@juliaelliott, I was kind of hoping Kaggle will stand by their own rules and this post will not be necessary. However, if the results are now final, this has to be said:
\n\nThe two top teams probably cheated. I suspect a few more as well.
\n\nKaggle has the means to verify this, but they had funny team members that joined and right away submitted >0.6 submissions (within hours). These were not very active in days that did not have over 5 submits. I would bet their submission are pretty similar.
\n\nAlthough they consolidated these allegedly (gosh i LOVE this word) fake members, so overall their submissions were not over the limit, the submissions per day definitely were over it.
\n\nThis has been mentioned a few times in the discussions.
", "messages": [ { "id": "640456", "postDate": "10/04/2019 03:26:01", "content": "@juliaelliott, I was kind of hoping Kaggle will stand by their own rules and this post will not be necessary. However, if the results are now final, this has to be said:
\n\nThe two top teams probably cheated. I suspect a few more as well.
\n\nKaggle has the means to verify this, but they had funny team members that joined and right away submitted >0.6 submissions (within hours). These were not very active in days that did not have over 5 submits. I would bet their submission are pretty similar.
\n\nAlthough they consolidated these allegedly (gosh i LOVE this word) fake members, so overall their submissions were not over the limit, the submissions per day definitely were over it.
\n\nThis has been mentioned a few times in the discussions.
", "rawMarkdown": "juliaelliott, I was kind of hoping Kaggle will stand by their own rules and this post will not be necessary. However, if the results are now final, this has to be said:\n\nThe two top teams probably cheated. I suspect a few more as well.\n\nKaggle has the means to verify this, but they had funny team members that joined and right away submitted >0.6 submissions (within hours). These were not very active in days that did not have over 5 submits. I would bet their submission are pretty similar.\n\nAlthough they consolidated these allegedly (gosh i LOVE this word) fake members, so overall their submissions were not over the limit, the submissions per day definitely were over it.\n\nThis has been mentioned a few times in the discussions.", "votes": null }, { "id": "640478", "postDate": "10/04/2019 04:07:28", "content": "We have completed our investigation and made removals where evidence demonstrated rules violations. If you have specific evidence that supports a claim of a violation that has gone undetected, you (and anyone) is welcome to submit that through our compliance.
\n\nWe do look closely at suspicious behavior but cannot act purely on hunch or suspicion, as any removal is not taken lightly. So statements about scores achieved in what is subjectively considered too quickly are not substantiated enough to be actionable without the evidence of an actual violation to back it.
", "rawMarkdown": "We have completed our investigation and made removals where evidence demonstrated rules violations. If you have specific evidence that supports a claim of a violation that has gone undetected, you (and anyone) is welcome to submit that through our [compliance](https://www.kaggle.com/compliance). \n\nWe do look closely at suspicious behavior but cannot act purely on hunch or suspicion, as any removal is not taken lightly. So statements about scores achieved in what is subjectively considered too quickly are not substantiated enough to be actionable without the evidence of an actual violation to back it.", "votes": null }, { "id": "640714", "postDate": "10/04/2019 07:24:45", "content": "https://www.kaggle.com/c/open-images-2019-object-detection/discussion/110029#latest-635452\nhttps://www.kaggle.com/c/open-images-2019-object-detection/discussion/109991#latest-634910
", "rawMarkdown": "https://www.kaggle.com/c/open-images-2019-object-detection/discussion/110029#latest-635452\nhttps://www.kaggle.com/c/open-images-2019-object-detection/discussion/109991#latest-634910", "votes": null }, { "id": "640738", "postDate": "10/04/2019 07:47:33", "content": "@juliaelliott, I will do my bestest with meta kaggle, but I am limited. I can't see the predictions submitted, the prediction submission timeline, the kaggle upload metadata. You have all this. I think it is beyond reasonable doubt from my casual observation during the competition that even if these where not fake accounts, they were definitely cooperating before merge.\nIf anyone can help with \"concrete\" evidence, please do. \nWhile i understand kaggle turning a blind eye to missing the deadline in the external data disclosure (which is probably just a misunderstanding), this seems, allegedly, premeditated
", "rawMarkdown": "juliaelliott, I will do my bestest with meta kaggle, but I am limited. I can't see the predictions submitted, the prediction submission timeline, the kaggle upload metadata. You have all this. I think it is beyond reasonable doubt from my casual observation during the competition that even if these where not fake accounts, they were definitely cooperating before merge.\nIf anyone can help with \"concrete\" evidence, please do. \nWhile i understand kaggle turning a blind eye to missing the deadline in the external data disclosure (which is probably just a misunderstanding), this seems, allegedly, premeditated", "votes": null }, { "id": "641626", "postDate": "10/04/2019 21:25:12", "content": "@juliaelliott, Meta kaggle is incomplete now and does not have any of the submissions, leaderboard and users data, so there is really nothing I can do. BTW, I suspect the top 3 teams, not only the top 2. However, as I said, without Meta kaggle I can not prove anything.
", "rawMarkdown": "juliaelliott, Meta kaggle is incomplete now and does not have any of the submissions, leaderboard and users data, so there is really nothing I can do. BTW, I suspect the top 3 teams, not only the top 2. However, as I said, without Meta kaggle I can not prove anything.", "votes": null }, { "id": "641824", "postDate": "10/05/2019 07:02:03", "content": "@moshel @juliaelliott If you look between the top 50 and top 80 you'll see another weird pattern between 80% of the competitors in this area : all around 0.45 and probably from the same region - and all novice. Looks like an assignement made for a course, and share between all the students :o)
", "rawMarkdown": "moshel @juliaelliott If you look between the top 50 and top 80 you'll see another weird pattern between 80% of the competitors in this area : all around 0.45 and probably from the same region - and all novice. Looks like an assignement made for a course, and share between all the students :o)", "votes": null }, { "id": "641845", "postDate": "10/05/2019 08:19:50", "content": "no, these are the people who managed to convince the google frcnn inception2 from the zoo to lower its th... this is what you get if you do that
", "rawMarkdown": "no, these are the people who managed to convince the google frcnn inception2 from the zoo to lower its th... this is what you get if you do that", "votes": null } ], "comments": [ { "id": 640478, "author_name": "juliaelliott", "author_url": "", "post_date": "10/04/2019 04:07:28", "content": "We have completed our investigation and made removals where evidence demonstrated rules violations. If you have specific evidence that supports a claim of a violation that has gone undetected, you (and anyone) is welcome to submit that through our compliance.
\n\nWe do look closely at suspicious behavior but cannot act purely on hunch or suspicion, as any removal is not taken lightly. So statements about scores achieved in what is subjectively considered too quickly are not substantiated enough to be actionable without the evidence of an actual violation to back it.
", "votes": null, "replies": [ { "id": 641626, "author_name": "moshel", "author_url": "", "post_date": "10/04/2019 21:25:12", "content": "@juliaelliott, Meta kaggle is incomplete now and does not have any of the submissions, leaderboard and users data, so there is really nothing I can do. BTW, I suspect the top 3 teams, not only the top 2. However, as I said, without Meta kaggle I can not prove anything.
", "votes": null, "replies": [] } ] }, { "id": 640714, "author_name": "nhlr21", "author_url": "", "post_date": "10/04/2019 07:24:45", "content": "https://www.kaggle.com/c/open-images-2019-object-detection/discussion/110029#latest-635452\nhttps://www.kaggle.com/c/open-images-2019-object-detection/discussion/109991#latest-634910
", "votes": null, "replies": [] }, { "id": 640738, "author_name": "moshel", "author_url": "", "post_date": "10/04/2019 07:47:33", "content": "@juliaelliott, I will do my bestest with meta kaggle, but I am limited. I can't see the predictions submitted, the prediction submission timeline, the kaggle upload metadata. You have all this. I think it is beyond reasonable doubt from my casual observation during the competition that even if these where not fake accounts, they were definitely cooperating before merge.\nIf anyone can help with \"concrete\" evidence, please do. \nWhile i understand kaggle turning a blind eye to missing the deadline in the external data disclosure (which is probably just a misunderstanding), this seems, allegedly, premeditated
", "votes": null, "replies": [] }, { "id": 641824, "author_name": "arthurllau", "author_url": "", "post_date": "10/05/2019 07:02:03", "content": "@moshel @juliaelliott If you look between the top 50 and top 80 you'll see another weird pattern between 80% of the competitors in this area : all around 0.45 and probably from the same region - and all novice. Looks like an assignement made for a course, and share between all the students :o)
", "votes": null, "replies": [ { "id": 641845, "author_name": "moshel", "author_url": "", "post_date": "10/05/2019 08:19:50", "content": "no, these are the people who managed to convince the google frcnn inception2 from the zoo to lower its th... this is what you get if you do that
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "640456": "juliaelliott, I was kind of hoping Kaggle will stand by their own rules and this post will not be necessary. However, if the results are now final, this has to be said:\n\nThe two top teams probably cheated. I suspect a few more as well.\n\nKaggle has the means to verify this, but they had funny team members that joined and right away submitted >0.6 submissions (within hours). These were not very active in days that did not have over 5 submits. I would bet their submission are pretty similar.\n\nAlthough they consolidated these allegedly (gosh i LOVE this word) fake members, so overall their submissions were not over the limit, the submissions per day definitely were over it.\n\nThis has been mentioned a few times in the discussions.", "640478": "We have completed our investigation and made removals where evidence demonstrated rules violations. If you have specific evidence that supports a claim of a violation that has gone undetected, you (and anyone) is welcome to submit that through our [compliance](https://www.kaggle.com/compliance). \n\nWe do look closely at suspicious behavior but cannot act purely on hunch or suspicion, as any removal is not taken lightly. So statements about scores achieved in what is subjectively considered too quickly are not substantiated enough to be actionable without the evidence of an actual violation to back it.", "640714": "https://www.kaggle.com/c/open-images-2019-object-detection/discussion/110029#latest-635452\nhttps://www.kaggle.com/c/open-images-2019-object-detection/discussion/109991#latest-634910", "640738": "juliaelliott, I will do my bestest with meta kaggle, but I am limited. I can't see the predictions submitted, the prediction submission timeline, the kaggle upload metadata. You have all this. I think it is beyond reasonable doubt from my casual observation during the competition that even if these where not fake accounts, they were definitely cooperating before merge.\nIf anyone can help with \"concrete\" evidence, please do. \nWhile i understand kaggle turning a blind eye to missing the deadline in the external data disclosure (which is probably just a misunderstanding), this seems, allegedly, premeditated", "641626": "juliaelliott, Meta kaggle is incomplete now and does not have any of the submissions, leaderboard and users data, so there is really nothing I can do. BTW, I suspect the top 3 teams, not only the top 2. However, as I said, without Meta kaggle I can not prove anything.", "641824": "moshel @juliaelliott If you look between the top 50 and top 80 you'll see another weird pattern between 80% of the competitors in this area : all around 0.45 and probably from the same region - and all novice. Looks like an assignement made for a course, and share between all the students :o)", "641845": "no, these are the people who managed to convince the google frcnn inception2 from the zoo to lower its th... this is what you get if you do that" }, "source": "meta" }