Kaggle and other platforms and communities have been holding deep learning/machine learnin/data mining online competitions to promote related research and commercialization. In such kind of competitions, it is often stateted that, e.g., third party dataset or the testing partition of a specific dataset, is not allowed for training.
However, is there any way to detect if a team has used forbidden data?
I do not think so. This (kaggle) platform is based on the given databases over particular tasks. The algorith/method/technique is however the choice of participent.
I think at least you can check it trough reproducability.
Meaning if your alogirthm/method or technique is not able to score similar results with the real original data then it seems obvious that there is something foul.
Especially if the difference is huge.
Also very often they make their own verify split so if you add more data and it includes the data of their verify split then it could be even more obvious.
Dear Guang Hua
This doc from kaggle might be useful, have a look
https://www.kaggle.com/docs/competitions
Kind Regards
Qamar Ul Islam
- Badges
- Science topic