018 |
| File source | https://arxiv.org/pdf/1812.00332.pdf |
| Author | Han Cai |
Licensing
|
File history
Click on a date/time to view the file as it appeared at that time.
| Date/Time | Thumbnail | Dimensions | User | Comment | |
|---|---|---|---|---|---|
| current | 01:57, 13 March 2023 | | 1,574 × 146 (35 KB) | ANUBHAVGARG (talk | contribs) | Uploaded a work by Han Cai from https://arxiv.org/pdf/1812.00332.pdf with UploadWizard |
You cannot overwrite this file.
File usage
The following page uses this file:
