Reproducing scientific papers » History » Version 1
Evgeniy Pavlovskiy, 2022-03-21 13:23
| 1 | 1 | Evgeniy Pavlovskiy | h1. Reproducing scientific papers |
|---|---|---|---|
| 2 | 1 | Evgeniy Pavlovskiy | |
| 3 | 1 | Evgeniy Pavlovskiy | Given: |
| 4 | 1 | Evgeniy Pavlovskiy | 1. A link to a scientific paper. |
| 5 | 1 | Evgeniy Pavlovskiy | 2. A link to a repository with the sourcecode for the paper. |
| 6 | 1 | Evgeniy Pavlovskiy | |
| 7 | 1 | Evgeniy Pavlovskiy | Requirements: |
| 8 | 1 | Evgeniy Pavlovskiy | 1. Report should have your own result on test or validation set. |
| 9 | 1 | Evgeniy Pavlovskiy | 2. The results should be compared with the paper's one. |
| 10 | 1 | Evgeniy Pavlovskiy | 3. A slide should contain your experience summary, including you positive feedback, highlighting problems of reproducing, discrepancy reason if present. |
| 11 | 1 | Evgeniy Pavlovskiy | |
| 12 | 1 | Evgeniy Pavlovskiy | Actions: |
| 13 | 1 | Evgeniy Pavlovskiy | [0. Find the source codes for the paper.] |
| 14 | 1 | Evgeniy Pavlovskiy | 1. Run training on a given ML-pipeline if you have enough resources. If no, find available models. If both no, select another paper. |
| 15 | 1 | Evgeniy Pavlovskiy | 2. Run test on the obtained model. |