Begin 30-Day Comment Period for Ibrahim Marcouch’s cmix-lex
James Bowery
We are now beginning the 30-day comment and verification period for Ibrahim Marcouch’s submission, cmix-lex, which shares credit with Kaido Orav and has exceeded the 1% improvement award threshold.
Source code:
https://github.com/blahem/cmix-lex
Improvement:
1.01% = 100*(1 - 109671639 / 110793128)Resource use:
CPU Allowed: 42.4757 = 70000 / 1648This entry has also raised a dispute that might have been avoided if Hutter Prize judging were as automated as some other compression competitions. Full automation would require contestants to pay for cloud-computing resources, not only for judging but also for development. One important feature of the Hutter Prize, from its beginning in the 2000s, has been that it remained accessible to talented contestants with limited resources.
Even though encode.su has an important history in lossless compression, it should not be treated as the official place to give notice about Hutter Prize matters. Comments during the 30-day evaluation period should be posted to the Hutter Prize Google group.
See:
http://prize.hutter1.net/#details
Priority notices, such as those involved in the present dispute, are especially important. It is not practical for the judging committee to search the web for such disclosures.
In this case, an out-of-protocol announcement on encode.su appeared to start the 30-day comment period. As a judge, I share responsibility for letting that happen, and I apologize to the parties involved. Hutter Prize entries are rare, so this kind of procedural issue is not something we exercise often.
On the substance of the current case:
Legally, Ibrahim’s use of Kaido Orav’s open-source code through cmix-lex appears to be permitted by Kaido’s GNU license. Kaido could have used a more restrictive license until he was ready to submit a Hutter Prize entry himself. A strictly legal reading would also be consistent with the Prize’s objective award criteria.
However, the relevant question is not only legal. It is also about algorithmic contribution.
The encode.su community, like any community, has its own sense of fair competition. How that should relate to Marcus Hutter’s goals for the Prize is ultimately Marcus’s decision.
Fabrice Bellard, I believe, suggested apportioning the award by factoring the code contributions. Ibrahim responded positively to that suggestion. Kaido then argued that the overwhelming algorithmic contribution was his. Ibrahim seems apologetic and may be willing to move away from his original suggestion of an equal split, while also maintaining that his work to make Kaido’s entry satisfy the judging system’s technical constraints was more than cosmetic.
For my own part, I would not have been as patient with Ibrahim if I had known then what I know now about the entry. My reason for trusting him was influenced by a bias toward talented contestants with limited resources. That has been an important part of the Hutter Prize’s history. Ibrahim’s situation as a student in financial difficulty resonated with me.
Priority notices, such as Kaido’s, should be posted to the Hutter Prize Google group or sent directly to the judging committee. That is a simple procedural requirement, but it matters.
Had Kaido posted to the Hutter Prize Google group, or emailed the judges, stating his intent to submit an entry based on fxcm_v26, the likely outcome would have been very different. In that case, I believe the appropriate response would have been:
Announce in the Hutter Prize Google group that anyone submitting an entry based on fxcm_v26 must first reach agreement with Kaido.
Advise Kaido to place a restrictive license on his proprietary code immediately, even though that would not undo the earlier public release.
As things stand, Marcus Hutter and Matt Mahoney are both disposed toward a 50-50 split. As usual, Marcus will make the final decision after the 30-day comment period.