|
|
|
|
|
|
by sytelus
2810 days ago
|
|
|
The paper claims that only few variables are documented so reproducibility is low. I think that's not a good measure because this complexity has kept increasing and it would require significant space in pdf to document every little detail. The good measure is if paper comes with code and it reproduces the claims. If I'm reading it right, this paper says AAAI and IJCAI had only 25% of papers with code. We really shouldn't be accepting AI papers without code. |
|
|